ghc 8 added backtraces on uncaught errors. This is great, but git-annex was
using error in many places for a error message targeted at the user, in
some known problem case. A backtrace only confuses such a message, so omit it.
Notably, commands like git annex drop that failed due to eg, numcopies,
used to use error, so had a backtrace.
This commit was sponsored by Ethan Aubin.
When used with an older version of ssh, any ServerAliveInterval in
~/.ssh/config will be overridden by .git/annex/ssh.config.
This commit was sponsored by Josh Taylor on Patreon.
So that stalled transfers will be noticed within about 3 minutes,
even if TCPKeepAlive is disabled or doesn't work.
Rather than setting with -o, use -F with another config file,
so that any settings in ~/.ssh/config or /etc/ssh/ssh_config overrides this.
Closes https://github.com/datalad/datalad/issues/1020
The use of runWriter in scanUnlockedFiles broke due to this change;
it failed with blocked indefinitely in mvar, because the database write
handle was taken while linkFromAnnex needed to also write to it (to update
the inode cache). So, switched to using a separate runWriter for each call
to addAssociatedFileFast. A little less efficient, but not greatly; the
writes should all still be cached.
In the case where the pointer file is in place, and not the content
of the object, lock's performNew was called with filemodified=True,
which caused it to try to repopulate the object from an unmodified
associated file, of which there were none. So, the content of the object
got thrown away incorrectly. This was the cause (although not the root
cause) of data loss in https://github.com/datalad/datalad/issues/1020
The same problem could also occur when the work tree file is modified,
but the object is not, and lock is called with --force. Added a test case
for this, since it's excercising the same code path and is easier to set up
than the problem above.
Note that this only occurred when the keys database did not have an inode
cache recorded for the annex object. Normally, the annex object would be in
there, but there are of course circumstances where the inode cache is out
of sync with reality, since it's only a cache.
Fixed by checking if the object is unmodified; if so we don't need to
try to repopulate it. This does add an additional checksum to the unlock
path, but it's already checksumming the worktree file in another case,
so it doesn't slow it down overall.
Further investigation found a similar problem occurred when smudge --clean
is called on a file and the inode cache is not populated. cleanOldKeys
deleted the unmodified old object file in this case. This was also
fixed by checking if the object is unmodified.
In general, use of getInodeCaches and sameInodeCache is potentially
dangerous if the inode cache has not gotten populated for some reason.
Better to use isUnmodified. I breifly auited other places that check the
inode cache, and did not see any immediate problems, but it would be easy
to miss this kind of problem.
An easy change now that supportedVersions is a list. Since v3 and v5 are
identical other than version number, just add v3 to the list.
This commit was sponsored by andrea rota.
Fixes a bug introduced with v6 mode that I didn't notice until now.
Probably not many v3 repos left out there, and upgrading them to v6 mode
is not disastrous, only a little premature.
This commit was sponsored by Riku Voipio
.. and have to be checked to see if they are a pointed to an annexed file.
Cases where such memory use could occur included, but were not limited to:
- git commit -a of a large unlocked file (in v5 mode)
- git-annex adjust when a large file was checked into git directly
Generally, any use of catKey was a potential problem.
Fix by using git cat-file --batch-check to check size before catting.
This adds another git batch process, which is included in the CatFileHandle
for simplicity.
There could be performance impact, anywhere catKey is used. Particularly
likely to affect adjusted branch generation speed, and operations on
unlocked files in v6 mode. Hopefully since the --batch-check and
--batch read the same data, disk buffering will avoid most overhead.
Leaving only the overhead of talking to the process over the pipe and
whatever computation --batch-check needs to do.
This commit was sponsored by Bruno BEAUFILS on Patreon.
Speeds up commands like "git-annex find --in remote" by over 50%.
Profiling showed that adjustGitEnv was 21% of the time and 37% of the
allocations of that command. It copied the environment each time with
getEnvironment.
The only repeated use of adjustGitEnv is in withIndexFile, which tends to
be run at least once per file. So, it was optimised by keeping a cache of
the environment, which can be reused.
There could be other better ways to optimise this. Maybe get the while
environment once at startup. But, then it would have to be serialized back
out each time running a child process, so I doubt that would be a net win.
It might be better to cache a version of the environment that is
pre-modified to use .git-annex/index. But, profiling doesn't show that
modifying the enviroment is taking any significant time.
key2file and file2key were top cost centers according to profiling.
The repeated use of replace was not efficient. This new approach is quite a
lot more efficient.
This commit was sponsored by Denis Dzyubenko on Patreon.
* sync: Previously, when run in a branch with a slash in its name,
such as "foo/bar", the sync branch was "synced/bar". That conflicted
with the sync branch used for branch "bar", so has been changed to
"synced/foo/bar".
* adjust: Previously, when adjusting a branch with a slash in its name,
such as "foo/bar", the adjusted branch was "adjusted/bar(unlocked)".
That conflicted with the adjusted branch used for branch "bar",
so has been changed to "adjusted/foo/bar(unlocked)"
* Also, running sync in an adjusted branch did not correctly sync
changes back to the parent branch when it had a slash in its name.
This bug has been fixed.
Eliminate use of Git.Ref.under and Git.Ref.basename; using
Git.Ref.underBase and Git.Ref.base make everything handle deep branches
correctly.
Probably noone was adjusting deep branches, and v6 is still experimental
anyway, so I'm not going to worry about the mess that was left by that bug.
In the case of git-annex sync, using a fixed git-annex with an old unfixed
one will mean they use different sync branches for a deep branch, and so
they may stop syncing until the old one is upgraded. However, that's only
a problem when syncing between repositories without going via a central
bare repository. Added a warning about this to the CHANGELOG, but it's
probably not going to affect many people at all.
This commit was sponsored by Riku Voipio.
This makes -Jn work with --json and --quiet, where before
setting -Jn disabled those options.
Concurrent json output is currently a mess though since threads output
chunks over top of one-another.
Only done in -J mode because only if there's concurrency can downloading
from two remotes be faster. Without concurrency, it's likely the case that
sequential downloads from the same remote are faster than switching back
and forth between two remotes.
There is some hairy MVar code here, but basically it just keeps
the activeremotes MVar full except when deciding which remote to assign
to a thread.
Also affects gets by sync --content -J
This commit was sponsored by Jochen Bartl.
This was disabled in commit 61ccf95004,
because only the assistant used them, and they were clutter. But, now
--failed also uses them.
Remove the failure log files after successful transfers. Should avoid
most of the clutter problems.
Commit 61ccf95004 mentions a subtle behavior
change, which has now been reverted:
There is one behavior change from this. If glacier is being used, and a
manual git annex get --from glacier fails because the file isn't available
yet, the assistant will no longer later see that failed transfer file and
retry the get.
Note that get --from foo --failed will get things that a previous get --from bar
tried and failed to get, etc. I considered making --failed only retry
transfers from the same remote, but it was easier, and seems more useful,
to not have the same remote requirement.
Noisy due to some refactoring into Types/
Use nextRandom to generate the random UUID, rather than using randomIO.
This gets fixes for the following two bugs in the uuid library.
However, this did not impact git-annex much, so a hard depedency has
not been added on uuid-1.3.12.
https://github.com/aslatter/uuid/issues/15
"v4 UUIDs are not that random"
This doesn't greatly affect git-annex, because even with only
2^64 possible UUIDs, the chance that two git-annex repositories
that are clones of the same git repo get the same UUID is miniscule.
And, git-annex generates only one UUID per run, so preducting
subsequent UUIDs is not a problem.
https://github.com/aslatter/uuid/issues/16
"Remove Random instance for UUID, or mark it as deprecated"
git-annex was using that instance; let's stop before it gets
deprecated or removed.
Show branch:file that is being operated on.
I had to make ActionItem a type and not a type class because
withKeyOptions' passed two different types of values when using the type
class, and I could not get the type checker to accept that.
This bug caused broken tree objects to get built by a later git annex sync.
This is a somewhat unlikely but not impossible situation, and the test
suite's union_merge_regression test tickled it when it was run on FAT.
Mostly the username is only used for the git committer or other display
purposes, and we can just fall back to a dummy value in these cases.
The only remaining place where an error is thrown is when starting local
pairing, which needs the username to be known.
The queue could potentially contain changes from before withAltRepo, and
get flushed inside the call, which would apply the changes to the modified
repo.
Or, changes could be queued in withAltRepo that were intended to affect
the modified repo, but don't get flushed until later.
I don't know of any cases where either happens, but better safe than sorry.
Note that this affect withIndexFile, which is used in git-annex branch
updates. So, it potentially makes things slower. Should not be by much;
the overhead consists only of querying the current queue a couple of times,
and potentially flushing changes queued within withAltRepo earlier, that
could have maybe been bundled with other later changes.
Notice in particular that the existing queue is not flushed when calling
withAltRepo. So eg when git annex add needs to stage files in the index,
it will still bundle them together efficiently.
Added guard in Annex.Transfer to prevent this problem at a deeper level.
I'm unhappy ith NoUUID, but having Maybe UUID instead wouldn't help either
if nothing checked that there was a UUID. Since there legitimately need to
be Remotes that do not have a UUID, I can't see a way to fix it at the type
level, short making there be two separate types of Remotes.
Removed the instance LensGpgEncParams RemoteConfig because it encouraged
code that does not take the RemoteGitConfig into account.
RemoteType's setup was changed to take a RemoteGitConfig,
although the only place that is able to provide a non-empty one is
enableremote, when it's changing an existing remote. This led to several
folow-on changes, and got RemoteGitConfig plumbed through.
This is actually worse than I thought; when git is being run with a
detached work tree, GIT_INDEX_FILE is treated as a path relative to CWD,
instead of the normal behavior of relative the top of the work tree.
This seems to make it basically impossible for any program that wants to
use GIT_INDEX_FILE to use anything other than an absolute path to it; there
are too many configurations to keep straight that can change how git
interprets what should be a simple relative path to a file.
(I have complained to the git developers.)
This affected git annex view. It turns out that some other places
that use GIT_INDEX_FILE were already working around the bug. I removed the
workaround from Annex.Branch since the new workaround will do.
Could not think of a foolproof way to detect if the old adjusted branch was
just behind the current branch. It's possible that the user amended the
adjusting commit at the head of the adjusted branch, for example.
I decided to bail in this situation, instead of just entering the old
branch, so that if git annex adjust succeeds the user is always in a
*current* adjusted branch, not some old and out of date one.
What could perhaps be done is enter the old branch and then update it. But
that seems too magical; the user may have rebased master or something or
may not want to propigate the changes from the old branch. Best to error
out.
It started exporting a isSymbolicLink which supports windows. But,
git-annex does no use symlinks on windows yet and this conflicts with the
function by the same name from unix-compat, so hide it.
git 2.8.1 (or perhaps 2.9.0) is going to prevent git merge from merging in
unrelated branches. Since the webapp's pairing etc features often combine
together repositories with unrelated histories, work around this behavior
change by setting GIT_MERGE_ALLOW_UNRELATED_HISTORIES when the assistant
merges.
Note though that this is not done for git annex sync's merges, so
it will follow git's default or configured behavior.
When git-annex is used with a git version older than 2.2.0, disable support for
adjusted branches, since GIT_COMMON_DIR is needed to update them and was first
added in that version of git.
Made all Annex.Perms file mode changing functions ignore errors when
core.sharedRepository is set, because the file might be owned by someone
else. I don't fancy getting bug reports about crashes due to set modes in
this configuration, which is a very foot-shooty configuration in the first
place.
The fsck warning is necessary because old repos kept files mode 444, which
doesn't allow locking them, and so if the mode remains 444 due to the file
being owned by someone else, the user should be told about it.
When annex.thin is set, adding an object will add the execute bits to the
work tree file, and this does mean that the annex object file ends up
executable.
This doesn't add any complexity that wasn't already present, because git
annex add of an executable file has always ingested it so that the annex
object ends up executable.
But, since an annex object file can be executable or not, when populating
an unlocked file from one, the executable bit is always added or removed
to match the mode of the pointer file.
This is how direct mode does it too, and somehow, for reasons that
currently escape me, this makes git merge not care if it's run with an
empty work tree.
Was using L.readFile, so the Handle would remain open until the garbage
collector got around to it. Changed to explicit open and close, so we know
it's always closed when the function returns.
This makes the direct mode to v6 upgrade able to be performed in one clone
of a repository without affecting other clones, which can continue using v5
and direct mode.
This does mean that it has to write out temp files containing updated
objects for the merge. So may use more disk space, and disk IO, but that
should generally win out over needing to launch N separate
git hash-object processes.
An unlocked present file does not have a pointer file in the worktree, so
info skipped counting it.
It may be that unused was also affected by the problem, but it seemed not
to be in my tests. I think because of the use of the associatedFilesFilter.
This fix slows down both info and unused a little bit, since they have to
query the contents of files from git, but only when handling unlocked files.
Only reverse adjust the changes in the commit, which means that adjustments
do not need to be generally cleanly reversable.
For example, an adjustment can unlock all locked files, but does not need
to worry about files that were originally unlocked when reversing, because
it will only ever be run on files that have been changed. So, it's ok
if it locks all files when reversed, or even leaves all files as-is when
reversed.
Using adjusted/unlocked/master made lots of git stuff dealing with "master"
complain that it was ambiguous. This new appoach is more like view branch
names, and shows the adjustment right there in the branch display even if
only the basename of the branch is shown.
There's a race here, but entering an adjusted branch for the first time is
not something to do when a commit is being made at the same time. Although,
may want to prevent the assistant from committing while entering the
adjusted branch.
So, it will pull and push the original branch, not the adjusted one.
And, for merging, it will use updateAdjustedBranch (not implemented yet).
Note that remaining uses of Git.Branch.current need to be checked too;
for things that should act on the original branch, and not the adjusted
branch.
"git annex adjust" may be a temporary interface, but works for a proof of
concept.
It is pretty fast at creating the adjusted branch. The main overhead is
injecting pointer files. It might be worth optimising that by reusing the
symlink target as the pointer file content. When I tried to do that,
the problem was that the clean filter doesn't use that same format, and so
git thought files had changed. Could be dealt with, perhaps make the clean
filter use symlink format for pointer files when on an adjusted branch?
But the real overhead is in checking out the branch, when git runs the
smudge filter once per file. That is perhaps too slow to be usable,
although it may only affect initial checkout of the branch, and not
updates. TBD.
* add, addurl, import, importfeed: When in a v6 repository on a crippled
filesystem, add files unlocked.
* annex.addunlocked: New configuration setting, makes files always be
added unlocked. (v6 only)
The problem with having the slashes unescaped is, it broke parsing, since
the parser takes the filename to get the part containing the key.
That particularly affected URL keys.
This makes the format be the same as symlinks point to, which keeps things
simple.
Existing pointer files will continue to work ok.
Before, the call to mkProgressUpdater created the directory as a
side-effect, but since that ignored failure to create it, this led to
a "does not exist" exception when the transfer lock file was created,
rather than a permissions error.
So, make sure the directory exists before trying to lock the file in it.
When a PermissionDenied exception is caught, skip making the transfer lock.
This lets downloads from readonly remotes happen.
If an upload is being tried, and the lock file can't be written due to
permissions, then probably the actual transfer will fail for the same
reason, so I think it's ok that it continues w/o taking the lock in that
case.
The type checker should have noticed this, but the changes to mapM
that make it accept any Traversable hid the fact that it was not being
passed a list at all. Thus, what should have returned an empty list most
of the time instead returned [""] which was treated as the name of the
associated file, with disasterout consequences.
When I have time, I should add a test case checking what sync --content
drops. I should also consider replacing mapM with one re-specialized to
lists.
Previously, it only flushed when the queue got larger than 1.
Also, make the queue auto-flush when items are added, rather than needing
to be flushed as a separate step. This simplifies the code and make it more
efficient too, as it avoids needing to read the queue out of the state to
check if it should be flushed.
The homomorphs are back, just encoded such that it doesn't crash in LANG=C
However, I noticed a bug in the old escaping; [pseudoSlash] was escaped the
same as ['/','/']. Fixed by using '%' to escape pseudoSlash. Which requires
doubling '%' to escape it, but that's already done in the escaping of
worktree filenames in a view, so is probably ok.
Linking the file to the tmp dir was not necessary in the clean
filter, and it caused the ctime to change, which caused git to think
the file was changed. This caused git status to get slow as it kept
re-cleaning unchanged files.
Fixes several bugs with updates of pointer files. When eg, running
git annex drop --from localremote
it was updating the pointer file in the local repository, not the remote.
Also, fixes drop ../foo when run in a subdir, and probably lots of other
problems. Test suite drops from ~30 to 11 failures now.
TopFilePath is used to force thinking about what the filepath is relative
to.
The data stored in the sqlite db is still just a plain string, and
TopFilePath is a newtype, so there's no overhead involved in using it in
DataBase.Keys.
WorkTree.lookupFile was finding a key for a file that's deleted from the
work tree, which is different than the v5 behavior (though perhaps the same
as the direct mode behavior). Fix by checking that the work tree file exists
before catting its key.
Hopefully this won't slow down much, probably the catKey is much more expensive.
I can't see any way to optimise this, except perhaps to make Command.Unused
check if work tree files exist before/after calling lookupFile. But,
it seems better to make lookupFile really only find keys for worktree files;
that's what it's intended to do.
Else, queued file stages won't have reached the index, and it won't find
everthing.
This evidently fixes a reversion in my work today, although I don't see how
I broke it. It didn't use to flush the queue first, before, and worked
somehow.
Test suite for v5 is back to 100% green now.
The smudge filter does need to be run, because if the key is in the local
annex already (due to renaming, or a copy of a file added, or a new file
added and its content has already arrived), git merge smudges the file and
this should provide its content.
This does probably mean that in merge conflict resolution, git smudges the
existing file, re-copying all its content to it, and then the file is
deleted. So, not efficient.
This is a behavior change for merge conflicts between locked files
that both pointed to the same key, in different ways.
Before, the conflict was resolved, but the file was renamed to .variant.
This was unnecessary, because there was only one variant.
Of course, this also handles conflicts between unlocked and locked, or even
two unlocked files with different pointer contents.
Since the file was present and locked, its annex object was not in the
inode cache. So, despite not needing to update the annex object when the
clean filter is run on the content by git merge, it does need to record the
inode cache of the annex object. Otherwise, the annex object will be
assumed to be bad, since its inode is not cached.
Several tricky parts:
* When the conflict is just between the same key being locked and unlocked,
the unlocked version wins, and the file is not renamed in this case.
* Need to update associated file map when conflict resolution renames
an unlocked file.
* git merge runs the smudge filter on the conflicting file, and actually
overwrites the file with the same content it had before, and so
invalidates its inode cache. This makes it difficult to know when it's
safe to remove such files as conflict cruft, without going so far as to
compare their entire contents.
Dealt with this by preventing the smudge filter from populating the file
when a merge is run. However, that also prevents the smudge filter being
run for non-conflicting files, so eg moving a file won't put its new
content into place.
* Ideally, if a merge or a merge conflict resolution renames an unlocked
file, the file in the work tree can just be moved, rather than copying
the content to a new worktree file.
This is attempted to be done in merge conflict resolution, but
due to git merge's behavior of running smudge filters, what actually
seems to happen is the old worktree file with the content is deleted and
rewritten as a pointer file, so doesn't get reused.
So, this is probably not as efficient as it optimally could be.
If that becomes a problem, could look into running the merge in a separate
worktree and updating the real worktree more efficiently, similarly to the
direct mode merge. However, the direct mode merge had a lot of bugs, and
I'd rather not use that more error-prone method unless really needed.
Decided it's too scary to make v6 unlocked files have 1 copy by default,
but that should be available to those who need it. This is consistent with
git-annex not dropping unused content without --force, etc.
* Added annex.thin setting, which makes unlocked files in v6 repositories
be hard linked to their content, instead of a copy. This saves disk
space but means any modification of an unlocked file will lose the local
(and possibly only) copy of the old version.
* Enable annex.thin by default on upgrade from direct mode to v6, since
direct mode made the same tradeoff.
* fix: Adjusts unlocked files as configured by annex.thin.
This optimisation was not necessary, and didn't work for v6 unlocked files.
Typically only a small number of files will be changed by a commit, so just
catKey them all.
This can happen when ingesting a new file in either locked or unlocked
mode, when some unlocked files in the repo use the same key, and the
content was not locally available before.
This fixes a race where the modified file ended up in annex/objects, and
the InodeCache stored in the database was for the modified version, so
git-annex didn't know it had gotten modified.
The race could occur when the smudge filter was running; now it gets the
InodeCache before generating the Key, which avoids the race.
In v5, lookupFile is supposed to only look at symlinks on disk (except when
in direct mode).
Note that v6 also has a bug when a locked file's symlink is deleted and is
replaced with a new file. It sees that a link is staged and gets that
key.
The problem is that shutdown is not always called, particularly in the test
suite. So, a database connection would be opened, possibly some changes
queued, and then not shut down.
One way this can happen is when using Annex.eval or Annex.run with a new
state. A better fix might be to make both of them call Keys.shutdown
(and be sure to do it even if the annex action threw an error).
Complication: Sometimes they're run reusing an existing state, so shutting
down a database connection could cause problems for other users of that
same state. I think this would need a MVar holding the database handle,
so it could be emptied once shut down, and another user of the database
connection could then start up a new one if it got shut down. But, what if
2 threads were concurrently using the same database handle and one shut it
down while the other was writing to it? Urgh.
Might have to go that route eventually to get the database access to run
fast enough. For now, a quick fix to get the test suite happier, at the
expense of speed.
This covers the case where multiple files have the same content and are
added with git add. Previously only the one that was linked to the annex
got its inode cached; now both are.
When a v6 unlocked files is removed from the work tree,
unused doesn't show it. When it gets removed from the index,
unused does show it. This is the same as a locked file.
If multiple files point to the same annex object, the user may want to
modify them independently, so don't use a hard link.
Also, check diskreserve when copying.
Before the smudge filter added a trailing newline, but other things that
wrote formatPointer to a file did not.
also some new pointer staging code to use later
This avoids querying the database when the content file doen't exist
(or otherwise fails the provided check). However, it does add overhead of
querying the database, and will certianly impact performance.
The Keys database can hold multiple inode caches for a given key. One for
the annex object, and one for each pointer file, which may not be hard
linked to it.
Inode caches for a key are recorded when its content is added to the annex,
but only if it has known pointer files. This is to avoid the overhead of
maintaining the database when not needed.
When the smudge filter outputs a file's content, the inode cache is not
updated, because git's smudge interface doesn't let us write the file. So,
dropping will fall back to doing an expensive verification then. Ideally,
git's interface would be improved, and then the inode cache could be
updated then too.
Renamed the db to keys, since it is various info about a Keys.
Dropping a key will update its pointer files, as long as their content can
be verified to be unmodified. This falls back to checksum verification, but
I want it to use an InodeCache of the key, for speed. But, I have not made
anything populate that cache yet.
This removes ambiguity, because while someone might have "WORM--foo" in a
file that's not intended to be a git-annex pointer file,
"annex/objects/WORM--foo" is less likely.
Also, 664cc987e8 had a caveat about symlink
targets being parsed as pointer files, and now the same parser is used for
both.
I did not include any hash directories before the key in the pointer file,
as they're not needed. However, if they were included, the parser would
still work ok.
Backend.lookupFile is changed to always fall back to catKey when
operating on a file that's not a symlink.
catKey is changed to understand pointer files, as well as annex symlinks.
Before, catKey needed a file mode witness, to be sure it was looking at a
symlink. That was complicated stuff. Now, it doesn't actually care if a
file in git is a symlink or not; in either case asking git for the content
of the file will get the pointer to the key.
This does mean that git-annex will treat a link
foo -> WORM--bar as a git-annex file, and also treats
a regular file containing annex/objects/WORM--bar as a git-annex file.
Calling catKey could make git-annex commands need to do more work than
before. This would especially be the case if a repo contained many regular
files, and only a few annexed files, as now git-annex will need to ask
git about the contents of the regular files.
Was not putting it inside the temp dir, but next to it!
This was just wrong, and it led to a longer filename that desired being
used, leading to some bug reports.
Since all places where a repo is used in direct mode need to have git-annex
upgraded before the repo can safely be converted to v6, the upgrade needs
to be manual for now.
I suppose that at some point I'll want to drop all the direct mode support
code. At that point, will stop supporting v5, and will need to auto-upgrade
any remaining v5 repos. If possible, I'd like to carry the direct mode
support for say, a year or so, to give people plenty of time to upgrade and
avoid disruption.
When core.sharedRepository is set, annex object files are not made mode
444, since that prevents a user other than the file owner from locking
them. Instead, a mode such as 664 is used in this case.
replaceFile created a temp file, which was guaranteed to not overlap with
another temp file. However, makeAnnexLink then deleted that file, in
preparation for making the symlink in its place. This caused a race, since
some other replaceFile could create a temp file, using the same name!
I was able to reproduce the race easily running git-annex add -J10 in a
directory with 100 files (all with different contents). Some files would
get ingested into the annex, but their annex links would fail to be added.
There could be other situations where this same problem could occur.
Perhaps when the assistant is adding a file, if the user manually also ran
git-annex add. Perhaps in cases not involving adding a file.
The new replaceFile makes a temprary directory, which is guaranteed to be
unique, and doesn't make a temp file in there. makeAnnexLink can thus
create the symlink without problem and the race is avoided.
Audited all calls to replaceFile to make sure that the old behavior of
providing an empty temp file was not relied on.
The general problem of asking for a temp file and deleting it as part of
the process of using it could reach beyond replaceFile. Did some quick
audits and didn't find other cases of it. Probably only symlink creation
stuff would tend to make that mistake, mostly.
Instead, only display transport error if the configlist output doesn't
include an annex.uuid line, even an empty one.
A recent change made git-annex init try to get all the remote uuids, and so
the transport error would be displayed by it. It was also displayed when
eg, copying files to a remote that had no uuid yet.
sync, merge, assistant: When git merge failed for a reason other than a
conflicted merge, such as a crippled filesystem not allowing particular
characters in filenames, git-annex would make a merge commit that could
omit such files or otherwise be bad. Fixed by aborting the whole merge
process when git merge fails for any reason other than a merge conflict.
Implemented with no additional overhead of compares etc.
This is safe to do for presence logs because of their locality of change;
a given repo's presence logs are only ever changed in that repo, or in a
repo that has just been actively changing the content of that repo.
So, we don't need to worry about a split-brain situation where there'd
be disagreement about the location of a key in a repo. And so, it's ok to
not update the timestamp when that's the only change that would be made
due to logging presence info.
sideAction is for things not generally related to the current action being
performed. And, it adds a newline after the side action. This was not the
right thing to use for stuff like "checksum", where doing a checksum is
part of the git annex get process, and indeed we want it to display
"(checksum...) ok"
By definition, a trusted repository is trusted to always have its location
tracking log accurate. Thus, it should never be in a position where content
is being dropped from it concurrently, as that would result in the location
tracking log not being accurate.
This avoids a failure where eg, we start with RecentlyVerifiedCopies
for all remotes, and so didn't do any active verification, which is
required.
Also, dedup the list of VerifiedCopies when checking if we have enough,
in case 2 copies of a UUID slip in.
See doc/bugs/concurrent_drop--from_presence_checking_failures.mdwn for
discussion about why 1 locked copy is all we can require, and how this
fixes concurrent dropping bugs.
Note that, since nothing yet generates a VerifiedCopyLock yet, this commit
breaks dropping temporarily.
There should be no behavior changes in this commit, it just adds a more
expressive data type and adjusts code that had been passing around a [UUID]
or sometimes a Maybe Remote to instead use [VerifiedCopy].
Although, since some functions were taking two different [UUID] lists,
there's some potential for me to have gotten it horribly wrong.
Also, rename lockContent to lockContentExclusive
inAnnexSafe should perhaps be eliminated, and instead use
`lockContentShared inAnnex`. However, I'm waiting on that, as there are
only 2 call sites for inAnnexSafe and it's fiddly.
In c6632ee5c8, it actually only handled
uploading objects to a shared repository. To avoid verification when
downloading objects from a shared repository, was a lot harder.
On the plus side, if the process of downloading a file from a remote
is able to verify its content on the side, the remote can indicate this
now, and avoid the extra post-download verification.
As of yet, I don't have any remotes (except Git) using this ability.
Some more work would be needed to support it in special remotes.
It would make sense for tahoe to implicitly verify things downloaded from it;
as long as you trust your tahoe server (which typically runs locally),
there's cryptographic integrity. OTOH, despite bup being based on shas,
a bup repo under an attacker's control could have the git ref used for an
object changed, and so a bup repo shouldn't implicitly verify. Indeed,
tahoe seems unique in being trustworthy enough to implicitly verify.
* When annex objects are received into git repositories, their checksums are
verified then too.
* To get the old, faster, behavior of not verifying checksums, set
annex.verify=false, or remote.<name>.annex-verify=false.
* setkey, rekey: These commands also now verify that the provided file
matches the key, unless annex.verify=false.
* reinject: Already verified content; this can now be disabled by
setting annex.verify=false.
recvkey and reinject already did verification, so removed now duplicate
code from them. fsck still does its own verification, which is ok since it
does not use getViaTmp, so verification doesn't happen twice when using fsck
--from.
I couldn't find a good way to make an *empty* index file (zero byte file
won't do), so I punted and just don't make index.lock when there's no index
yet. This means some other git process could race and write an index file
at the same time as the merge is ongoing, in theory. Only happens in new
repos though.
Oh boy, not again. So, another place that the filesystem encoding needs to
be applied. Yay.
In passing, I changed decodeBS so if a NUL is embedded in the input, the
resulting FilePath doesn't get truncated at that NUL. This was needed to
make prop_b64_roundtrips pass, and on reviewing the callers of decodeBS, I
didn't see any where this wouldn't make sense. When a FilePath is used to
operate on the filesystem, it'll get truncated at a NUL anyway, whereas if
a String is being used for something else, it might conceivably have a NUL
in it, and we wouldn't want it to get truncated when going through
decodeBS.
(NB: There may be a speed impact from this change.)
I think that the problem was caused by windows not having a concept of an
env var that is set, but to the empty string. So, GIT_ANNEX_SSHOPTION
got set to "" and was not seen as set at all.
Easy fix, which also makes git-annex sync a little faster is to not set
GIT_SSH, when GIT_ANNEX_SSHOPTION has no options. Might as well let git use
ssh per usual in this case, no need to run git-annex as the proxy ssh
command..
* proxy: Fix proxy git commit of non-annexed files in direct mode.
* proxy: If a non-proxied git command, such as git revert
would normally fail because of unstaged files in the work tree,
make the proxied command fail the same way.
* Perform a clean shutdown when --time-limit is reached.
This includes running queued git commands, and cleanup actions normally
run when a command is finished.
* fsck: Commit incremental fsck database when --time-limit is reached.
Previously, some of the last files fscked did not make it into the
database when using --time-limit.
Note that this changes Annex.addCleanup hooks, to run after --time-limit
expires. Fsck was using such a hook to clean up after a
--incremental-schedule, and that shouldn't run when --time-limit exipires
it. So, instead, moved that cleanup code to be run by cleanupIncremental.
Resulted in some data type juggling.
This is needed because when preferred content matches on files,
the second pass would otherwise want to drop all keys. Using a bloom filter
avoids this, and in the case of a false positive, a key will be left
undropped that preferred content would allow dropping. Chances of that
happening are a mere 1 in 1 million.
This makes git annex unused use around 48 mb more memory than it did before,
but the massive increase in accuracy makes this worthwhile for all but the
smallest systems.
Also, I want to use the bloom filter for sync --all --content, to avoid
dropping files that the preferred content doesn't want, and 1/1000
false positives would be far too many in that use case, even if it were
acceptable for unused.
Actual memory use numbers:
1000: 21.06user 3.42system 0:26.40elapsed 92%CPU (0avgtext+0avgdata 501552maxresident)k
1000000: 21.41user 3.55system 0:26.84elapsed 93%CPU (0avgtext+0avgdata 549496maxresident)k
10000000: 21.84user 3.52system 0:27.89elapsed 90%CPU (0avgtext+0avgdata 549920maxresident)k
Based on these numbers, 10 million seemed a better pick than 1 million.
This removes a bit of complexity, and should make things faster
(avoids tokenizing Params string), and probably involve less garbage
collection.
In a few places, it was useful to use Params to avoid needing a list,
but that is easily avoided.
Problems noticed while doing this conversion:
* Some uses of Params "oneword" which was entirely unnecessary
overhead.
* A few places that built up a list of parameters with ++
and then used Params to split it!
Test suite passes.
The content file may not be owned by the user running git-annex, in which
case, setting the owner write bit was not enough to let lockContent
act on the file. However, with some core.sharedRepository configs, the file
should be writable by the user's group. So, the thing to do is to call
thawContent on it.
It was returning Just False in this situation, which differed from indirect
mode behavior. I don't think this led to any actual problems; things that
checked if the file being dropped was present just failed to fail, and
instead reported it wasn't present, possibly incorrectly.
Hmm, it's possible that this could have made git annex fsck --from remote
update the location log wrongly, if a remote was in direct mode, and was in
the middle of trying to drop a key, and the drop later failed.
Also cleaned up the code, avoiding creating a lock file if we're going to
open it for create later anyway.
And, if there's an exception while preparing to lock the file, but not at
the point of actually taking the lock, throw an exception, instead of
silently not locking and pretending to succeed.
And, on Windows, always use lock file, even if the repo somehow got into
indirect mode (maybe with cygwin git..)
The one exception is in Utility.Daemon. As long as a process only
daemonizes once, which seems reasonable, and as long as it avoids calling
checkDaemon once it's already running as a daemon, the fcntl locking
gotchas won't be a problem there.
Annex.LockFile has it's own separate lock pool layer, which has been
renamed to LockCache. This is a persistent cache of locks that persist
until closed.
This is not quite done; lockContent stil needs to be converted.
Should be no behavior changes, just simplified code.
The only actual difference is it doesn't truncate the lock file.
I think that was a holdover from when transfer info was written to the lock
file.
Only the assistant uses these, and only the assistant cleans them up, so
make only git annex transferkeys write them,
There is one behavior change from this. If glacier is being used, and a
manual git annex get --from glacier fails because the file isn't available
yet, the assistant will no longer later see that failed transfer file and
retry the get. Hope no-one depended on that old behavior.
The setDifferences that got added to initialize turns out to make a git
commit, and before ensureCommit has been used. Thus, repo init can fail
when the system has a broken hostname etc.
Move the ensureCommit to the very first thing to avoid this kind of breakage.
This works, and seems fairly robust. Clean get of 20 files at -J3. At -J10,
there are some messages about ssh multiplexing, probably due to a race
spinning up the ssh connection cacher. But, it manages to get all the files
ok regardless.
The progress bars are a scrambled mess though, due to bugs in
ascii-progress, which I've already filed. Particularly this one:
https://github.com/yamadapc/haskell-ascii-progress/issues/8
Came up with a generic way to filter out progress messages while keeping
errors, for commands that use stderr for both.
--json mode will disable command outputs too.
This might be overkill; I only know I need it in ls-files, but other git
commands can also do their own globbing, it turns out, and I am pretty sure
I never want them too when git-annex is using them as plumbing.
Test suite still passes and it looks ok.
This was introduced by commit 450ee53ab6
However, the same problem could affect other calls to programPath,
specifically some on the assistant. So, I fixed it at a deeper level.
Seems to work, but still experimental until it's been tested more.
When repositories are on filesystems not supporting symlinks, the .git dir
symlink trick cannot be used. Since we're going to be in direct mode
anyway, the .git dir symlink is not strictly needed.
However, I have not fixed the code that creates new annex symlinks to
handle this case -- the committed symlinks will be wrong.
git annex sync happens to currently fail in a submodule using direct mode,
because there's no HEAD ref. That also needs to be dealt with to get
this fully working in crippled filesystems.
Leaving http://github.com/datalad/datalad/issues/44 open until these issues
are dealt with.
Most of the time, there will be no discreprancy between programPath and
readProgramFile.
But, the programFile might have been written by an old version of git-annex
that is still installed, while a newer one is currently running. In this
case, we want to run the same one that's currently running.
This is especially important for things like the GIT_SSH=git-annex used for
ssh connection caching.
The only code that still uses readProgramFile directly is the upgrade code,
which needs to know where the standalone git-annex was installed, in order to
upgrade it.
Turns out sqlite does not like having its database deleted out from
underneath it. It might suffice to empty the table, but I would rather
start each fsck over with a new database, so I added a lock file, and
running incremental fscks use a shared lock.
This leaves one concurrency bug left; running two concurrent fsck --more
will lead to: "SQLite3 returned ErrorBusy while attempting to perform step."
and one or both will fail. This is a concurrent writers problem.
* sync: Use the ssh-options git config when doing git pull and push.
* remotedaemon: Use the ssh-options git config.
Note that the rename env var means that if a new git-annex calls an old one
for git-annex ssh, or a new calls an old, nothing much will go wrong;
just ssh caching won't happen.
I hope this doesn't impact speed much -- it does have to pull out a value
from Annex state every time it accesses the branch now.
The test case I dropped has never caught any problems that I can remember,
and would have been rather difficult to convert.
Eliminated complexity and future proofed. The most important change is that
all functions over Difference are now total; any Difference that can be
expressed should be handled. Avoids needs for sanity checking of inputs,
and version skew with the future.
Also, the difference.log now serializes a [Difference], not a Differences.
This saves space and keeps it simpler.
Note that [Difference] might contain conflicting differences (eg,
[Version5, Version6]. In this case, one of them needs to consistently win
over the others, probably based on Ord.
* init: Repository tuning parameters can now be passed when initializing a
repository for the first time. For details, see
http://git-annex.branchable.com/tuning/
* merge: Refuse to merge changes from a git-annex branch of a repo
that has been tuned in incompatable ways.
This is necessary for interop between inode caches created on unix and
windows. Which is more important than supporting inodecaches for large keys
with the wrong size, which are broken anyway.
There should be no slowdown from this change, except on Windows.
Avoid using fileSize which maxes out at just 2 gb on Windows.
Instead, use hFileSize, which doesn't have a bounded size.
Fixes support for files > 2 gb on Windows.
Note that the InodeCache code only needs to compare a file size,
so it doesn't matter it the file size wraps. So it has been
left as-is. This was necessary both to avoid invalidating existing inode
caches, and because the code passed FileStatus around and would have become
more expensive if it called getFileSize.
This commit was sponsored by Christian Dietrich.
Reverts 965e106f24
Unfortunately, this caused breakage on Windows, and possibly elsewhere,
because parentDir and takeDirectory do not behave the same when there is a
trailing directory separator.
parentDir is less safe than takeDirectory, especially when working
with relative FilePaths. It's really only useful in loops that
want to terminate at /
This commit was sponsored by Audric SCHILTKNECHT.
This fixes 9 test suite failures. There are some tricky things going on
with the paths to the index file, and git's working directory, which
are hard to get right with relative paths. So, I switched back to absolute
here, at least for now.
Only 2 test suite failures remain on this branch, but there are other
potential problems the test suite doesn't catch. Including some calls to
setCurrentDirectory -- I was wrong and git-annex does do that in a few
places, like when generating a view.
This allows the git repository to be moved while git-annex is running in
it, with fewer problems.
On Windows, this avoids some of the problems with the absurdly small
MAX_PATH of 260 bytes. In particular, git-annex repositories should
work in deeper/longer directory structures than before. See
http://git-annex.branchable.com/bugs/__34__git-annex:_direct:_1_failed__34___on_Windows/
There are several possible ways this change could break git-annex:
1. If it changes its working directory while it's running, that would
be Bad News. Good news everyone! git-annex never does so. It would also
break thread safety, so all such things were stomped out long ago.
2. parentDir "." -> "" which is not a valid path. I had to fix one
instace of this, and I should probably wipe all calls to parentDir out
of the git-annex code base; it was never a good idea.
3. Things like relPathDirToFile require absolute input paths,
and code assumes that the git repo path is absolute and passes it to it
as-is. In the case of relPathDirToFile, I converted it to not make
this assumption.
Currently, the test suite has 16 failures.
This allows bypassing the direct mode guard in a safe way to do all sorts
of things including git revert, git mv, git checkout ...
This commit was sponsored by the WikiMedia Foundation.
Didn't know that this library existed!
This includes making git-annex not re-exec itself on start on windows, and
making the test suite on Windows run tests without forking.
Added a Default instance for TrustLevel, and was able to use that to clear
up several other parts of the code too.
This commit was sponsored by Stephan Schulz
Found these with:
git grep "^ " $(find -type f -name \*.hs) |grep -v ': where'
Unfortunately there is some inline hamlet that cannot use tabs for
indentation.
Also, Assistant/WebApp/Bootstrap3.hs is a copy of a module and so I'm
leaving it as-is.
This fixes all instances of " \t" in the code base. Most common case
seems to be after a "where" line; probably vim copied the two space layout
of that line.
Done as a background task while listening to episode 2 of the Type Theory
podcast.
* New annex.hardlink setting. Closes: #758593
* init: Automatically detect when a repository was cloned with --shared,
and set annex.hardlink=true, as well as marking the repository as
untrusted.
Had to reorganize Logs.Trust a bit to avoid a cycle between it and
Annex.Init.
This avoids cp -a overriding the default mode acls that the user might have
set in a git repository.
With GNU cp, this behavior change should not be a breaking change, because
git-anex also uses rsync sometimes in the same situation, and has only ever
preserved timestamps when using rsync.
Systems without GNU cp will no longer use cp -a, but instead just cp.
So, timestamps will no longer be preserved. Preserving timestamps when
copying between repos is not guaranteed anyway.
Closes: #729757
This fixed one bug where it needed to be and wasn't (in Assistant.Unused).
And also found one place where lockContent was used unnecessarily (by
drop --from remote).
A few other places like uninit probably don't really need to lockContent,
but it doesn't hurt to do call it anyway.
This commit was sponsored by David Wagner.
Also fixes a test suite failures introduced in recent commits, where
inAnnexSafe failed in indirect mode, since it tried to open the lock file
ReadWrite. This is why the new checkLocked opens it ReadOnly.
This commit was sponsored by Chad Horohoe.
The nice refactoring in ec7dd0446a
highlighted a bug in lockContent -- when the content is not present,
this incorrectly created an empty lock file, using the same filename
as the content file.
This seems like it could result in empty objects, which fsck would detect
and complain about. Both drop and move --to call lockContent, as does
Remote.Git.dropKey -- I think we got lucky and this bug didn't show up
because both all of those only operate on files that are present. So
this bug could only manifest if there was a race, and a file's content
was dropped at just the wrong time, just as another process was about to
drop it. (And then only if the other process's dropping failed, otherwise
it'd delete the empty object file.)
Hmm, move --from also called lockContent. Unnecessarily, since the content
is not being removed from the local annex. In this case, the combination of
the 2 bugs could result in an empty lock file being written, and then if
the download of the content failed, left in the object directory as the
content.
This commit also optimises lockContent, avoiding an unncessary
doesFileExist test and instead just catching the exception that's thrown
when the file doesn't exist.
This commit was sponsored by Justine Lam.
Added a convenience Utility.LockFile that is not a windows/posix
portability shim, but still manages to cut down on the boilerplate around
locking.
This commit was sponsored by Johan Herland.
This was a bug, but it was only used for ssh locks and by the hook special
remote locking. At least in the case of ssh locks, the lock files happened
to already exist before this tried to use them, so the bug didn't cause
anything to break.
This does mean that eg, copying multiple files to a local remote will
become slightly slower, since it now restarts git-cat-file after each copy.
Should not be significant slowdown.
The reason git-cat-file is run on the remote at all is to update its
location log. In order to add an item to it, it needs to get the current
content of the log. Finding a way to avoid needing to do that would be a
good path to avoiding this slowdown if it does become a problem somehow.
This commit was sponsored by Evan Deaubl.
(With the exception of daemon pid locking.)
This fixes at part of #758630. I reproduced the assistant locking eg, a
removable drive's annex journal lock file and forking a long-running
git-cat-file process that inherited that lock.
This did not affect Windows.
Considered doing a portable Utility.LockFile layer, but git-annex uses
posix locks in several special ways that have no direct Windows equivilant,
and it seems like it would mostly be a complication.
This commit was sponsored by Protonet.
replaceFileOr was broken and ran the rollback action always.
Luckily, for replaceFile, the rollback action was safe to run, since it
just nuked a temp file that had already been moved into place.
However, when `git annex direct` used replaeFileOr, its rollback printed a
scary message:
/home/joey/tmp/rrrr/.git/annex/misctmp/tmp32268: rename: does not exist (No such file or directory)
There was actually no bad result though.
Removed old extensible-exceptions, only needed for very old ghc.
Made webdav use Utility.Exception, to work after some changes in DAV's
exception handling.
Removed Annex.Exception. Mostly this was trivial, but note that
tryAnnex is replaced with tryNonAsync and catchAnnex replaced with
catchNonAsync. In theory that could be a behavior change, since the former
caught all exceptions, and the latter don't catch async exceptions.
However, in practice, nothing in the Annex monad uses async exceptions.
Grepping for throwTo and killThread only find stuff in the assistant,
which does not seem related.
Command.Add.undo is changed to accept a SomeException, and things
that use it for rollback now catch non-async exceptions, rather than
only IOExceptions.
For example, I had a copy to a remote that was failing for an unknown
reason. This let me see the exception was createDirectory: permission
denied; the underlying problem being a permissions issue.
Putting a callback in the Retriever type allows for the callback to
remove the retrieved file when it's done with it.
I did not really want to make Retriever be fixed to Annex Bool,
but when I tried to use Annex a, I got into some type of type mess.
Needed for eg, Remote.External.
Generally, any Retriever that stores content in a file is responsible for
updating the meter, while ones that procude a lazy bytestring cannot update
the meter, so are not asked to.
Slightly tricky as they are not normal UUIDBased logs, but are instead maps
from (uuid, chunksize) to chunkcount.
This commit was sponsored by Frank Thomas.
I think this is a git behavior change, but have not checked to be sure.
Conflict cruft used to look like $foo~HEAD, but now just $foo is left
behind as conflict cruft.
With test case.
Running `git annex direct` would cause loss of data, because the object
was moved to a temp file, which it then tried to replace the work tree file
with, and on failure, the temp file got deleted. Now it's instead moved
back into the annex object location.
Minor because normally only 1 FD is leaked per git-annex run. However,
the test suite leaks a few hundred FDs, and this broke it on the Debian
autobuilders, which seem to have a tigher than usual ulimit.
The leak was introduced by the lazy getDirectoryContents' that was
introduced in e6330988dd in order to scale to
millions of journal files -- if the lazy list was never fully consumed, the
directory handle did not get closed.
Instead, pull in openDirectory/readDirectory/closeDirectory code that I
already developed and submitted in a patch to the haskell directory library
earlier. Using this in journalDirty avoids the place that the lazy list
caused a problem. And using it in stageJournal eliminates the need for
getDirectoryContents'.
The getJournalFiles* functions are switched back to using the regular
strict getDirectoryContents. I'm not sure if those always consume the whole
list, so this avoids any leak. And the things that call those are things
like git annex unused, which also look at every file committed to the
git-annex branch, so would need more work to scale to insane numbers of
files anyway.
When one side is an annexed symlink, and the other side is a non-annexed symlink.
In this case, git-merge does not replace the annexed symlink in the work
tree with the non-annexed symlink, which is different from it's handling of
conflicts between annexed symlinks and regular files or directories.
So, while git-annex generated the correct merge commit, the work tree
didn't get updated to reflect it.
See comments on bug for additional analysis.
Did not add this to the test suite yet; just unloaded a truckload of firewood
and am feeling lazy.
This commit was sponsored by Adam Spiers.
Eg after git-annex add has run on 2 million files in one go.
Slightly unhappy with the neeed to use a temp file here, but I cannot see
any other alternative (see comments on the bug report).
This commit was sponsored by Hamish Coleman.
Support users who have set commit.gpgsign, by disabling gpg signatures for
git-annex branch commits and commits made by the assistant.
The thinking here is that a user sets commit.gpgsign intending the commits
that they manually initiate to be gpg signed. But not commits made in the
background, whether by a deamon or implicitly to the git-annex branch.
gpg signing those would be at best a waste of CPU and at worst would fail,
or flood the user with gpg passphrase prompts, or put their signature on
changes they did not directly do.
See Debian bug #753720.
Also makes all commits done by git-annex go through a few central control
points, to make such changes easier in future.
Also disables commit.gpgsign in the test suite.
This commit was sponsored by Antoine Boegli.
When annex.genmetadata is set, metadata from the feed is added to files
that are imported from it.
Reused the same feedtitle and itemtitle, feedauthor, itemauthor, etc names
that are used in --template.
Also added title and author, which are the item title/author if available,
falling back to the feed title/author. These are more likely to be common
metadata fields.
(There is a small bit of dupication here, but once git gets
around to packing the object, it will compress it away.)
The itempubdate field is not included in the metadata as a string; instead
it is used to generate year and month fields, same as is done when adding
files with annex.genmetadata set.
This commit was sponsored by Amitai Schlair, who cooincidentially
is responsible for ikiwiki generating nice feed metadata!
http://marc.info/?l=git&m=140262402204212&w=2
This git bug manifested on FAT and Windows as the test suite failing in 3
places. All involved merge conflict resolution. It turned out that the
associated file mappings were getting messed up, and that happened because
this git bug lost track of what files were supposed to be symlinks.
This commit was sponsored by Eric Kidd.
Rather than calculating the TSDelta once, and caching it, this now
reads the inode sential file's InodeCache file once, and then each time a
new InodeCache is generated, looks at the sentinal file to get the current
delta.
This way, if the time zone changes while git-annex is running, it will
adapt.
This adds some inneffiency, but only on Windows, and only 1 stat per new
file added. The worst innefficiency is that `git annex status` and
`git annex sync` will now (on Windows) stat the inode sentinal file once per
file in the repo.
It would be more efficient to use getCurrentTimeZone, rather than needing
to stat the sentinal file. This should be easy to do, once the time
package gets my bugfix patch.
This commit was sponsored by Jürgen Lüters.
On Windows, changing the time zone causes the apparent mtime of files to
change. This confuses git-annex, which natually thinks this means the files
have actually been modified (since THAT'S WHAT A MTIME IS FOR, BILL <sheesh>).
Work around this stupidity, by using the inode sentinal file to detect if
the timezone has changed, and calculate a TSDelta, which will be applied
when generating InodeCaches.
This should add no overhead at all on unix. Indeed, I sped up a few
things slightly in the refactoring.
Seems to basically work! But it has a big known problem:
If the timezone changes while the assistant (or a long-running command)
runs, it won't notice, since it only checks the inode cache once, and
so will use the old delta for all new inode caches it generates for new
files it's added. Which will result in them seeming changed the next time
it runs.
This commit was sponsored by Vincent Demeester.
It was possible for a interrupted sync or merge in direct mode to
leave the work tree out of sync with the last recorded commit.
This would result in the next commit seeing files missing from the work
tree, and committing their removal.
Now, a direct mode merge happens not only in a throwaway work tree, but using
a temporary index file, and without any commits or index changes
being made until the real work tree has been updated. If the merge is
interrupted, the work tree may have some updated files, but worst case a
commit will redundantly commit changes that come from the merge.
This commit was sponsored by Tony Cantor.
There was a tricky bit here, when it does combine, the edit form is shown,
and so the info needs to be committed to the new repository, but then
pulled into the current one. And caches need to be invalidated for it
to be visible in the edit form.
Allow any encoding to be used, as with filenames (but utf8 is the sane
choice). Affects metadata and repository descriptions, and preferred
content expressions.
The question of what's the right encoding for the git-annex branch is a
vexing one. utf-8 would be a nice choice, but this leaves the possibility
of bad data getting into a git-annex branch somehow, and this resulting in
git-annex crashing with encoding errors, which is a failure mode I want to
avoid.
(Also, preferred content expressions can refer to filenames, and filenames
can have any encoding, so limiting to utf-8 would not be ideal.)
The union merge code already took care to not assume any encoding for a
file. Except it assumes that any \n is a literal newline, and not part of
some encoding of a character that happens to contain a newline. (At least
utf-8 avoids using newline for anything except liternal newlines.)
Adapted the git-annex branch code to use this same approach.
Note that there is a potential interop problem with Windows, since
FileSystemEncoding doesn't work there, and instead things are always
decoded as utf-8. If someone uses non-utf8 encoding for data on the
git-annex branch, this can lead to an encoding error on windows. However,
this commit doesn't actually make that any worse, because the union merge
code would similarly fail with an encoding error on windows in that
situation.
This commit was sponsored by Kyle Meyer.
To use, set GIT_ANNEX_SSHASKPASS to point to a fifo or regular file
(FIFO is better, avoids touching disk or multiple readers) that contains
the password. Then set SSH_ASKPASS=git-annex, and when ssh runs it, it will
tell ssh the password.
This is not yet used..
Only fsck and reinject and the test suite used the Backend, and they can
look it up as needed from the Key. This simplifies the code and also speeds
it up.
There is a small behavior change here. Before, all commands would warn when
acting on an annexed file with an unknown backend. Now, only fsck and
reinject show that warning.
For sync, saves 1 ssh connection per remote. For remotedaemon, the same
ssh connection that is already open to run git-annex-shell notifychanges
is reused to pull from the remote.
Only potential problem is that this also enables connection caching
when the assistant syncs with a ssh remote. Including the sync it does
when a network connection has just come up. In that case, cached ssh
connections are likely to be stale, and so using them would hang.
Until I'm sure such problems have been dealt with, this commit needs to
stay on the remotecontrol branch, and not be merged to master.
This commit was sponsored by Alexandre Dupas.
Motivation: Hook scripts for nautilus or other file managers
need to provide the user with feedback that a file is being downloaded.
This commit was sponsored by THM Schoemaker.
Note that this is a nearly entirely free feature. The data was already
stored in the metadata log in an easily accessible way, and already was
parsed to a time when parsing the log. The generation of the metadata
fields may even be done lazily, although probably not entirely (the map
has to be evaulated to when queried).
For example "standard or (include=otherdir/*)" or even "not standard"
Note that the implementation avoids any potential for loops (if a
standard preferred content expression itself mentioned standard).
This commit was sponsored by Jochen Bartl.
Old ssh did not check the hostname passed to -O stop, so I had used "any".
But now ssh does check it! I think this happened as part of the client-side
hostname canonicalization changes in 6.5p1, but have not verified that
introduced the problem.
The symptom was that it would try to dns lookup "any", which often caused a
bit of a delay at shutdown. And the old ssh connection kept running, so
it would do it over and over again.
Fixed by using localhost, which hopefully reliably resolves to some address
that ssh will accept.. Also nukeFile the socket after ssh has been asked to
shutdown, just in case.
unused: In direct mode, files that are deleted from the work tree are no longer incorrectly detected as unused.
Direct mode `git annex info` slows down a bit due to more stringent
checking, but not by a lot.
This is a new feature, it was not handled before, since it's a bit of an
edge case. However, it can be handled exactly the same as a file/dir
conflict, just leave the non-annexed item alone.
While implementing this, the core resolveMerge' function got a lot simpler
and clearer. Note especially that where before there was an asymetric call to
stagefromdirectmergedir, now graftin is called symmetrically in both cases.
And, in order to add that `graftin us`, the current branch needed to be
known (if there is no current branch, there cannot be a merge conflict).
This led to some cleanups of how autoMergeFrom behaved when there is no
current branch.
This commit was sponsored by Philippe Gauthier.
Added test cases for both ways this can happen, with a conflict involving a
file, or a directory.
Cleaned up resolveMerge to not touch the work tree in direct mode, which
turned out to be the only way to handle things.. And makes it much nicer.
Still need to run test suite on windows.
In the case of a conflicted merge where the remote adds a directory, and we
have a file (which is checked in), resolveMerge' will create the link,
and so the fix for 1192d98721 looked at that,
thought it was an unannexed file (it's not in the oldref), and preserved
it.
This is a hacky fix. It would be better for resolveMerge' to not update the
work tree, at least in direct mode, and only stage the changes, which
mergeDirectCleanUp could then move into tree. I want to make that change,
but this is not the time to do it.
Using the extract(1) program to do the heavy lifting.
Decided to make git-annex run pre-commit-annex when committing. Since
git-annex pre-commit also runs it, it'll be run when git commit is run too,
via the pre-commit hook. This basically gives back the pre-commit hook
that git-annex took away. The implementation avoids repeatedly looking
for the hook script when the assistant is running and committing
repeatedly; only checks if the hook is available once.
To make the script simpler, made git-annex metadata -s field?=value
only set a field when it's not already got a value.
This commit was sponsored by bak.
Note that negated globs are not supported. Would have complicated the code
to add them, without changing the data type serialization in a
non-backwards-compatable way.
This commit was sponsored by Denver Gingerich.
This allows eg, putting .git/annex/tmp on a ram disk, if the disk IO
of temp object files is too annoying (and if you don't want to keep
partially transferred objects across reboots).
.git/annex/misctmp must be on the same filesystem as the git work tree,
since files are moved to there in a way that will not work cross-device,
as well as symlinked into there.
I first wanted to put the tmp objects in .git/annex/objects/tmp, but
that would pose transition problems on upgrade when partially transferred
objects existed.
git annex info does not currently show the size of .git/annex/misctemp,
since it should stay small. It would also be ok to make something clean it
out, periodically.
Performance impact: When adding a large tree of new files, this needs
to do some git cat-file queries to check if any of the files already
existed and might need a metadata copy. I tried a benchmark in a copy
of my sound repository (so there was already a significant git tree
to check against.
Adding 10000 small files, with a cold cache:
before: 1m48.539s
after: 1m52.791s
So, impact is 0.0004 seconds per file added. Which seems acceptable, so did
not add some kind of configuration to enable/disable this.
This commit was sponsored by Lisa Feilen.
When constructing views, metadata is available about the location of the
file in the view's reference branch. Allows incorporating parts of the
directory hierarchy in a view.
For example `git annex view tag=* podcasts/=*` makes a view in the form
tag/showname.
Performance impact: I benchmarked git annex view tag=* in the conference
proceedings repo to take 6.459s before this change, and 6.544s after.
FWIW, I considered making the syntax for this be podcasts/*, which might
be easier for the user to learn. However, I think it's not as good:
* The user has to then juggle two different syntaxes, and podcasts/* will
be expanded by the shell so they also need to quote it, while podcasts/=*
is unlikely to be expanded by the shell.
* It would allow for things like podcasts/*/* and *.mp3 which do not
map well into views.
This commit was sponsored by Aurélien Pinceaux.
While writing this documentation, I realized that there needed to be a way
to stay in a view like tag=* while adding a filter like tag=work that
applies to the same field.
So, there are really two ways a view can be refined. It can have a new
"field=explicitvalue" filter added to it, which does not change the
"shape" of the view, but narrows the files it shows.
Or, it can have a new view added, which adds another level of
subdirectories.
So, added a vfilter command, which takes explicit values to add to the
filter, and rejects changes that would change the shape of the view.
And, made vadd only accept changes that change the shape of the view.
And, changed the View data type slightly; now components that can match
multiple metadata values can be visible, or not visible.
This commit was sponsored by Stelian Iancu.
So the user can now switch to a view and then move files around within it
to manage metadata. For example, moving a file into a new directory
when in the tags=* view adds a tag to it.
Implementation is fairly efficient. One diff-index, which is no more
expensive than the first stage of a git commit, followed by possibly
some cat-file --batch traffic to find the key (when deleting a file).
Very similar to what's done in direct mode when committing. And like
direct mode when updating the WC after a merge, it has to buffer the
diff-tree values in order to make 2 passes over them.
When not in a view, pre-commit now does one extra git symbolic-ref,
which is tiny overhead.
This commit was sponsored by Andrew Eskridge.
I was careful to write the code so its clear how laziness memoizes it,
although it's likely that much less explicit currying would have had
the same effect. Verified that the memoization works using a Debug.Trace.
Removed instance, got it all to build using fromRef. (With a few things
that really need to show something using a ref for debugging stubbed out.)
Then added back Read instance, and made Logs.View use it for serialization.
This changes the view log format.
(And a vpop command, which is still a bit buggy.)
Still need to do vadd and vrm, though this also adds their documentation.
Currently not very happy with the view log data serialization. I had to
lose the TDFA regexps temporarily, so I can have Read/Show instances of
View. I expect the view log format will change in some incompatable way
later, probably adding last known refs for the parent branch to View
or something like that.
Anyway, it basically works, although it's a bit slow looking up the
metadata. The actual git branch construction is about as fast as it can be
using the current git plumbing.
This commit was sponsored by Peter Hogg.
Promosing work toward metadata driven filter branches. A few methods
to construct them are stubbed out; all the data types and pure code
seems good.
This commit was sponsored by Walter Somerville.
Adds metadata log, and command.
Note that unsetting field values seems to currently be broken.
And in general this has had all of 2 minutes worth of testing.
This commit was sponsored by Julien Lefrique.
ef24751922 described a bug moving between
remotes in direct mode; I can no longer reproduce it with this strange
workaround removed. Also test suite still passes. Hope the broken code just
got fixed in the meantime.
Potentially fixes some FD leak if an action on an opened file handle fails
for some reason. There have been some hard to reproduce reports of
git-annex leaking FDs, and this may solve them.
Seems that locking of annexed objects when they're being dropped was broken
in direct mode:
* When taking the lock before dropping, it created the .git/annex/objects
file, as an empty file. It seems that the dropping code deleted that,
but that is not right, and for all I know could in some situation cause
a corrupted object to leak out.
* When the lock was checked, it actually tried to open each direct mode
file, and checked if it was locked. Not the same lock used above, and
could also fail if some consumer of the file locked it.
Fixed this, and added windows support by switching direct mode to lock a
.lck file.
Several places assumed this would not happen, and when the AssociatedFile
was Nothing, did nothing.
As part of this, preferred content checks pass the Key around.
Note that checkMatcher is sometimes now called with Just Key and Just File.
It currently constructs a FileMatcher, ignoring the Key. However, if it
constructed a FileKeyMatcher, which contained both, then it might be
possible to speed up parts of Limit, which currently call the somewhat
expensive lookupFileKey to get the Key.
I have not made this optimisation yet, because I am not sure if the key is
always the same. Will need some significant checking to satisfy myself
that's the case..
Checking .gitattributes adds a full minute to a git annex find looking for
files that don't have enough copies. 2:25 increasts to 3:27. I feel this is
too much of a slowdown to justify making it the default. So, exposed two
versions of the preferred content expression, a slow one and a fast but
approximate one.
I'm using the approximate one in the default preferred content expressions
to avoid slowing down the assistant.
* Add numcopiesneeded preferred content expression.
* Client, transfer, incremental backup, and archive repositories
now want to get content that does not yet have enough copies.
This means the asssistant will make copies of files that don't yet
meet the configured numcopies, even to places that would not normally want
the file.
For example, if numcopies is 4, and there are 2 client repos and
2 transfer repos, and 2 removable backup drives, the file will be sent
to both transfer repos in order to make 4 copies. Once a removable drive
get a copy of the file, it will be dropped from one transfer repo or the
other (but not both).
Another example, numcopies is 3 and there is a client that has a backup
removable drive and two small archive repos. Normally once one of the small
archives has a file, it will not be put into the other one. But, to satisfy
numcopies, the assistant will duplicate it into the other small archive
too, if the backup repo is not available to receive the file.
I notice that these examples are fairly unlikely setups .. the old behavior
was not too bad, but it's nice to finally have it really correct.
.. Almost. I have skipped checking the annex.numcopies .gitattributes
out of fear it will be too slow.
This commit was sponsored by Florian Schlegel.
* numcopies: New command, sets global numcopies value that is seen by all
clones of a repository.
* The annex.numcopies git config setting is deprecated. Once the numcopies
command is used to set the global number of copies, any annex.numcopies
git configs will be ignored.
* assistant: Make the prefs page set the global numcopies.
This global numcopies setting is needed to let preferred content
expressions operate on numcopies.
It's also convenient, because typically if you want git-annex to preserve N
copies of files in a repo, you want it to do that no matter which repo it's
running in. Making it global avoids needing to warn the user about gotchas
involving inconsistent annex.numcopies settings.
(See changes to doc/numcopies.mdwn.)
Added a new variety of git-annex branch log file, that holds only 1 value.
Will probably be useful for other stuff later.
This commit was sponsored by Nicolas Pouillard.
I've been disliking how the command seek actions were written for some
time, with their inversion of control and ugly workarounds.
The last straw to fix it was sync --content, which didn't fit the
Annex [CommandStart] interface well at all. I have not yet made it take
advantage of the changed interface though.
The crucial change, and probably why I didn't do it this way from the
beginning, is to make each CommandStart action be run with exceptions
caught, and if it fails, increment a failure counter in annex state.
So I finally remove the very first code I wrote for git-annex, which
was before I had exception handling in the Annex monad, and so ran outside
that monad, passing state explicitly as it ran each CommandStart action.
This was a real slog from 1 to 5 am.
Test suite passes.
Memory usage is lower than before, sometimes by a couple of megabytes, and
remains constant, even when running in a large repo, and even when
repeatedly failing and incrementing the error counter. So no accidental
laziness space leaks.
Wall clock speed is identical, even in large repos.
This commit was sponsored by an anonymous bitcoiner.
Similar to the assistant, this honors any configured preferred content
expressions.
I am not entirely happpy with the implementation. It would be nicer if
the seek function returned a list of actions which included the individual
file gets and copies and drops, rather than the current list of calls to
syncContent. This would allow getting rid of the somewhat reundant display
of "sync file [ok|failed]" after the get/put display.
But, do that, withFilesInGit would need to somehow be able to construct
such a mixed action list. And it would be less efficient than the current
implementation, which is able to reuse several values between eg get and
drop.
Note that currently this does not try to satisfy numcopies when
getting/putting files (numcopies are of course checked when dropping
files!) This makes it like the assistant, and unlike get --auto
and copy --auto, which do duplicate files when numcopies is not yet
satisfied. I don't know if this is the right decision; it only seemed to
make sense to have this parallel the assistant as far as possible to start
with, since I know the assistant works.
This commit was sponsored by Øyvind Andersen Holm.
This adds a http HEAD before the download is done. That was already the
case when the assistant was running, and it seems worth it to avoid filling
up the whole disk, like happened to my server today.
This allows a remote to store a piece of arbitrary state associated with a
key. This is needed to support Tahoe, where the file-cap is calculated from
the data stored in it, and used to retrieve a key later. Glacier also would
be much improved by using this.
GETSTATE and SETSTATE are added to the external special remote protocol.
Note that the state is left as-is even when a key is removed from a remote.
It's up to the remote to decide when it wants to clear the state.
The remote state log, $KEY.log.rmt, is a UUID-based log. However,
rather than using the old UUID-based log format, I created a new variant
of that format. The new varient is more space efficient (since it lacks the
"timestamp=" hack, and easier to parse (and the parser doesn't mess with
whitespace in the value), and avoids compatability cruft in the old one.
This seemed worth cleaning up for these new files, since there could be a
lot of them, while before UUID-based logs were only used for a few log
files at the top of the git-annex branch. The transition code has also
been updated to handle these new UUID-based logs.
This commit was sponsored by Daniel Hofer.
The assistant's commit code also always avoids git commit, for simplicity.
Indirect mode sync still does a git commit -a to catch unstaged changes.
Note that this means that direct mode sync no longer runs the pre-commit
hook or any other hooks git commit might call. The git annex pre-commit
hook action for direct mode is however explicitly run. (The assistant
already ran git commit with hooks disabled, so no change there.)
0980f3dae6 broke support for local remotes
from direct mode repos, because the relative path was taken to be from the
gitdir, rather than from the work tree.
This works around horribleness in the Mavericks cpp, which falls over on
the #if when configure is running. Moving it avoids the file being built at
that point.
But it's also a location that makes sense..
Because that allowed writing to symlinks of files that are not present,
which followed the link and put bad content in an object location.
fsck: Fix up .git/annex/object directory permissions.
This commit was sponsored by an anonymous bitcoin donor.
This works for both direct and indirect mode.
It may need some performance tuning.
Note that unlike git status, it only shows the status of the work tree, not
the status of the index. So only one status letter, not two .. and since
files that have been added and not yet committed do not differ between the
work tree and the index, they are not shown. Might want to add display of
the index vs the last commit eventually.
This commit was sponsored by an unknown bitcoin contributor, whose
contribution as been going up lately! ;)
Now that direct mode sets core.bare=true, git's normal prohibition about
pushing into the currently checked out branch doesn't work.
A simple fix for this would be an update hook which blocks the pushes..
but git hooks must be executable, and git-annex needs to be usable on eg,
FAT, which lacks x bits.
Instead, enabling direct mode switches the branch (eg master) to a special
purpose branch (eg annex/direct/master). This branch is not pushed when
syncing; instead any changes that git annex sync commits get written to
master, and it's pushed (along with synced/master) to the remote.
Note that initialization has been changed to always call setDirect,
even if it's just setDirect False for indirect mode. This is needed because
if the user has just cloned a direct mode repo, that nothing has synced
with before, it may have no master branch, and only a annex/direct/master.
Resulting in that branch being checked out locally too. Calling setDirect False
for indirect mode moves back out of this branch, to a new master branch,
and ensures that a manual "git push" doesn't push changes directly to
the annex/direct/master of the remote. (It's possible that the user
makes a commit w/o using git-annex and pushes it, but nothing I can do
about that really.)
This commit was sponsored by Jonathan Harrington.
This guarantees that stopping an existing socket never fails.
This might be the route out of the mess of needing to worry about socket
lengths in general. However, it would need quite a lot of refactoring
to make every place in git-annex that runs ssh run it with a cwd that was
determined by the location of its connection caching socket. If this
wasn't already such a mess, I'd consider even the thought of that API a bad
idea..
The control socket path passed to ssh needs to be 17 characters shorter
than the maximum unix domain socket length, because ssh appends stuff to it
to make a temporary filename. Closes: #725512
Also, take the shorter of the relative and the absolute paths to the
socket. Typically the relative path will be a lot shorter (unless
deep inside a subdirectory of the repository), and so using it will
avoid flirting with the maximum safe socket lenghts in more situations,
and so lead to less breakage if all my attempts at fixing this are
still buggy.
My implementation does not guard against double locking of the journal. But
it does ensure that the journal is always locked when operated on, by using
a type that is only produced by lockJournal, and which is required as a
parameter of all functions that operate on the journal.
Note that I had to add the fooStale functions for cases where it does not
make sense to lock the journal when querying it. I was more concerned about
ensuring that anything that modifies the journal is locked.
setJournalFile's implementation ensures that any query of the journal will
get one value or the other atomically, even if the journal is being changed
at the time.
This may not strictly be needed -- the transition code bypasses the
journal. However, this ensures that the git-annex branch is only
committed with the journal locked. This will allow for further
improvements.
Overridable with --user-agent option.
Not yet done for S3 or WebDAV due to limitations of libraries used --
nether allows a user-agent header to be specified.
This commit sponsored by Michael Zehrer.
Since 006cf7976f was incomplete, not being
able to get the right mode of the file when the index differs from HEAD,
this is a final workaround. Only buffering the start of the file
in this case avoids leaking memory.
This does not prevent git-cat-file being asked to output the whole file,
which needs to be consumed, and can be slow. But this only happens in a
rare edge case.
Done using a mode witness, which ensures it's fixed everywhere.
Fixing catFileKey was a bear, because git cat-file does not provide a
nice way to query for the mode of a file and there is no other efficient
way to do it. Oh, for libgit2..
Note that I am looking at tree objects from HEAD, rather than the index.
Because I cat-file cannot show a tree object for the index.
So this fix is technically incomplete. The only cases where it matters
are:
1. A new large file has been directly staged in git, but not committed.
2. A file that was committed to HEAD as a symlink has been staged
directly in the index.
This could be fixed a lot better using libgit2.
The second commit had some bad refs which resulted in the race detection
code running. But that commit was unnecessary anyway, it only was there to
merge in the other refs.
Wrote nice pure transition calculator, and ugly code to stage its results
into the git-annex branch. Also had to split up several Log modules
that Annex.Branch needed to use, but that themselves used Annex.Branch.
The transition calculator is limited to looking at and changing one file at
a time. While this made the implementation relatively easy, it precludes
transitions that do stuff like deleting old url log files for keys that are
being removed because they are no longer present anywhere.
Works, more or less. --dead is not implemented, and so far a new branch
is made, but keys no longer present anywhere are not scrubbed.
git annex sync fails to push the synced/git-annex branch after a forget,
because it's not a fast-forward of the existing synced branch. Could be
fixed by making git-annex sync use assistant-style sync branches.
When quvi is installed, git-annex addurl automatically uses it to detect
when an page is a video, and downloads the video file.
web special remote: Also support using quvi, for getting files,
or checking if files exist in the web.
This commit was sponsored by Mark Hepburn. Thanks!
Requires git 1.8.4 or newer. When it's installed, a background
git check-ignore process is run, and used to efficiently check ignores
whenever a new file is added.
Thanks to Adam Spiers, for getting the necessary support into git for this.
A complication is what to do about files that are gitignored but have
been checked into git anyway. git commands assume the ignore has been
overridden in this case, and not need any more overriding to commit a
changed version.
However, for the assistant to do the same, it would have to run git ls-files
to check if the ignored file is in git. This is somewhat expensive. Or it
could use the running git-cat-file process to query the file that way,
but that requires transferring the whole file content over a pipe, so it
can be quite expensive too, for files that are not git-annex
symlinks.
Now imagine if the user knows that a file or directory tree will be getting
frequent changes, and doesn't want the assistant to sync it, so gitignores
it. The assistant could overload the system with repeated ls-files checks!
So, I've decided that the assistant will not automatically commit changes
to files that are gitignored. This is a tradeoff. Hopefully it won't be a
problem to adjust .gitignore settings to not ignore files you want the
assistant to autocommit, or to manually git annex add files that are listed
in .gitignore.
(This could be revisited if git-annex gets access to an interface to check
the content of the index w/o forking a git command. This could be libgit2,
or perhaps a separate git cat-file --batch-check process, so it wouldn't
need to ship over the whole file content.)
This commit was sponsored by Francois Marier. Thanks!
Started with a problem when running addurl on a really long url,
because the whole url is munged into the filename. Ended up doing
a fairly extensive review for places where filenames could get too large,
although it's hard to say I'm not missed any..
Backend.Url had a 128 character limit, which is fine when the limit is 255,
but not if it's a lot shorter on some systems. So check the pathconf()
limit. Note that this could result in fromUrl creating different keys
for the same url, if run on systems with different limits. I don't see
this is likely to cause any problems. That can already happen when using
addurl --fast, or if the content of an url changes.
Both Command.AddUrl and Backend.Url assumed that urls don't contain a
lot of multi-byte unicode, and would fail to truncate an url that did
properly.
A few places use a filename as the template to make a temp file.
While that's nice in that the temp file name can be easily related back to
the original filename, it could lead to `git annex add` failing to add a
filename that was at or close to the maximum length.
Note that in Command.Add.lockdown, the template is still derived from the
filename, just with enough space left to turn it into a temp file.
This is an important optimisation, because the assistant may lock down
a bunch of files all at once, and using the same template for all of them
would cause openTempFile to iterate through the same set of names,
looking for an unused temp file. I'm not very happy with the relatedTemplate
hack, but it avoids that slowdown.
Backend.WORM does not limit the filename stored in the key.
I have not tried to change that; so git annex add will fail on really long
filenames when using the WORM backend. It seems better to preserve the
invariant that a WORM key always contains the complete filename, since
the filename is the only unique material in the key, other than mtime and
size. Since nobody has complained about add failing (I think I saw it
once?) on WORM, probably it's ok, or nobody but me uses it.
There may be compatability problems if using git annex addurl --fast
or the WORM backend on a system with the 255 limit and then trying to use
that repo in a system with a smaller limit. I have not tried to deal with
those.
This commit was sponsored by Alexander Brem. Thanks!
This is ok to do now that the socket filename never needs to be mapped back
to a hostname.
Short hostnames will still appear in the clear, which is less obfuscated.
So this cannot possibly make ssh connection caching fail for a hostname it
used to work for.
Turns out that with -O stop -S socketfile, ssh does not need the real
hostname, or port to be specificed. This is because it simply talks to the
ssh behind the socket and tells it to stop. So, can eliminate the
conversion back from a socketfile to host and port. Which will allow using
shorter filenames for sockets in the future.
If the file is > 8192 bytes, it's certianly not a symlink file.
And if it contains nuls or newlines or whitespace, it's certianly
not a link to annexed content. But it might be a tarball containing
a git-annex repo.
This hack is only needed on FAT filesystems, so there's no point in doing
it the rest of the time. And it's possible for there to be a false
positive, so it's best to avoid the hack when possible.
Test suite on windows failed running git annex init in a bare clone of an
annexed repo. The annex directory didn't exist when it tried to write the
inode sentinal file.
Yeah, that didn't actually work. Got error messages like it couldn't read
from the control socket, so probably ssh doesn't really support that on
Windows, at least the cygwin ssh build I'm using.
On Windows, that means the file could still be open when later code wants
to delete it, which fails. Since we're only reading 8k anyway, just read
it, strictly. However, avoid reading the whole file strictly, so no
getContentsStrict here.
If the cleanup of a single file fails for some reason, continue
to clean up other files.
This could happen because of a race. The merge pulls in a change to a file,
which gets changed locally at the same time.
Made fromDirect check that a file in the tree has good content (and is not
a broken symlink either) before copying it to another file that has the
same key.
Made replaceFile clean up the temp file if the action that creates it, or
the file replacement action fails.
This was also tripped by the test suite's automatic conflict resolution
test. Which also shows BTW that an unnecessary copy of content is done
sometimes when merging in direct mode. Not going to try to speed that up
now.
This bug was turned up by the test suite, running fsck in direct mode.
A repository was cloned, was put into direct mode, was fscked, and fsck
incorrectly said that no copy existed of a file, that was actually present
in origin.
This turned out to occur because fsck first did a Annex.Branch.change,
recording that it did not locally have the file. That was recorded in the
journal. Since neither the git annex direct not the fsck had yet needed to
read any info from the branch, but had only made changes to it, the
origin/git-annex branch was not yet merged in. So the journal got a
location log entry written to it, but this did not include
the location log info for the origin. When fsck then did a
Annex.Branch.get, it trusted the journal was cosnsitent, and returned it,
again w/o merging from origin/git-annex. This latter behavior is the
actual bug.
Refer to commit e9bfa8eaed for the thinking
behind it being ok to make a change to a file on the branch, without
first merging the branch. That thinking still stands. However, it means
that files in the journal cannot be trusted to be consistent if the branch
has not been merged. So, to fix, just enure the branch gets merged, even
when reading from the journal.
In tests, this does not seem to cause any extra merging. Except, of course,
in the one case described above. But git annex add, etc, are able to make
changes w/o first merging the branch.
The bug was in movein, which just replaceFile'd the file with a symlink,
even if it already had the desired content, before trying to pull the
content out of the annex and replace the symlink with it.
That was ok-ish for non conflicted merges, where if the file existed it would
be an old version of the content. But for conflicted merges, the automatic
merge resolver has already run, and will have already put the desired
content into the file for the local variant.
Also, made removeDirect not trust that the associated files map is correct.
Only if it can verify that another file has the content will it not move it
into .git/annex/objects.
As seen in this bug report, the lifted exception handling using the StateT
monad throws away state changes when an action throws an exception.
http://git-annex.branchable.com/bugs/git_annex_fork_bombs_on_gpg_file/
.. Which can result in cached values being redundantly calculated, or other
possibly worse bugs when the annex state gets out of sync with reality.
This switches from a StateT AnnexState to a ReaderT (MVar AnnexState).
All changes to the state go via the MVar. So when an Annex action is
running inside an exception handler, and it makes some changes, they
immediately go into affect in the MVar. If it then throws an exception
(or even crashes its thread!), the state changes are still in effect.
The MonadCatchIO-transformers change is actually only incidental.
I could have kept on using lifted-base for the exception handling.
However, I'd have needed to write a new instance of MonadBaseControl
for the new monad.. and I didn't write the old instance.. I begged Bas
and he kindly sent it to me. Happily, MonadCatchIO-transformers is
able to derive a MonadCatchIO instance for my monad.
This is a deep level change. It passes the test suite! What could it break?
Well.. The most likely breakage would be to code that runs an Annex action
in an exception handler, and *wants* state changes to be thrown away.
Perhaps the state changes leaves the state inconsistent, or wrong. Since
there are relatively few places in git-annex that catch exceptions in the
Annex monad, and the AnnexState is generally just used to cache calculated
data, this is unlikely to be a problem.
Oh yeah, this change also makes Assistant.Types.ThreadedMonad a bit
redundant. It's now entirely possible to run concurrent Annex actions in
different threads, all sharing access to the same state! The ThreadedMonad
just adds some extra work on top of that, with its own MVar, and avoids
such actions possibly stepping on one-another's toes. I have not gotten
rid of it, but might try that later. Being able to run concurrent Annex
actions would simplify parts of the Assistant code.
Before, if a direct mode repo had one or more associated files that
were modifed, moving the object into it would overwrite the associated
files with the pristine object.
Now, modified associated files are left unchanged. To ensure that,
when an object is moved into a direct mode repo, it's not thrown away,
it gets stored in indirect mode.
This fixes a bug with git annex add in direct mode. If some files already
existed in the tree pointing at the same key as a file that was just added,
and their content was not present, add neglected to copy the content to
those files.
I also changed the behavior of moveAnnex slightly: When content is moved
into the annex in direct mode, it does not overwrite any content already
present in direct mode files. That content may be modified after all.
On Windows with Cygwin, checking out a git-annex repo will create symlinks
on disk, so we need to always try to read the symlink, even when
core.symlinks says they're not supported.
Seems that Windows doesn't allow deleting a file that the same process has open.
Here the inode cache file was read and a the value from it gets used later.
But due to laziness, the old file is still open when it gets deleted. Adding
strictness avoids this problem. Of course, the file is small, so it's no
problem to read it all strictly, so this is probably an improvement even
outside of Windows.
That's needed in files used to build the configure program.
For the other files, I'm keeping my __WINDOWS__ define, as I find that much easier to type.
I may search and replace it to use the mingw32_HOST_OS thing later.
It's possible for files in indirect mode to have a direct mode mapping
file. Probably from when they were in direct mode. In this case,
toDirectGen tried to copy the content from the direct mode file that the
mapping said had it. But, being in indirect mode, it didn't really have the
content. So it did nothing. This fix makes it always move the content from
.git/annex/objects/ when it's there.
I don't know why, but I can't seem to set the environment variables inside
git-annex to work around the git error caused by android's crappy username
and hostname settings. This workaround works, and that's all that's good
about it.
Git fails on Android, because it gets some weird domain for local host like
"localhost.(none)". This works around that. I made it always set EMAIL when
GECOS workaround was needed (unless EMAIL is already set). It might be
nicer to try to get the hostname.domain as git does, and only set it if
that fails. But I don't want to be stuck trying to exactly duplicate
whatever git is doing.
A content directory can be frozen in direct mode. One way this can happen
is if the content is transferred before direct mode has a mapping for it,
so it's stored in the content directory.
So, we need to thaw the content directory before doing things with it.
The root of the problem is that toInodeCache sees a non-symlink, and so
goes on and generates a new inode cache for the dummy symlink.
Any place that toInodeCache, or sameFileStatus, or genInodeCache are called
may need to deal with this case. Although many of them are ok. For example,
prepSendAnnex calls sameInodeCache, which calls genInodeCache.. but if
the file content is not present, the InodeCache generated for its standin
file is appropriately not the same, and so it returns Nothing.
I've audited some, but have to say I'm not happy with this; it should be
handled at the type level somehow, or a toInodeCache wrapper be used that
is aware of dummy symlinks.
(The Watcher already dealt with it, via the guardSymlinkStandin function.)
Introduced a new per-remote option 'annex-rsync-transport' to specify
the remote shell that it to be used with rsync. In case the value is
'ssh', connections are cached unless 'sshcaching' is unset.
This looks at the string one char at a time, which is hardly efficient..
but more than good enough for expanding variables in
relatively short command lines.
Fixed by storing a list of cached inodes for a key, instead of just one.
Backwards compatability note: An old git-annex version will fail to parse
an inode cache file that has been written by a new version, and has
multiple items. It will succees if just one. So old git-annexes will have
even worse behavior when there are duplicated files, if that is possible.
I don't think it will be a problem. (Famous last words.)
Also, note that it doesn't expire old and unused inode caches for a key.
It would be possible to add this if needed; just look through the
associated files for a key and if there are more cached inodes, throw out
any not corresponding to associated files. Unless a file is being copied
repeatedly and the old copy deleted, this lack of expiry should not be a
problem.
* since this is a crippled filesystem anyway, git-annex doesn't use
symlinks on it
* so there's no reason to use the mixed case hash directories that we're
stuck using to avoid breaking everyone's symlinks to the content
* so we can do what is already done for all bare repos, and make non-bare
repos on crippled filesystems use the all-lower case hash directories
* which are, happily, all 3 letters long, so they cannot conflict with
mixed case hash directories
* so I was able to 100% fix this and even resuming `git annex add` in the
test case will recover and it will all just work.
This avoids commit churn by the assistant when eg,
replacing a file with a symlink.
But, just as importantly, it prevents the working tree being left with a
deleted file if git-annex, or perhaps the whole system, crashes at the
wrong time.
(It also probably avoids confusing displays in file managers.)
I would have sort of liked to put this in .gitattributes, but it seems
it does not support multi-word attribute values. Also, making this a single
config setting makes it easy to only parse the expression once.
A natural next step would be to make the assistant `git add` files that
are not annex.largefiles. OTOH, I don't think `git annex add` should
`git add` such files, because git-annex command line tools are
not in the business of wrapping git command line tools.
There are two types of equality here, and which one is right varies,
so this forces me to consider and choose between them.
Based on this, I learned that the commit in git anex sync was
always doing a strong comparison, even when in a repository where
the inodes had changed. Fixed that.
This fixes the issue mentioned in the last commit.
Turns out just collecting UUID of clients behind a XMPP remote is
insufficient (although I should probably still do it for other reasons),
because a single remote repo might be connected via both XMPP and local
pairing. So a way is needed to know when a push was received from any
client using a given XMPP remote over XMPP, as opposed to via ssh.
Make manualPull send push requests over XMPP.
When reconnecting with remotes, those that are XMPP remotes cannot
immediately be pulled from and scanned, so instead maintain a set of
(probably) desynced remotes, and put XMPP remotes on it. (This set could be
used in other ways later, if we can detect we're out of sync with other
types of remotes.)
The merger handles detecting when a XMPP push is received from a desynced
remote, and triggers a scan then, if they have in fact diverged.
This has one known bug: A single XMPP remote can have multiple clients
behind it. When this happens, only the UUID of one client is recorded
as the UUID of the XMPP remote. Pushes from the other XMPP clients will not
trigger a scan. If the client whose UUID is expected responds to the push
request, it'll work, but when that client is offline, we're SOL.
Pass subcommand as a regular param, which allows passing git parameters
like -c before it. This was already done in the pipeing set of functions,
but not the command running set.
The only thing lost is ./ghci
Speed: make fast used to take 20 seconds here, when rebuilding from
touching Command/Unused.hs. With cabal, it's 29 seconds.
This resulted in a lot of user complains that git annex init had git
telling them they needed to run git config --global user.email .. which
didn't work because even HOME was not passed into git.
Refactored annex link code into nice clean new library.
Audited and dealt with calls to createSymbolicLink.
Remaining calls are all safe, because:
Annex/Link.hs: ( liftIO $ createSymbolicLink linktarget file
only when core.symlinks=true
Assistant/WebApp/Configurators/Local.hs: createSymbolicLink link link
test if symlinks can be made
Command/Fix.hs: liftIO $ createSymbolicLink link file
command only works in indirect mode
Command/FromKey.hs: liftIO $ createSymbolicLink link file
command only works in indirect mode
Command/Indirect.hs: liftIO $ createSymbolicLink l f
refuses to run if core.symlinks=false
Init.hs: createSymbolicLink f f2
test if symlinks can be made
Remote/Directory.hs: go [file] = catchBoolIO $ createSymbolicLink file f >> return True
fast key linking; catches failure to make symlink and falls back to copy
Remote/Git.hs: liftIO $ catchBoolIO $ createSymbolicLink loc file >> return True
ditto
Upgrade/V1.hs: liftIO $ createSymbolicLink link f
v1 repos could not be on a filesystem w/o symlinks
Audited and dealt with calls to readSymbolicLink.
Remaining calls are all safe, because:
Annex/Link.hs: ( liftIO $ catchMaybeIO $ readSymbolicLink file
only when core.symlinks=true
Assistant/Threads/Watcher.hs: ifM ((==) (Just link) <$> liftIO (catchMaybeIO $ readSymbolicLink file))
code that fixes real symlinks when inotify sees them
It's ok to not fix psdueo-symlinks.
Assistant/Threads/Watcher.hs: mlink <- liftIO (catchMaybeIO $ readSymbolicLink file)
ditto
Command/Fix.hs: stopUnless ((/=) (Just link) <$> liftIO (catchMaybeIO $ readSymbolicLink file)) $ do
command only works in indirect mode
Upgrade/V1.hs: getsymlink = takeFileName <$> readSymbolicLink file
v1 repos could not be on a filesystem w/o symlinks
Audited and dealt with calls to isSymbolicLink.
(Typically used with getSymbolicLinkStatus, but that is just used because
getFileStatus is not as robust; it also works on pseudolinks.)
Remaining calls are all safe, because:
Assistant/Threads/SanityChecker.hs: | isSymbolicLink s -> addsymlink file ms
only handles staging of symlinks that were somehow not staged
(might need to be updated to support pseudolinks, but this is
only a belt-and-suspenders check anyway, and I've never seen the code run)
Command/Add.hs: if isSymbolicLink s || not (isRegularFile s)
avoids adding symlinks to the annex, so not relevant
Command/Indirect.hs: | isSymbolicLink s -> void $ flip whenAnnexed f $
only allowed on systems that support symlinks
Command/Indirect.hs: whenM (liftIO $ not . isSymbolicLink <$> getSymbolicLinkStatus f) $ do
ditto
Seek.hs:notSymlink f = liftIO $ not . isSymbolicLink <$> getSymbolicLinkStatus f
used to find unlocked files, only relevant in indirect mode
Utility/FSEvents.hs: | Files.isSymbolicLink s = runhook addSymlinkHook $ Just s
Utility/FSEvents.hs: | Files.isSymbolicLink s ->
Utility/INotify.hs: | Files.isSymbolicLink s ->
Utility/INotify.hs: checkfiletype Files.isSymbolicLink addSymlinkHook f
Utility/Kqueue.hs: | Files.isSymbolicLink s = callhook addSymlinkHook (Just s) change
all above are lower-level, not relevant
Audited and dealt with calls to isSymLink.
Remaining calls are all safe, because:
Annex/Direct.hs: | isSymLink (getmode item) =
This is looking at git diff-tree objects, not files on disk
Command/Unused.hs: | isSymLink (LsTree.mode l) = do
This is looking at git ls-tree, not file on disk
Utility/FileMode.hs:isSymLink :: FileMode -> Bool
Utility/FileMode.hs:isSymLink = checkMode symbolicLinkMode
low-level
Done!!
Now getKeysPresent checks that the key's content, not only its directory,
exists. In direct mode, the inode cache file is used as a standin for the
content.
removeAnnex always removes the inode cache file, and drop and move --from
always call removeAnnex, even if the object does not seem to be inAnnex,
to ensure it's always deleted.
This reverts commit 57780cb3a4.
This was buggy, it caused the direct mode cache to be lost when dropping
keys, so when the file is gotten back, it's stored in indirect mode.
Note to self: Do not attempt bug fixes at 6 am!
In indirect mode, now checks the inode cache to detect changes to a file.
Note that a file can still be changed if a process has it open for write,
after landing in the annex.
In direct mode, some checking of the inode cache was done before, but
from a much later point, so fewer modifications could be detected. Now it's
as good as indirect mode.
On crippled filesystems, no lock down is done before starting to add a
file, so checking the inode cache is the only protection we have.
git annex init probes for crippled filesystems, and sets direct mode, as
well as `annex.crippledfilesystem`.
Avoid manipulating permissions of files on crippled filesystems.
That would likely cause an exception to be thrown.
Very basic support in Command.Add for cripped filesystems; avoids the lock
down entirely since doing it needs both permissions and hard links.
Will make this better soon.
Been meaning to do this for some time; Android port was last straw.
Note that newer versions of the uuid library have a Data.UUID.V4 that
generates random UUIDs slightly more cleanly, but Debian has an old version
of the library, so I do it slightly round-about.
These files were left behind, and made getKeysPresent find keys that were
not present. It would be expensive to make getKeysPresent check that the
actual key files are present (it just lists the directories). But that's not
needed if we just clean up the stale cache and mapping files.
To handle systems that were in direct mode and got switched back with stale
direct mode files, made cleanObjectLoc remove all files in the key's directory.
git annex unused will still list keys that are gone but for which the stale
direct mode files exists. To deal with that, made dropunused remove the key's
directory even if the key does not seem to be present.
The most common way for a mapping to be stale is when a file was deleted,
or renamed. Nothing updates the mappings for deletions yet.
But they can also become stale in other ways. For example a file can
be modified.
So, the mapping is not trusted to be consistent. When we get a key,
only replace symlinks that still point to that key with its content.
When we drop a key, only put back symlinks for files that still have
the direct mode content.
Avoid a crash if a mapping contains files that no longer exist.
This could happen because eg, one was deleted and a commit has not yet been
done to update the mapping.
Fix path calculation.
Sometimes it seems that git-cat-file --batch stops getting info for
files in the current repo, when ":file" is fed to it. I have not reproduced
this at the command line, but only when using git annex whereis and git
annex move inside a direct mode repo. Those failed, because cat-file
returned "file missing". OTOH, git annex find works fine, despite passing
the same file to cat-file. It seems that the failing commands first asked
cat-file to show a file on the git-annex branch. Perhaps it got "stuck" on
that branch? But I cannot repoduce it running cat-file by hand. Most
strange. HEAD is a workaround for this extreme weirdness, since I spent a
good 2 hours struggling with it already.
The expensive scan uses lookupFile, but in direct mode, that doesn't work
for files that are present. So the scan was not finding things that are
present that need to be uploaded. (It did find things not present that
needed to be downloaded.)
Now lookupFile also works in direct mode. Note that it still prefers
symlinks on disk to info committed to git, in direct mode. This is
necessary to make things like Assistant.Threads.Watcher.onAddSymlink
work correctly, when given a new symlink not yet checked into git (or
replacing a file checked into git).
Now there's a Config type, that's extracted from the git config at startup.
Note that laziness means that individual config values are only looked up
and parsed on demand, and so we get implicit memoization for all of them.
So this is not only prettier and more type safe, it optimises several
places that didn't have explicit memoization before. As well as getting rid
of the ugly explicit memoization code.
Not yet done for annex.<remote>.* configuration settings.
When a file is changed in direct mode, the old content is probably lost
(at least from the local repo), and bookeeping needs to be updated to
reflect this.
Also, synthetic add events are generated at assistant startup, so
make it detect when the file has not really changed, and avoid re-adding
it.
This does add the overhead of querying the runing git cat-file for the
key that's recorded in git for the file, each time a file is added or
modified in direct mode.
However, I don't yet have a reliable way to deal with files being modified
while they're being transferred. I have code that detects it on the sending
side, but the receiver is still free to move the wrong content into its
annex, and record that it has the content. So that's not acceptable, and
I'll need to work on it some more.
However, at this point I can use a direct mode repository as a remote and
transfer files from and to it.
Also for dropping objects in direct mode.
Checking presence reliably needs a cache of mtime, size, and inode.
This way, if a file is modified, keys that point to it are no longer
present.
Also, the code for restoring the symlink when removing objects is
unnecessarily messy. calcGitLink was generating links starting with
"../../remote/.git/", when running "git annex move --from remote".
I put in a workaround, but calcGitLink should probably be fixed.
There is not yet support for getting objects from repositories in direct
mode; it still looks for content in .git/annex/objects, and there's no
once place I can change to fix that.
Also, getting objects from direct mode repositories is problematic since
the can be changed while the object is being transferred. It probably needs
to quarantine it first.
* get/copy --auto: Transfer data even if it would exceed numcopies,
when preferred content settings want it.
* drop --auto: Fix dropping content when there are no preferred content
settings.
This cache prevented noticing changes made by another process.
The case I just ran into involved the assistant dropping a file, which
cached its presence info. Then the same file was downloaded again,
but the assistant didn't know its presence info had changed.
I don't see a way to keep this cache. Will instead rely on the OS level
file cache, for files in the journal. May need to add more higher-level
caching of info that it's ok to have a potentially stale copy of,
although much of git-annex already does so.
When in a subdir, both the normal filepath, and the filepath relative to
the top of the git repo are needed for matching. The former for key lookup,
and the latter for include/exclude to match against. Previously, key lookup
didn't work in this situation.
The standalone build does not bundle its own ssh, so should be built
to support as wide an array of ssh versions as possible, so turn off
connection caching.
Unfortunatly, as implemented this forces a full rebuild when building the
standalone binary, and of course it makes it somewhat slower.
This is not ideal, but neither is probing the ssh version every time it's
run (slow), or once when initializing a repo (fragile).
Setting GIT_INDEX_FILE clobbers the rest of the environment, making git
not read ~/.gitconfig, and blow up if GECOS didn't have a name for the
user.
I'm not entirely happy with getEnvironment being run every time now,
that's somewhat expensive. It may make sense to just set GIT_COMMITTER_*
and GIT_AUTHOR_*, but I worry that clobbering the rest could break PATH,
or GIT_PATH, or something else that might be used by a command run in here.
And caching the environment is not a good idea either; it can change..
I'm using transfer for most things, both removable drives and cloud
storage, because it's the safest choice. We'll see if it makes sense
to prompt for the group when setting this up, or let the user pick
something else after the fact.
Both when queueing downloads, and uploads, consults the preferred content
settings.
I didn't make it check yet when requeing failed transfers or queuing
deferred downloads; dealing with the preferred content settings (or indeed,
other settings) changing while the assistant is running still needs work.
I'm down to 9 places in the code that can produce unwaited for zombies.
Most of these are pretty innocuous, at least for now, are only
used in short-running commands, or commands that run a set of
actions and explicitly reap zombies after each one.
The one from Annex.Branch.files could be trouble later,
since both Command.Fsck and Command.Unused can trigger it,
and the assistant will be doing those eventally. Ditto the one in
Git.LsTree.lsTree, which Command.Unused uses.
The only ones currently affecting the assistant though, are
in Git.LsFiles. Several threads use several of those.
(And yeah, using pipes or ResourceT would be a less ad-hoc approach,
but I don't really feel like ripping my entire code base apart right
now to change a foundation monad. Maybe one of these days..)
Fix resuming of downloads, which do not have a transfer info file to read.
When checking upload progress, use the MVar, rather than re-reading
the info file.
Catch exceptions in the transfer action. Required a tryAnnex.
Branch.get is not able to see changes that have been staged to the index
but not committed. This is a limitation of git cat-file --batch; when
reading from the index, as opposed to from a branch, it does not notice
changes made after the first time it reads the index.
So, had to revert the changes made in 1f73db3469
to make annex.alwayscommit=false stage changes.
Also, ensure that Branch.change and Branch.get always see changes
at all points during a commit, by not deleting journal files when
staging to the index. Delete them only after committing the branch.
Before, there was a race during commits where a different git-annex
could see out-of-date info from the branch while a commit was in progress.
That's also done when updating the branch to merge in remote branches.
In the case where the local git-annex branch has had changes pushed into it
that are not yet reflected in the index, and there are journalled changes
as well, a merge commit has to be done.
Commits used to be made to the git-annex branch whenever there were
journalled changes from a previous command, and the current command looked
up the value of a file. This no longer happens.
This means that transferkey, which is a oneshot command that stages
changes, can be run multiple times by the assistant, without each of them
committing the changes made by the command before. Which will be a lot
faster and use less space by batching up the commits.
Commits still happen if a remote git-annex branch has been changed and is
merged in.
Found a very cheap way to determine when a disconnected remote has
diverged, and has new content that needs to be transferred: Piggyback on
the git-annex branch update, which already checks for divergence.
However, this does not check if new content has appeared locally while
disconnected, that should be transferred to the remote.
Also, this does not handle cases where the two git repos are in sync,
but their content syncing has not caught up yet.
This code could have its efficiency improved:
* When multiple remotes are synced, if any one has diverged, they're
all queued for transfer scans.
* The transfer scanner could be told whether the remote has new content,
the local repo has new content, or both, and could optimise its scan
accordingly.
Make Utility.Process wrap the parts of System.Process that I use,
and add debug logging to them.
Also wrote some higher-level code that allows running an action
with handles to a processes stdin or stdout (or both), and checking
its exit status, all in a single function call.
As a bonus, the debug logging now indicates whether the process
is being run to read from it, feed it data, chat with it (writing and
reading), or just call it for its side effect.
Test suite now passes with -threaded!
I traced back all the hangs with -threaded to System.Cmd.Utils. It seems
it's just crappy/unsafe/outdated, and should not be used. System.Process
seems to be the cool new thing, so converted all the code to use it
instead.
In the process, --debug stopped printing commands it runs. I may try to
bring that back later.
Note that even SafeSystem was switched to use System.Process. Since that
was a modified version of code from System.Cmd.Utils, it needed to be
converted too. I also got rid of nearly all calls to forkProcess,
and all calls to executeFile, which I'm also doubtful about working
well with -threaded.
While I was in there, I noticed and fixed a bug in the queue size
calculations. It was never encountered only because Queue.add was
only ever run with 1 file in the list.
There's a race adding a new file to the annex: The file is moved to the
annex and replaced with a symlink, and then we git add the symlink. If
someone comes along in the meantime and replaces the symlink with
something else, such as a new large file, we add that instead. Which could
be bad..
This race is fixed by avoiding using git add, instead the symlink is
directly staged into the index.
It would be nice to make `git annex add` use this same technique.
I have not done so yet because it currently runs git update-index once per
file, which would slow does `git annex add`. A future enhancement would be
to extend the Git.Queue to include the ability to run update-index with
a list of Streamers.
This allows the queue to be used in a single process for multiple possibly
conflicting commands, like add and rm, without running them out of order.
This assumes that running the same git subcommand with different parameters
cannot itself conflict.
Baked into the code was an assumption that a repository's git directory
could be determined by adding ".git" to its work tree (or nothing for bare
repos). That fails when core.worktree, or GIT_DIR and GIT_WORK_TREE are
used to separate the two.
This was attacked at the type level, by storing the gitdir and worktree
separately, so Nothing for the worktree means a bare repo.
A complication arose because we don't learn where a repository is bare
until its configuration is read. So another Location type handles
repositories that have not had their config read yet. I am not entirely
happy with this being a Location type, rather than representing them
entirely separate from the Git type. The new code is not worse than the
old, but better types could enforce more safety.
Added support for core.worktree. Overriding it with -c isn't supported
because it's not really clear what to do if a git repo's config is read, is
not bare, and is then overridden to bare. What is the right git directory
in this case? I will worry about this if/when someone has a use case for
overriding core.worktree with -c. (See Git.Config.updateLocation)
Also removed and renamed some functions like gitDir and workTree that
misused git's terminology.
One minor regression is known: git annex add in a bare repository does not
print a nice error message, but runs git ls-files in a way that fails
earlier with a less nice error message. This is because before --work-tree
was always passed to git commands, even in a bare repo, while now it's not.
annex.ssh-options, annex.rsync-options, annex.bup-split-options.
And adjust types to avoid the bugs that broke several config settings
recently. Now "annex." prefixing is enforced at the type level.
A bit tricky to avoid printing it twice in a row when there are queued git
commands to run and journal to stage.
Added a generic way to run an action that may output multiple side
messages, with only the first displayed.
This is incomplete, it does not honor it yet for hash directories
and other annex bookkeeping files. Some of that is not needed for a bare
repo; some of it may be.
getConfig got a remote-specific config, and this confusing name caused it
to be used a couple of places that only were interested in global configs.
Rename to getRemoteConfig and make getConfig only get global configs.
There are no behavior changes here, but remote.<name>.annex-web-options
never actually worked (and per-remote web options is a very unlikely to be
useful case so I didn't make it work), so fix the documentation for it.
Don't check if configure indicated checks won't work. This should fix a
FTBFS on mipsel, where configure correctly detects the checks won't work,
while garbage is returned for disk space info at git-annex runtime. It also
means that, when built via cabal, disk space checks are not enabled,
unfortunatly.
.. Allowing it to be used by things in constant space!
Random statistics: git annex status has gone from taking 239 mb
of memory and 26 seconds in a repo, to 8 mb and 13 seconds.
The trick here is the unsafeInterleaveIO, and the form of the function's
recursion, which I cribbed heavily from System.IO.HVFS.Utils.recurseDirStat.
The difference is, this one goes to a limited depth and avoids statting
everything.
Before, it leaked space due to caching lists of keys. Now all necessary
data about keys is calculated as they stream in.
The "nearly constant" is due to getKeysPresent, which builds up a lot
of [] thunks as it traverses .git/annex/objects/. Will deal with it later.
Now changes are staged into the branch's index, but not committed,
which avoids growing a large journal. And sync and merge always
explicitly commit, ensuring that even when they do nothing else,
they commit the staged changes.
Added a flag file to indicate that the branch's journal contains
uncommitted changes. (Could use git ls-files, but don't want to run
that every time.)
In the future, this ability to have uncommitted changes staged in the
journal might be used on remotes after a series of oneshot commands.
To avoid commits of data to the git-annex branch after each command
is run, set annex.alwayscommit=false. Its data will then be committed
less frequently, when a merge or sync is done.
I was able to reproduce this on linux using the kernel's nfs server and
mounting localhost:/. Determined that removing the directory fails when
the just-deleted file in it was locked. Considered dropping the lock
before removing the directory, but this would complicate parts of the code
that should not need to worry about locking. So instead, ignore the failure
to remove the directory in this case.
While I was at it, made it attempt to remove both levels of hash
directories, in case they're empty.
useful when adding hundreds of thousands of files on a system with plenty
of memory.
git add gets quite slow in such a large repository, so if the system has
more than the ~32 mb of memory the queue can use by default, it's a useful
optimisation to increase the queue size, in order to decrease the number
of times git add is run.
The list of files had to be retained until the end so it could be deleted.
Also, a list of update-index lines was generated and only then fed into it.
Now everything streams in constant space.
When hashing the files, the entire list of shas was read strictly.
That was entirely unnecessary, since there's a cleanup action run
after they're consumed.
Now gitattributes are looked up, efficiently, in only the places that
really need them, using the same approach used for cat-file.
The old CheckAttr code seemed very fragile, in the way it streamed files
through git check-attr.
I actually found that cad8824852
was still deadlocking with ghc 7.4, at the end of adding a lot of files.
This should fix that problem, and avoid future ones.
The best part is that this removes withAttrFilesInGit and withNumCopies,
which were complicated Seek methods, as well as simplfying the types
for several other Seek methods that had a Backend tupled in.
Done by adding a oneshot mode, in which location log changes are written to
the journal, but not committed. Taking advantage of git-annex's existing
ability to recover in this situation.
This is used by git-annex-shell and other places where changes are made to
a remote's location log.
Ssh connection caching is now enabled automatically by git-annex. Only one
ssh connection is made to each host per git-annex run, which can speed some
things up a lot, as well as avoiding repeated password prompts. Concurrent
git-annex processes also share ssh connections. Cached ssh connections are
shut down when git-annex exits.
Note: The rsync special remote does not yet participate in the ssh
connection caching.
For a local git remote, can symlink the file.
For a git remote using rsync, can preseed any local content.
There are a few reasons to use fsck --from on a normal git remote.
One is if it's using gitosis or similar, and you don't have shell access
to run git annex locally. Another reason could be if you just want to
fsck certian files of a bare remote.
This way, the build log will indicate whether StatFS can be relied on.
I've tested all the failing architectures now, and on all of them,
the StatFS code now returns Nothing, rather than Just nonsense.
Also, if annex.diskreserve is set on a platform where StatFS is not
working, git-annex will complain.
Also, the Makefile was missing the sources target used when building with
cabal.
This needs to run git log on the location log files to get at all past
versions of the file, which tends to be a bit slow.
It would be possible to make a version optimised for showing the location
logs for every key. That would only need to run git log once, so would be
faster, but it would need to process an enormous amount of data, so
would not speed up the individual file case.
In the future it would be nice to support log --format. log --json also
doesn't work right yet.
Could have just used hGetContentsStrict here, but that would require
storing all the shas in memory. Since this is called at the end of a
git-annex run, it may have created a *lot* of shas, so I avoid that memory
use and stream them out like before.
Dealing with a race without using locking is exceedingly difficult and tricky.
Fully tested, I hope.
There are three places left where the branch can be updated, that are not
covered by the race recovery code. Let's prove they're all immune to the
race:
1. tryFastForwardTo checks to see if a fast-forward can be done,
and then does git-update-ref on the branch to fast-forward it.
If a push comes in before the check, then either no fast-forward
will be done (ok), or the push set the branch to a ref that can
still be fast-forwarded (also ok)
If a push comes in after the check, the git-update-ref will
undo the ref change made by the push. It's as if the push did not come
in, and the next git-push will see this, and try to re-do it.
(acceptable)
2. When creating the branch for the very first time, an empty index
is created, and a commit of it made to the branch. The commit's ref
is recorded as the current state of the index. If a push came in
during that, it will be noticed the next time a commit is made to the
branch, since the branch will have changed. (ok)
3. Creating the branch from an existing remote branch involves making
the branch, and then getting its ref, and recording that the index
reflects that ref.
If a push creates the branch first, git-branch will fail (ok).
If the branch is created and a racing push is then able to change it
(highly unlikely!) we're still ok, because it first records the ref into
the index.lck, and then updating the index. The race can cause the
index.lck to have the old branch ref, while the index has the newly pushed
branch merged into it, but that only results in an unnecessary update of
the index file later on.
The last branch ref that the index was updated to is stored in
.git/annex/index.lck, and the index only updated when the current
branch ref differs.
(The .lck file should later be used for locking too.)
Some more optimization is still needed, since there is some redundancy in
calls to git show-ref.
Always merge the git-annex branch into .git/annex/index before making a
commit from the index.
This ensures that, when the branch has been changed in any way
(by a push being received, or changes pulled directly into it, or
even by the user checking it out, and committing a change), the index
reflects those changes.
This is much too slow; it needs to be optimised to only update the
index when the branch has really changed, not every time.
Also, there is an unhandled race, when a change is made to the branch
right after the index gets updated. I left it in for now because it's
unlikely and I didn't want to complicate things with additional locking
yet.
I had to, I hope temporarily, lose my nice Annex newtype, and use a type
synonym. This because I cannot find a way to derive a MonadBaseControl
instance of the Annex newtype. I've emailed Bas van Dijk in hope he can
help get the newtype back.
Otherwise appears to build & work.
Supporting multiple directory hash types will allow converting to a
different one, without a flag day.
gitAnnexLocation now checks which of the possible locations have a file.
This means more statting of files. Several places currently use
gitAnnexLocation and immediately check if the returned file exists;
those need to be optimised.
The only fully supported thing is to have the main repository on one disk,
and .git/annex on another. Only commands that move data in/out of the annex
will need to copy it across devices.
There is only partial support for putting arbitrary subdirectories of
.git/annex on different devices. For one thing, but this can require more
copies to be done. For example, when .git/annex/tmp is on one device, and
.git/annex/journal on another, every journal write involves a call to
mv(1). Also, there are a few places that make hard links between various
subdirectories of .git/annex with createLink, that are not handled.
In the common case without cross-device, the new moveFile is actually
faster than renameFile, avoiding an unncessary stat to check that a file
(not a directory) is being moved. Of course if a cross-device move is
needed, it is as slow as mv(1) of the data.
The branch may not exist, if .git/annex has been copied over from another
repo (or a corrupted repo). I suppose it could also have gotten deleted
somehow. Without this, there is a confusing failure.
In git, a Ref can be a Sha, or a Branch, or a Tag. I added type aliases for
those. Note that this does not prevent mixing up of eg, refs and branches
at the type level. Since git really doesn't care, except rare cases like
git update-ref, or git tag -d, that seems ok for now.
There's also a tree-ish, but let's just use Ref for it. A given Sha or Ref
may or may not be a tree-ish, depending on the object type, so there seems
no point in trying to represent it at the type level.
Before, a merge was first calculated, by running various actions that
called git and built up a list of lines, which were at the end sent
to git update-index. This necessarily used space proportional to the size
of the diff between the trees being merged.
Now, lines are streamed into git update-index from each of the actions in
turn.
Runtime size of git-annex merge when merging 50000 location log files
drops from around 100 mb to a constant 4 mb.
Presumably it runs quite a lot faster, too.
Avoids doing auto-merging in commands that don't need fully current
information from the git-annex branch. In particular, git annex add no
longer needs to auto-merge. Affected commands: Anything that doesn't
look up data from the branch, but does write a change to it.
It might seem counterintuitive that we can change a value without first
making sure we have the current value. This optimisation works because
these two sequences are equivilant:
1. pull from remote
2. union merge
3. read file from branch
4. modify file and write to branch
vs.
1. read file from branch
2. modify file and write to branch
3. pull from remote
4. union merge
After either sequence, the git-annex branch contains the same logical content
for the modified file. (Possibly with lines in a different order or
additional old lines of course).
git-annex-shell inannex now returns always 0, 1, or 100 (the last when
it's unclear if content is currently in the index due to it currently being
moved or dropped).
(Actual locking code still not yet written.)
The lock will only persist during the perform stage, so the content must
be removed from the annex then, rather than in the cleanup stage.
(No lock is actually taken yet.)
Many functions took the repo as their first parameter. Changing it
consistently to be the last parameter allows doing some useful things with
currying, that reduce boilerplate.
In particular, g <- gitRepo is almost never needed now, instead
use inRepo to run an IO action in the repo, and fromRepo to get
a value from the repo.
This also provides more opportunities to use monadic and applicative
combinators.
Avoid ever using read to parse a non-haskell formatted input string.
show :: Key is arguably still show abuse, but displaying Keys as filenames
is just too useful to give up.
Thanks Valentin Haenel for a test case showing how non-fast-forward merges
could result in an ongoing pull/merge/push cycle.
While the git-annex branch is fast-forwarded, git-annex's index file is still
updated using the union merge strategy as before. There's no other way to
update the index that would be any faster.
It is possible that a union merge and a fast-forward result in different file
contents: Files should have the same lines, but a union merge may change
their order. If this happens, the next commit made to the git-annex branch
will have some unnecessary changes to line orders, but the consistency
of data should be preserved.
Note that when the journal contains changes, a fast-forward is never attempted,
which is fine, because committing those changes would be vanishingly unlikely
to leave the git-annex branch at a commit that already exists in one of
the remotes.
The real difficulty is handling the case where multiple remotes have all
changed. git-annex does find the best (ie, newest) one and fast forwards
to it. If the remotes are diverged, no fast-forward is done at all. It would
be possible to pick one, fast forward to it, and make a merge commit to
the rest, I see no benefit to adding that complexity.
Determining the best of N changed remotes requires N*2+1 calls to git-log, but
these are fast git-log calls, and N is typically small. Also, typically
some or all of the remote refs will be the same, and git-log is not called to
compare those. In the real world I expect this will almost always add only
1 git-log call to the merge process. (Which already makes N anyway.)
Specifically, disabled trying to update the git-annex branch on the remote,
since that data is never used by operations that act on such remotes.
Also, when copying content to such a remote, skip committing the presence
information changes to its git-annex branch. Leaving it in the journal there
is ok: Any command run on the remote that needs the info will flush the
journal.
This may partially solve this bug:
http://git-annex.branchable.com/bugs/fails_to_handle_lot_of_files/
Although I still see unreaped git processes piling up when doing a copy --to.
Another process may stage journalled files before the lock is
taken, so need to get the list of journalled files afterwards.
It's unfortunate this means getting the directory contents twice,
but it seems better to do that than sometimes take the lock
unnecessarily.
