Large speed up to importing trees from special remotes that contain a lot
of files, by only processing changed files.
Benchmarks:
Importing from a special remote that has 10000 files, that have all been
imported before, and 1 new file sped up from 26.06 to 2.59 seconds.
An import with no change and 10000 unchanged files sped up from 24.3 to
1.99 seconds.
Going up to 20000 files, an import with no changes sped up from
125.95 to 3.84 seconds.
Sponsored-by: k0ld on Patreon
For simplicity, I've not tried to make it handle History yet, so when
there is a history, a full import will still be done. Probably the right
way to handle history is to first diff from the current tree to the last
imported tree. Then, diff from the current tree to each of the
historical trees, and recurse through the history diffing from child tree
to parent tree.
I don't think that will need a record of the previously imported
historical trees, and so Logs.Import doesn't store them. Although I did
leave room for future expansion in that log just in case.
Next step will be to change importTree to importChanges and modify
recordImportTree et all to handle it, by using adjustTree.
Sponsored-by: Brett Eisenberg on Patreon
This gets the trees built, but it does not use them. Next step will be
to remember the tree for next time an import is done, and diff between
old and new trees to find the files that have changed.
Added --missing to the mktree parameters. That only disables a check, so
it's ok to do everywhere mktree is used. It probably also speeds up
mktree to disable the check.
Note that git fsck does not complain about the resulting tree objects
that point to shas that are not in the repository. Even with --strict.
A quick benchmark, importing 10000 files, this slowed it down
from 2:04.06 to 2:04.28. So it will more than pay for itself.
Sponsored-by: Luke Shumaker on Patreon
Speed up importing trees from special remotes somewhat by avoiding
redundant writes to sqlite database.
Before, import would write to both the git-annex branch and also to the
sqlite database. But then the next time it was run, needsUpdateFromLog
would see the branch had changed, so run updateFromLog, which would make
the same writes to the sqlite database a second time.
Now import writes only to the git-annex branch. The next time it's run,
needsUpdateFromLog sees that the branch has changed and so calls
updateFromLog, which updates the sqlite database.
Why defer the write to the sqlite database like this? It seems that it
could write to the database as it goes, and at the end call
recordAnnexBranchTree to indicate that the information in the git-annex
branch has all been written to the cidsdb. That would avoid the second
import doing extra work.
But, there could be other processes running at the same time, and one of
them may update the git-annex branch, eg merging a remote git-annex branch
into it. Any cids logs on that merged git-annex branch would not be
reflected in the cidsdb yet. If the import then called
recordAnnexBranchTree, the cidsdb would never get updated with that merged
information.
I don't think there's a good way to prevent, or to detect that situation.
So, it can't call recordAnnexBranchTree at the end. So it might as well
wait until the next run and do updateFromLog then. It could instead do
updateFromLog at the end, but it's going to check needsUpdateFromLog
at the beginning anyway.
Note that the database writes were queued, so there is already a cidmap
that is used to remember changes that the current process has made.
So, omitting database writes can't change the behavior of the current
process.
Also note that thirdpartypopulatedimport uses recordcidkeyindb, which
reflects what it already did. That code path does not use the cidmap,
but does not need to query it either. It might be possible to make that
code path also only update the git-annex branch and not the db, but I
haven't checked.
Sponsored-by: Noam Kremen on Patreon
I noticed git-annex was using a lot of CPU when downloading from youtube,
and was not displaying progress. Turns out that yt-dlp (and I think also
youtube-dl) sometimes only knows an estimated size, not the actual size,
and displays the progress output slightly differently for that. That broke
the parser. And, the parser was feeding chunks that failed to parse back
as a remainder, which caused it to try to re-parse the entire output each
time, so it got slower and slower.
Using --progress-template like this should avoid parsing problems as well
as future proof against output changes. But it will work with only yt-dlp.
So, this seemed like the right time to deprecate youtube-dl, and default
to yt-dlp when available.
git-annex will still use youtube-dl if that's all that's available.
However, since the progress parser for youtube-dl was buggy, and I don't
want to maintain two different progress parsers (especially since
youtube-dl is no longer in debian unstable having been replaced by
yt-dlp), made git-annex no longer try to parse youtube-dl's progress.
Also, updated docs for yt-dlp being default. It did not seem worth
renaming annex.youtube-dl-options and annex.youtube-dl-command.
Note that yt-dlp does not seem to document the fields available in the
progress template. I found them by reading the source and looking at
the templates it uses internally. Also note that the use of "i" (rather
than "s") in progressTemplate makes it display floats rounded to integers;
particularly the estimated total size can be a float. That also does not
seem to be documented but I assume is a python thing?
Sponsored-by: Joshua Antonishen on Patreon
Consistency with sync and internal consistency is more important than
consistency with the assistant, which is not itself consistent about
what it does when run in a subdirectory.
Note that with -C, it will still commit staged changes to files outside
the directory. Like sync does. Presumably if the user is manually
staging things, then running this command, they intend to build up a
commit.
Sponsored-by: unqueued on Patreon
The obvious way to fix this would be to adapt lines to split on null.
However, it's actually nontrivial to rewrite lines. In particular it has a
weird implementation to avoid a space leak. See:
https://gitlab.haskell.org/ghc/ghc/-/issues/4334
Also, while that is a small amount of code, it's covered by a rather
complex copyright and I'd have to include that copyright in git-annex.
So, I opted to filter out the trailing empty string instead.
Sponsored-by: Dartmouth College's Datalad project
assist: New command, which is the same as git-annex sync but with
new files added and content transferred by default.
(Also this fixes another reversion in git-annex sync,
--commit --no-commit, and --message were not enabled, oops.)
See added comment for why git-annex assist does commit staged
changes elsewhere in the work tree, but only adds files under
the cwd.
Note that it does not support --no-commit, --no-push, --no-pull
like sync does. My thinking is, why should it? If you want that
level of control, use git commit, git annex push, git annex pull.
Sync only got those options because pull and push were not split
out.
Sponsored-by: k0ld on Patreon
