Added annex.bwlimit and remote.name.annex-bwlimit config that works for git
remotes and many but not all special remotes.
This nearly works, at least for a git remote on the same disk. With it set
to 100kb/1s, the meter displays an actual bandwidth of 128 kb/s, with
occasional spikes to 160 kb/s. So it needs to delay just a bit longer...
I'm unsure why.
However, at the beginning a lot of data flows before it determines the
right bandwidth limit. A granularity of less than 1s would probably improve
that.
And, I don't know yet if it makes sense to have it be 100ks/1s rather than
100kb/s. Is there a situation where the user would want a larger
granularity? Does granulatity need to be configurable at all? I only used that
format for the config really in order to reuse an existing parser.
This can't support for external special remotes, or for ones that
themselves shell out to an external command. (Well, it could, but it
would involve pausing and resuming the child process tree, which seems
very hard to implement and very strange besides.) There could also be some
built-in special remotes that it still doesn't work for, due to them not
having a progress meter whose displays blocks the bandwidth using thread.
But I don't think there are actually any that run a separate thread for
downloads than the thread that displays the progress meter.
Sponsored-by: Graham Spencer on Patreon
This was maybe a real bug too, although I don't know what circumstances
it would be a problem. See comment for analysis of this windows drive
letter wackyness issue.
Sponsored-by: Brock Spratlen on Patreon
This is to track down what file in .git/annex/ is being written to via a
temp file when the repository is read-only.
Sponsored-by: Dartmouth College's Datalad project
Except when configuration makes curl be used. It did not seem worth
trying to tail the file when curl is downloading.
But when an interrupted download is resumed, it does not read the whole
existing file to hash it. Same reason discussed in
commit 7eb3742e4b76d1d7a487c2c53bf25cda4ee5df43; that could take a long
time with no progress being displayed. And also there's an open http
request, which needs to be consumed; taking a long time to hash the file
might cause it to time out.
Also in passing implemented it for git and external special remotes when
downloading from the web. Several others like S3 are within striking
distance now as well.
Sponsored-by: Dartmouth College's DANDI project
IncrementalVerifier moved to Utility.Hash, which will let Utility.Url
use it later.
It's perhaps not really specific to hashing, but making a separate
module just for the data type seemed unncessary.
Sponsored-by: Dartmouth College's DANDI project
This eliminates the distinction between decodeBS and decodeBS', encodeBS
and encodeBS', etc. The old implementation truncated at NUL, and the
primed versions had to do extra work to avoid that problem. The new
implementation does not truncate at NUL, and is also a lot faster.
(Benchmarked at 2x faster for decodeBS and 3x for encodeBS; more for the
primed versions.)
Note that filepath-bytestring 1.4.2.1.8 contains the same optimisation,
and upgrading to it will speed up to/fromRawFilePath.
AFAIK, nothing relied on the old behavior of truncating at NUL. Some
code used the faster versions in places where I was sure there would not
be a NUL. So this change is unlikely to break anything.
Also, moved s2w8 and w82s out of the module, as they do not involve
filesystem encoding really.
Sponsored-by: Shae Erisson on Patreon
Eg, showImprecise 1 1.99 returned "1.1" rather than "2". The 9 rounded
upward to 10, and that was wrongly used as the decimal, rather than
carrying the 1.
Sponsored-by: Jack Hill on Patreon
Clear visible progress bar first.
Removed showSideActionAfter because it can't be used in reconcileStaged
(import loop). Instead, it counts the number of files it
processes and displays it after it's seen a sufficient to know it's
taking a while.
Sponsored-by: Dartmouth College's Datalad project
Avoids users thinking this scan is a big deal, when it's not in the
majority of repos.
showSideActionAfter has some ugly caveats, since it has to display in
the background of another action. I could not see a better way to do it
and it works fine in this particular case. It also doesn't really belong
in Annex.Concurrent, but cannot go in Messages due to an import loop.
Sponsored-by: Dartmouth College's Datalad project
ghc 8.8.4 seems to have changed something that broke code that has been
successfully using forkProcess since 2012. Likely a change to GC internals.
Since forkProcess has never had clear documentation about how to
use it safely, avoid using it at all. Instead, when git-annex needs to
daemonize itself, re-run the git-annex command, in a new process group
and session.
This commit was sponsored by Luke Shumaker on Patreon.
Fix behavior of several commands, including reinject, addurl, and rmurl
when given an absolute path to an unlocked file, or a relative path that
leaves and re-enters the repository.
To avoid slowing down all the cases where the paths are already ok
with an unncessary call to getCurrentDirectory, put in an optimisation
in relPathCwdToFile. That will probably also speed up other parts of
git-annex by some small amount, but I have not benchmarked.
Note that I did not convert branchFileRef, because it seems likely that
it will be used with a file that is not provided by the user, so is already
in a sane format. This is certainly true for the way git-annex uses it,
though maybe arguable to the extent Git.Ref is a reusable library.
This uses a DebugSelector, rather than debug levels, which will allow
for a later option like --debug-from=Process to only
see debuging about running processes.
The module name that contains the thing being debugged is used as the
DebugSelector (in most cases; does not need to be a hard and fast rule).
Debug calls were changed to add that. hslogger did not display
that first parameter to debugM, but the DebugSelector does get
displayed.
Also fastDebug will allow doing debugging in places that are used in
tight loops, with the DebugSelector coming from the Annex Reader
essentially for free. Not done yet.
New error message:
Remote foo not usable by git-annex; setting annex-ignore
http://localhost/foo/config download failed: Configuration of annex.security.allowed-ip-addresses does not allow accessing address ::1
If git config parse fails, or the git config file is not available at the url,
a better error message for that is also shown.
This commit was sponsored by Mark Reidenbach on Patreon.
Checksum as content is received from a remote git-annex repository, rather
than doing it in a second pass.
Not tested at all yet, but I imagine it will work!
Not implemented for any special remotes, and also not implemented for
copies from local remotes. It may be that, for local remotes, it will
suffice to use rsync, rely on its checksumming, and simply return Verified.
(It would still make a checksumming pass when cp is used for COW, I guess.)
As yet unused.
Backend.External could perhaps implement it too, although that would
involve sending chunks of data to it via a pipe or something, so likely
to be slow.
Note this does find things in PATH that are not executable.
Like searchPath use, the executable bit is not checked. Thing is,
there does not seem to be a binding for access(), which would be the
right way to check that the right execute bit is set. Anyway, if it's in
PATH and it's a file, it's probably fine to treat it as something that
was intended to be executable.
This commit was sponsored by Brock Spratlen on Patreon.
Fix an oddity in matching options and preferred content expressions such as
"foo (bar or baz)", which was incorrectly handled as if it were "(foo or
bar) and baz)" rather than the intended "foo and (bar or baz)"
Seemed like a change to consume should be able to handle this case
better, but I was having trouble writing it that way, so instead added
a separate pass that inserts the implicit ands explicitly. Also added
several test cases to make sure versions with and without explicit ands
generate the same.
Seems that dropDrive on windows only drops eg c:/ but not a leading /
while on linux, it does drop a leading / (which is what it considers
to be equivilant to a drive letter. I had been relying on it to drop
both. So need to drop leading directory separators.
Also, if the quickcheck generated input is eg "c:c:c:c:foo",
dropDrive will only drop the first one, leaving a path that's
still not relative. So instead of using dropDrive, just remove the
colons from the path.
Directory special remotes with importtree=yes now avoid unncessary overhead
when inodes of files have changed, as happens whenever a FAT filesystem
gets remounted.
A few unusual edge cases of modifications won't be detected and
imported. I think they're unusual enough not to be a concern. It would
be possible to add a config setting that controls whether to compare
inodes too, but does not seem worth bothering the user about currently.
I chose to continue to use the InodeCache serialization, just with the
inode zeroed. This way, if I later change my mind or make it
configurable, can parse it back to an InodeCache and operate on it. The
overhead of storing a 0 in the content identifier log seems worth it.
There is a one-time cost to this change; all directory special remotes
with importtree=yes will re-hash all files once, and will update the
content identifier logs with zeroed inodes.
This commit was sponsored by Brett Eisenberg on Patreon.
It was just slapping on a path separator to the front of the path to
make it absolute, but on windows, a path like "//foo/bar" actually
has a network "drive" of "//foo" and so that broke the test case.
Since "a:foo" is a somehow relative path on windows
(who knows how), drop any drive from the input. But dropDrive also drops
any leading path separator, making the input path relative. So now
it should be safe to slapp on a leading path separator.
Useful for eg, replicating failures in ghci. No need for this to be a
smart constructor, as long as it's used with valid filepaths, it's ok
and if not the test breaks.
This was not a good test, it broke the requirement that
relPathDirToFileAbs take absolute paths. And it failed when the two
input paths were eg, the same but differently normalized.
Replaced with some tests of the real basics of that function.
And vice-versa, but it's better to use '/' for portability.
Notably, standardPreferredContent contains "archive/*" and that might not
match if the filename ends up coming in with the slashes the other way
around.
Adding new instance for Integer, and some parsers for more parameters.
The conversion of readish to readMaybe is done because a serialized
exit code cannot contain additional text after the number.
New config annex.stalldetection, remote.name.annex-stalldetection, which
can be used to deal with remotes that stall during transfers, or are
sometimes too slow to want to use.
This commit was sponsored by Luke Shumaker on Patreon.
This old code will now be useful for git-annex beyond the assistant.
git-annex won't use the CheckTransferrer part, and won't run transferkeys
as a batch process, and will want withTransferrer to not shut down
transferkeys processes. Still, the rest of this is a good fit for what I
need now.
Also removed some dead code, and simplified a little bit.
This commit was sponsored by Mark Reidenbach on Patreon.
Json objects not yet handled, and some other special cases, but this is
the bulk of the messages.
For progress meters, POSIXTime does not have a Read instance (or a
suitable Show instance), so had to switch to using a Double for progress
meters.
This commit was sponsored by Ethan Aubin on Patreon.
This is an edge case, which happened to be triggered by the P2P protocol
seeing DATA 0. When reading 0 bytes, getting an empty string does
not mean the handle has reached EOF.
I verified there was in fact a bug, where get of an empty file followed
by another file would get the empty file and then fail
with "handle is closed". This fixes it.
This commit was sponsored by Boyd Stephen Smith Jr. on Patreon.
9cb250f7be got the ones in RawFilePath,
but there were others that used the one from unix-compat, which fails at
runtime on windows. To avoid this,
import System.PosixCompat.Files hiding removeLink
This commit was sponsored by Ethan Aubin.
It does use it on both stdout and stderr. It seems unlikely the problem
could really affect stdout, but the unix implementation of it combines
both into a single handle in any case.
It had lost the hAcceptEncoding header that is set as part of the
overriding of http-client's default decompression of compressed files.
Seems likely that would have caused resuming of compressed files to fail
in some cases.
This commit was sponsored by Brett Eisenberg on Patreon.
Fix a bug that could make resuming a download from the web fail when the
entire content of the file is actually already present locally.
What a mess that Request can throw exceptions or not, depending on how
it's configured. Makes it very hard if you need to handle some specific
http status codes in a function like this! Implementing everything two
ways did not seem appealing, if possible at all, so I decided to
override the Request if it did come configured to throw exception on
non-2xx http status. Other exceptions, like from http-client-restricted,
or due to a redirect to a non-http url, still get thrown.
This commit was sponsored by Luke Shumaker on Patreon.
This removeLink was introduced in commit
e505c03bcc, which replaced code
that used removeFile on Windows. So, I know git-annex did not used to do
anything other than removeFile on Windows. If there were symlinks it
wanted to remove, this would not work on windows, but of course it does
not use symlinks on windows.
Unfortunately, there is no hGetNewLineMode. This seems like an oversight
that should be fixed in ghc, but for now, I paper over it with a windows
hack.
The problem with the old version seemed to be that hWaitForInput blocks
rather than timing out when being run concurrently with hGetLine on the
same handle.
This passes the bench test, and also works when run concurrently on
different handles.
This seems to show that hWaitForInput does not seem to behave as
documented. It does not time out, so blocks forever in this situation.
This is with a 0 timeout and with larger timeouts. Unsure why, it looked
like it should work.
All properties changed to use them, except for
prop_encode_c_decode_c_roundtrip, which already filtered to ascii
for other reasons.
A few modules had to be split out, because Setup does not build-depend
on QuickCheck.
instance Arbitrary [Char] allows that, and it's not a legal part of a
filename so can break processing them.
Noticed when prop_view_roundtrips failed.
The instance Arbitrary AssociatedFile avoids this problem.
This commit was sponsored by Mark Reidenbach on Patreon.
Lots of nice wins from this in avoiding unncessary work, and I think
nothing got slower.
This commit was sponsored by Boyd Stephen Smith Jr. on Patreon.
nukeFile replaced with removeWhenExistsWith removeLink, which allows
using RawFilePath. Utility.Directory cannot use RawFilePath since setup
does not depend on posix.
This commit was sponsored by Graham Spencer on Patreon.
Had to split out some modules because getWorkingDirectory needs unix,
which is not a build-dep of configure.
This commit was sponsored by Brock Spratlen on Patreon.
This will break a lot of stuff that uses it, but once fixed should lead
to better performance.
Mostly mechanical.
Changes of note:
* upFrom now uses isPathSeparator, which is better on Windows where
there is not just one
* splitShortExtensions used to take the length of a string,
which would count wide unicode characters as a single character.
Changing to B.length changes that. Note that, git-annex's
annexMaxExtensionLength already changed to the length in bytes
before this change. This function is only used in generating views,
and the small behavior change should not be a problem.
* relHome still uses FilePath because it didn't seem worth changing(?)
This commit was sponsored by Jack Hill on Patreon.
The problem was this line:
cleanup = and <$> sequence (map snd v)
That caused all of v to be held onto until the end, when the cleanup action
was run.
I could not seem to find a bang pattern that avoided the leak, so I
resorted to a IORef, rather clunky, but not a performance problem because
it will only be written once per git ls-files, so typically just 1 time.
This commit was sponsored by Mark Reidenbach on Patreon.
inet_addr was removed, but all this needs is localhost, so hardcoding it
should work fine.
It may be that this windows ifdef is no longer needed. It was added in 2013
with a note that getAddrInfo didn't work on windows, but it seems likely
such a problem would have been fixed since.
Which lets progress be displayed when doing concurrent downloads.
Amoung other things, like --json-progress etc.
The youtube-dl output is no longer displayed, except for any errors.
This commit was sponsored by Denis Dzyubenko on Patreon.
Import small files into git, the same as is done when importing with content.
Which means, for small files, --no-content does download them.
If the largefiles expression needs the file content available
(due to mimetype or mimeencoding being used), the import will fail.
This commit was sponsored by Jake Vosloo on Patreon.
This was the last one marked as a zombie. There might be others I don't
know about, but except for in the hypothetical case of a thread dying
due to an async exception before it can wait on a process it started, I
don't know of any.
It would probably be safe to remove the reapZombies now, but let's wait
and so that in its own commit in case it turns out to cause problems.
This commit was sponsored by Boyd Stephen Smith Jr. on Patreon.
matchNeedsFileContent is not used yet, but shows how to add information
about terminals. That one would be needed for
https://git-annex.branchable.com/todo/sync_fast_import/
Note the tricky bit in Annex.FileMatcher.call where it folds over the
included matcher to propagate the information.
This commit was sponsored by Svenne Krap on Patreon.
getPid returns Nothing if the process has already been stopped, and in that
case, the pid will not be displayed. I think that would only happen if
waitForProcess or similar gets called more than once on the same process
handle though.
getPid on unix has an overhead of only a MVar read. On Windows it needs to
make a syscall, so will be probably more expensive. While the added expense
happens even when debug logging is disabled, it should be small enough
compared with the overhead of starting a process that it's not a problem.
(It does occur to me that a debugM that took an IO String could only run it
when debugging is really enabled, which would improve performance. It does
not seem possible to use the current hslogger interface to do that though;
it does not expose the information that would be needed.)
No behavior changes (hopefully), just adding SeekInput and plumbing it
through to the JSON display code for later use.
Over the course of 2 grueling days.
withFilesNotInGit reimplemented in terms of seekHelper
should be the only possible behavior change. It seems to test as
behaving the same.
Note that seekHelper dummies up the SeekInput in the case where
segmentPaths' gives up on sorting the expanded paths because there are
too many input paths. When SeekInput later gets exposed as a json field,
that will result in it being a little bit wrong in the case where
100 or more paths are passed to a git-annex command. I think this is a
subtle enough problem to not matter. If it does turn out to be a
problem, fixing it would require splitting up the input
parameters into groups of < 100, which would make git ls-files run
perhaps more than is necessary. May want to revisit this, because that
fix seems fairly low-impact.
Potentially fixes https://git-annex.branchable.com/bugs/concurrent_git-annex-copy_to_s3_special_remote_fails/
although I don't know if it does.
My thinking is, ResourceT may allocate a resource and then free it,
and a unforced thunk to that resource could result in reading memory
that has since been overwritten by something else, or in a SEGV,
depending. While that seems kind of like a bug in ResourceT to me, if it
is what's happening, this will avoid it. If it's not, this doesn't
really hurt much since the values are all smallish.
This commit was sponsored by Graham Spencer on Patreon.
That made Utility.FileMode depend on unix, but Utility.Tmp now depends
on it, and is used by Setup, which does not. So it was easiest to remove
this, especially since it's not used.
Fixed several cases where files were created without file mode bits that
the umask would usually set. This included exports to the directory special
remote, torrent files used by the bittorrent special remote, hooks written
by git-annex init, and some log files in .git/annex/
Audited all calls, looking for ones that didn't want the umask bits to be
set. All such turned out to already set the specific restrictive file mode
they wanted.
Also audited for other calls to openTempFile, and all are ok,
except for viaTmp which will need further work.
Remote.Directory fixed to set umask mode when writing to an export,
although it has another one using viaTmp that's not fixed.
Will make exports that are published via a http server running as
another user work, for example.
Remote.BitTorrent fixed to set umask mode when downloading the torrent
file. Normally this does not matter as that file does not hang around
after the download, but if a bittorrent download were started by one user,
got interrupted and then another user ran it, this will let them access
the torrent file created by the first user.
Also tested what happens if the other special remote has importtree=yes
and exporttree=yes, and in that case, download via httpalso works too,
without needing to implement any importtree methods here.
It might be possible to make it automatically set exporttree=yes if the
--sameas does. Didn't try, will probably be layering issues.
Or perhaps it should be inherited by sameas like some
other configs? But then, wouldn't it also make sense to inherit
importree=yes? But as shown here, it's not needed by this kind of
remote.
They normally shutdown when the GNUPGHOME directory is deleted, but on
NFS they keep the directory from being deleted. And also, this avoids
a number of them piling up while the test suite is running.
Fixes reversion in 8.20200617 that made annex.pidlock being enabled result
in some commands stalling, particularly those needing to autoinit.
Renamed runsGitAnnexChildProcess to make clearer where it should be
used.
Arguably, it would be better to have a way to make any process git-annex
runs have the env var set. But then it would need to take the pid lock
when running any and all processes, and that would be a problem when
git-annex runs two processes concurrently. So, I'm left doing it ad-hoc
in places where git-annex really does run a child process, directly
or indirectly via a particular git command.
This probably avoids the situation that caused the exception to be
thrown. It also makes sure that both threads end up canceled in the end,
while before the exception from wait outt could have caused errt to
never be waited on.
Since there's a race here, and since Kyle saw an exception leak out,
which I have not been able to reproduce that. See my comment for what
I think might be going on.
Note that, I used tryNonAsync, because it seems a later tryNonAsync
caught the exception. I don't actually understand how it did, as I
understand exception classification, it's the data type, not the way it
was thrown. One possibility is that the async exception may have been wrapped
in some other, non-async exception, and Show displayed it the same way.
Reduce the number of directories listed in libdirs, which makes the linker
check a lot less dead ends looking for directories.
Eliminated some directories that didn't really contain shared libraries,
or only contained the linker.
That left only 2, one in lib and one in usr/lib, so consolidate those two.
Doing it this way, rather than just consolidating all libs that might exist
into a single directory means that, if there are optimised versions of some
libs, eg in lib/subarch/foo.so, and lib/subarch2/foo.so, they don't get
moved around in a way that would make the linker pick the wrong one.
This is only implemented for git-annex get so far. It makes git-annex
get nearly twice as fast in a repo with 10k files, all of them present!
But, see the TODO for some caveats.
Turns out the %(rest) trick was not needed. Instead, just maintain a
list of files we've asked for, and each cat-file response is for the
next file in the list.
This actually benchmarks 25% faster than before! Very surprising, but it
must be due to needing to shove less data through the pipe, and parse
less.
Otherwise use the vendored copy as before.
The library is in Debian testing but not stable. Once it reaches
stable, the vendored copy can be removed.
Did not add it to debian/control because IIRC that's used to build
git-annex on stable too, possibly. However, the Debian maintainer will
probably want to make the package depend on libghc-http-client-restricted-dev
This commit was sponsored by Ilya Shlyakhter on Patreon.
Otherwise use the vendored copy as before.
The library is in Debian testing but not stable. Once it reaches
stable, the vendored copy can be removed.
Did not add it to debian/control because IIRC that's used to build
git-annex on stable too, possibly. However, the Debian maintainer will
probably want to make the package depend on libghc-git-lfs-dev.
This commit was sponsored by Ilya Shlyakhter on Patreon.
Clean build under ghc 8.8.3, which seems to do better at finding cases
where two imports both provide the same symbol, and warns about one of
them.
This commit was sponsored by Ilya Shlyakhter on Patreon.
Network.HTTP.Client exports makeConnection since 0.5.3.
Debian stable has a newer version than 0.5.3, so bumping the
min version seems better than adding an ifdef.
That made eg git-annex get of an unlocked file hang until the
annex.pidlocktimeout and then fail.
This fix should be fully thread safe no matter what else git-annex is
doing.
Only using runsGitAnnexChildProcess in the one place it's known to be a
problem. Could audit for all places where git-annex runs itself as a child
and add it to all of them, later.
Some recent changes to use mask missed that async exceptions can still
be thrown inside it. The goal is to make sure a block of cleanup code
runs entirely, w/o being interrupted by an async exception, so use
uninterruptibleMask.
Also, converted a few to bracket, which is nicer.
Tested the forcerestart code path and it works.
The hairy part is, what if an async exception is caught when it's in
restart?
If it's in the part that stops the old process, the old process
is left in the handle. The next attempt to use the CoProcessHandle
will then throw an IO exception, which will result in restart getting
run again. So I think this will work, but have not actually tested it.
The use of withMVarMasked lets it start the new process and fill the
mvar with it, even if there's an async exception at that point.
Note that exceptions are masked while running forcerestart, so
do not need to worry about an async exception being thrown while it's
recovering from an async exception.
Audited for openFile and openFd, and this fixes all the ones I found
where an async exception could prevent the file getting closed.
Except for the lock pool, which is a whole other can of worms.
Remove old code that can be trivially implemented using async in a much
nicer way (that is async exception safe).
I've audited all forkOS calls (except for ones in the assistant),
and this was the last remaining one that is not async exception safe.
The rest look ok to me.
This handles all createProcessSuccess callers, and aside from process
pools, the complete conversion of all process running to async exception
safety should be complete now.
Also, was able to remove from Utility.Process the old API that I now
know was not a good idea. And proof it was bad: The code size went *down*,
despite there being a fair bit of boilerplate for some future API to
reduce.
