Summary:
This allows automation to know which modified or added files are LFS for
specific changesets.
Test Plan: Added a test
Reviewers: #mercurial, davidsp
Reviewed By: davidsp
Subscribers: mjpieters
Differential Revision: https://phabricator.intern.facebook.com/D5091591
Signature: t1:5091591:1495163606:138638d8ccc57b8ed6c1e324750ec5dc15c198a0
Summary:
When a bundle is created, the bundle could be exchanged via copy and possibly
eventually reach to a publishing repo. If we don't upload LFS blobs, hg server
could have revisions that can never be checked out or verified.
So let's just assume bundles generated by `hg bundle` will be public, and
upload LFS blobs automatically, without depending on other code review tooling.
Note: there is a `preoutgoing` hook which will be triggered in this case,
however it's not useful since it does not have the `outgoing` information.
Test Plan: Will add a new test
Reviewers: davidsp, #mercurial, rmcelroy
Reviewed By: rmcelroy
Subscribers: rmcelroy, mjpieters
Differential Revision: https://phabricator.intern.facebook.com/D5050945
Signature: t1:5050945:1494958648:0c46ab1d85755838ba189bbb1e0673882922bb58
Summary:
The `filectx.cmp` fast path allows us to show "binary file changed" diff output
when two binaries have the same size but different content.
Test Plan: Added a test case
Reviewers: davidsp, #mercurial, rmcelroy
Reviewed By: rmcelroy
Subscribers: mjpieters
Differential Revision: https://phabricator.intern.facebook.com/D5067266
Signature: t1:5067266:1494954783:a41e8213782afe24618ef2930c9576f21610fd3e
Summary:
Calculating what blobs to upload should be fast since we use changelog `files`
information. Users will see `lfs: uploading ...` very soon. So let's downgrade
`lfs: computing set of blobs to upload` to a debug message to make `-v`
cleaner.
Test Plan: Updated existing test.
Reviewers: davidsp, #mercurial, rmcelroy
Reviewed By: rmcelroy
Subscribers: rmcelroy, mjpieters
Differential Revision: https://phabricator.intern.facebook.com/D5066600
Signature: t1:5066600:1494954115:c00925a1930f9b53e914078d32b6c4e4161099ee
Summary:
Displaying total bytes to upload is currently inaccurate - the server could
already have some objects so only a subset of selected objects will be
uploaded.
Besides, we pass `pointers` to related upload and download APIs so `total`
could be calculated from them.
This diff removes the inaccurate "need to upload" message and unnecessary
parameters. An accurate message will be added in a later patch.
Reviewers: #mercurial, davidsp, rmcelroy
Reviewed By: rmcelroy
Subscribers: mjpieters
Differential Revision: https://phabricator.intern.facebook.com/D5057470
Signature: t1:5057470:1494949554:07e637a8b58c894598b3f0c2dd53f80a5ade25a8
Summary:
`filelog.filenode` could be `None` when it's unknown - like working copy. That
breaks `isbinary` test. Let's modify `_islfs` check to return False to disable
lfs code path in that case.
Test Plan: Added a new test.
Reviewers: #mercurial, rmcelroy
Reviewed By: rmcelroy
Subscribers: rmcelroy, mjpieters
Differential Revision: https://phabricator.intern.facebook.com/D5034058
Signature: t1:5034058:1494431790:a02b69addf5d2f9b6b9a27ad71ed4f136b2bfd2b
Summary:
`lfs.blobstore` is a path used to store local blobs. Some of the blobs are not
uploaded yet so they couldn't be easily discarded. Although blobs downloaded
from the server could be removed if necessary.
It does not make much sense to make this a config option - the data should be
stored reliably. It's also dangerous to allow write to arbitrary paths under
`repo.vfs`.
This diff makes the local blob store fixed path at `.hg/store/lfs/objects`.
The choice is similar to `.git/lfs/objects`, but with `store`, shared repo
could share their lfs stores.
Test Plan: Modified existing test.
Reviewers: #mercurial, rmcelroy
Reviewed By: rmcelroy
Subscribers: rmcelroy, mjpieters
Differential Revision: https://phabricator.intern.facebook.com/D5009664
Signature: t1:5009664:1493978440:c3f6351d0ea0cca2ce3caa9f7260c5d65bcc0e5f
Summary:
Per discussion with @davidsp, we want to stick to Git-LFS specification and
avoid non-standard behavior. The chunking behavior will happen at LFS server
transparently.
The direct motivation for this is to make it possible to implement an
efficient `filectx.cmp` that just compares hashes.
Test Plan: Updated existing test
Reviewers: #mercurial, durham
Reviewed By: durham
Subscribers: mjpieters, davidsp
Differential Revision: https://phabricator.intern.facebook.com/D5001827
Signature: t1:5001827:1493914639:c58694873e79a8ca910bb8ee01bf593885896664
Summary:
Usually LFS files are binary files. But there could be exceptions. This diff
adds a new customized field `x-is-binary` to record those exceptions.
The `filectx.isbinary` API is changed to use that metadata as a fast path.
This allows us to provide a transparent user experience (whether a file is
stored in LFS or not does not affect its original `isbinary` property),
while still being able to skip loading the LFS blob if the file is binary.
Test Plan: Added a new test case
Reviewers: #mercurial, rmcelroy
Reviewed By: rmcelroy
Subscribers: rmcelroy, mjpieters
Differential Revision: https://phabricator.intern.facebook.com/D5001117
Signature: t1:5001117:1493892406:2a8ebd42d0ae0dbae39b87b9ea16db22b99f3d12
Summary:
This diff changes lfs `revision(raw=False)` output to include hg filelog
metadata. The LFS blob does not contain filelog metadata as before.
This hurts performance if there is a rename, or the binary starts with the
magic `\1\n`. But compatibility is greatly improved - it's now possible to swap
a non-lfs revision with mercurial rename to a lfs revision, and easier to be
compatible with remotefilelog (namely, remotefilelog defers filelog.add until
commit hash is known).
Test Plan: Modified existing test.
Reviewers: #mercurial, durham
Reviewed By: durham
Subscribers: rmcelroy, durham, mjpieters
Differential Revision: https://phabricator.intern.facebook.com/D4906074
Signature: t1:4906074:1492559513:09b25fc1026d4ce8fd784a044d6724f12e8bda45
Summary:
The "chunking" feature is not specified by [the current Git LFS standard](21e1695220/docs/spec.md).
Therefore avoid using it if possible - if there is only one chunk, use the
standard specification (`https://git-lfs.github.com/spec/v1`).
An upload message is slightly changed to be more accurate.
Test Plan: Changed existing tests
Reviewers: davidsp, #mercurial, rmcelroy
Reviewed By: rmcelroy
Subscribers: rmcelroy, mjpieters
Differential Revision: https://phabricator.intern.facebook.com/D4856527
Signature: t1:4856527:1491638822:1c6f555b706e7bb22dd9090afa156f2161bf9f7f
Summary:
This diff adds bundle support for lfs:
- Let `hg bundle` use changegroup3 instead of changegroup2 to record revlog
flags.
- Hook related functions so `hg -R bundle.hg` works with LFS.
Test Plan: Added a test
Reviewers: #mercurial, rmcelroy
Reviewed By: rmcelroy
Subscribers: rmcelroy, mjpieters
Differential Revision: https://phabricator.intern.facebook.com/D4856335
Signature: t1:4856335:1491638660:d56ba54ea5f59262f009418b9c7e48c8f2a25ed6
Summary:
Per discussion with @davidsp, it's better for LFS to not store Mercurial
filelog metadata, which is currently used to store rename information. That has
many advantages:
- Large blobs could be reused across renames
- No need to special handle files starting with `\1\n`
- P4 LFS server implementation is much easier
- remotefilelog LFS support is easier and cleaner
That said, the rename information is stored as lfs metadata using the
non-standard `x-hg-copy`, `x-hg-copyrev` keys. So they still exist and are
functional.
The disadvantage is that rename gets no longer hashed, which is probably fine.
Test Plan: Added a test
Reviewers: davidsp, #sourcecontrol, rmcelroy
Reviewed By: rmcelroy
Subscribers: jsgf, rmcelroy, stash, mjpieters, davidsp
Differential Revision: https://phabricator.intern.facebook.com/D4849764
Signature: t1:4849764:1491580506:1d80ad476b9cbd6773843cb52aee6745f478a0b0
Summary:
The diff cleans up messages shown to the user. It makes verbose messages
gated by `if ui.verbose`, and simplifies some words.
The resulting user experience is, when there is no large file involved, lfs
shows nothing. When there are largefiles being downloaded or uploaded, show
progress bar if it takes long. The progress bar is the only user visible
output from lfs by default.
Test Plan: `rt test-lfs.t`
Reviewers: #mercurial, simonfar
Reviewed By: simonfar
Subscribers: simonfar, mjpieters
Differential Revision: https://phabricator.intern.facebook.com/D4813297
Signature: t1:4813297:1491213089:c9ec363eb65df4c85282c90b230da3321b93b5e0
Summary:
Previously, the remote blobstore could be either `git-lfs` or `dummy`. The
application code does not really care, it only wants a "remote" blobstore.
This diff adds a factory method and makes `git-lfs` and `dummy` stores private.
The `@staticmethod get(vfs)` interface is also removed as it's duplicated and
unnecessary - as long as mercurial calls `reposetup`, the blob store objects
are set, and they cannot be missing.
The error message about an unsupported store is also changed to be consistent
with mercurial style. A test was added to test the error.
Test Plan: `rt test-lfs.t`
Reviewers: #mercurial, simonfar
Reviewed By: simonfar
Subscribers: simonfar, mjpieters
Differential Revision: https://phabricator.intern.facebook.com/D4813023
Signature: t1:4813023:1491382755:b9d8ec6518141d0ba8263e16c53f430ce80c39f0
Summary:
The bypass option limits lfs's functionality to only skip hash checks. It is
intended to be used server-side, to make it more predictable - the server
never interacts with the lfs blob service.
Test Plan: Added a test case
Reviewers: #mercurial, rmcelroy
Reviewed By: rmcelroy
Subscribers: rmcelroy, simonfar, mjpieters
Differential Revision: https://phabricator.intern.facebook.com/D4811828
Signature: t1:4811828:1491301114:c2e3c4200ce4cc84b9c5872a8b9a040176bb002a
Summary:
This diff makes the `test-lfs.t` much more stronger. It reveals a lot of core
hg issues in this area. I'll send patches to fix them all.
Regarding on lfs, there are some changes:
- An existence check in its push hook was added. Otherwise pushing a revision
with rename will cause crash.
- The "read" processor is responsible for downloading blobs, and translate
raw revision to lfs text. It should always return lfs text. But it may
return raw revision text on error currently. That error handler was
removed to avoid further damage.
Test Plan: Added new test cases. I also added `hg verify` to sanity check things are good.
Reviewers: #mercurial, rmcelroy
Reviewed By: rmcelroy
Subscribers: rmcelroy, mjpieters
Differential Revision: https://phabricator.intern.facebook.com/D4792641
Signature: t1:4792641:1490868946:8f09c84dc9ebda3889d6a1ea04c49a06acbf38a3
Summary:
Previously `lfs.blobstore` must be a relative path. If an absolute
blobstore path is set, it will traceback because vfs audit fails:
```
File "hg/mercurial/revlog.py", line 1356, in _processflags
text, vhash = writetransform(self, text)
File "fb-hgext/hgext3rd/lfs/wrapper.py", line 69, in writetostore
blobstore.local.get(self.opener).write(storeid, chunk)
File "fb-hgext/hgext3rd/lfs/blobstore.py", line 41, in write
fp = self._opener(self.filename(storeid), 'w+', atomictemp=True)
File "hg/mercurial/vfs.py", line 344, in __call__
self.audit(path)
File "hg/mercurial/pathutil.py", line 64, in __call__
raise error.Abort(_("path contains illegal component: %s") % path)
Abort: path contains illegal component: /home/quark/lfslocalblobstore/d7/dbc611df1fe7dfacfe267a2bfd32ba8fc27ad16aa72af7e6c553a120b92f18
```
That was because the code was using `repo.vfs`. This diff adds a new `lfsvfs`
to avoid the issue. The `lfsvfs` also did the correct filename check (the
old `re.match` check will not match the whole string), so `blobstore.local`
could be simplified a lot.
Test Plan:
A new test case was added to make sure absolute blobstore path works. I also
did some cleanups for the test file to de-dup hgrc, and avoid writing files
outside `$TESTTMP`.
Reviewers: #mercurial, rmcelroy
Reviewed By: rmcelroy
Subscribers: rmcelroy, mjpieters, remi
Differential Revision: https://phabricator.intern.facebook.com/D4785084
Signature: t1:4785084:1490693306:b42eef8e53af107897c2e1bc2984c090bdd2e465
Summary:
A bunch of modifications to get the test pass with the new lfs code.
- Move `lfs` to `hgext3rd`. The code was supposed for hg-core. For now, we do
them in fb-hgext to speed up the process
- Remove the windows test, which is not supported by `run-tests.py` and is
duplicated with `test-lfs.t`.
- Do import `mercurial.i18n._` correctly.
- Change some i18n logic a bit so it's more translator-friendly.
- Change `revlog.RevlogError` to `error.RevlogError`.
- Avoid direct symbol import of `mercurial.util.bytecount`, which will fail the
upstream importchecker test.
- Fix various lint issues like lines being too long etc.
- Document lfs config options.
Test Plan: `arc unit`
Reviewers: #sourcecontrol, rmcelroy
Reviewed By: rmcelroy
Subscribers: mjpieters
Differential Revision: https://phabricator.intern.facebook.com/D4772216
Signature: t1:4772216:1490401458:1ad3c18ab80e1d31085d0b6b4c630e62a7dc7930
