Summary: The test uses the tracing framework to examine paths being prefetched.
Reviewed By: sfilipco
Differential Revision: D18405644
fbshipit-source-id: 6f3e72c1c084e0bb4950b050e3217bf811e15dc1
Summary:
For glob patterns, tree matcher has a better 'visitdir' implementation that
can greatly speed up some pattern matching. So try to use it instead.
This should make things like `hg status -n -c "glob:path/to/something**"` much
faster.
With this change:
% ~/hg/hg files 'glob:skycastle/src/test/resources**' --time --pager=off | wc -l :(
time: real 0.110 secs (user 0.070+0.000 sys 0.020+0.000)
118
Without this change:
% hg files 'glob:skycastle/src/test/resources**' --time --pager=off | wc -l
time: real 16.700 secs (user 15.660+0.000 sys 1.310+0.000)
118
A config was added so we can turn this off if it causes issues.
The fsmonitor code was slightly changed since treematcher.visitdir does not
like trailing slash.
Reviewed By: matthewdippel
Differential Revision: D18305373
fbshipit-source-id: 2ff927041ba8bab6912f125a7ae465be1a0ea659
Summary:
Two small issues:
Use `matchfn` instead of `__call__`, since `basematcher` has:
def __call__(self, fn):
return self.matchfn(fn)
def matchfn(self, f):
return False
Avoid using forceincludematcher when unnecessary.
Reviewed By: sfilipco
Differential Revision: D18414143
fbshipit-source-id: f7a81f88ac922f75b5a145e8094fab4dc8bd1681
Summary:
The `repo.sparsematch(*revs)` API is expected to return a union matcher that
unions sparse configurations from multiple revs.
I expect `repo.sparsematch(a, b, c)` to match more files than
`repo.sparsematch(a, b)`. However, that's not true if all sparse profiles in
`repo.sparsematch(a, b)` are empty (which, turns it into an always matcher),
and `repo.sparsematch(c)` is not empty.
Fix it by using the alwaysmatcher if any of the revs passed to `sparsematch`
has an empty profiler, instead of requiring all revs to all being empty
profiles.
Reviewed By: sfilipco
Differential Revision: D18414145
fbshipit-source-id: 357e90961acb9b2ae59fbdf84509f8201f629d44
Summary: This is used in an upcoming diff.
Reviewed By: sfilipco
Differential Revision: D18414144
fbshipit-source-id: 92a68b36f24b01d87c55d4a5f52de4efcc50221e
Summary:
Right now, there is no read-only API to get directory information.
In an upcoming change I'd like to get a mapping between tree ids and paths.
So add an `tree.dirs()` API to do that.
Reviewed By: sfilipco
Differential Revision: D18414146
fbshipit-source-id: bf47071b052e512e65fdaafe71742274bed4131f
Summary: This gives us a way to check what exactly are being fetched.
Reviewed By: sfilipco
Differential Revision: D18405645
fbshipit-source-id: 9f1c928eaee521562d7f40dae56d477ad501aae1
Summary:
Previously I thought it's possible to use `[pattern, "!%s/*" % pattern]` to
emulate non-recursive rules. However, that's not true.
Consider this rule `**/a`, if we translate it to `["**/a", "!**/a/*"]`, then
consider this path `a/b/a`:
- Its parent directory `a/b` matches the negative pattern `**/a/*` so the
directory wouldn't be visited.
- If `a/b` is not visited, then `a/b/a` might be missed.
- But `a/b/a` does match the pattern `**/a`.
I couldn't find a way to express the non-recursive `**a` pattern using
recursive patterns. Therefore, change TreeMatcher APIs to do non-recursive by
default. The callsite can still append `/**` to all rules to get the old
behavior.
Reviewed By: sfilipco
Differential Revision: D18342228
fbshipit-source-id: a97144c2a73c6d35fbfbd0d55ea3661ad4a9ac12
Summary: Add a Python binding so it can be used in the Python world.
Reviewed By: DurhamG
Differential Revision: D10861609
fbshipit-source-id: c70ed514229fbcdf99556dd26760f76e027d8aa2
Summary:
The `snapshot` extension expects to be able to call `extra()` on the smartlog
context, so make the method available.
Reviewed By: quark-zju
Differential Revision: D18456550
fbshipit-source-id: 9d1620e1b5fbb0866e728c7b378d37b4de9dbe21
Summary:
Apply some automatic formatting fixes to a couple files that disagreed from
the automatic clang-format and black style.
Reviewed By: sfilipco
Differential Revision: D18446028
fbshipit-source-id: 48e4a82b5b8975b930f08ae56fa449e80b0439c8
Summary:
`subset.filter` has the time complexity `O(len(subset))`, which can be huge.
Since we already know the one single revision, use `baseset` to construct the
resulting revset directly. This makes a huge difference in performance.
Reviewed By: sfilipco
Differential Revision: D18441368
fbshipit-source-id: 5094803e8968a5ddd4aaf4909a72f239325b5fff
Summary:
If there is no `.arcconfig`, the Phabricator client will fail to be created.
Let's just wrap it in the `try` block to ignore the issue.
Reviewed By: sfilipco
Differential Revision: D18440072
fbshipit-source-id: b1141df2513991c8cd23817ae5257aa740ce0c3a
Summary:
blackbox sync shows up as taking 100ms multiple times in tracing profiling
result on Windows. In particular, `atomicwrite` takes about 20ms. `sync()`
calls `atomicwrite` multiple times. They add up to 100ms.
For now, improve it by increasing the buffer time so `sync()` gets called less
frequently. If it continues to be a problem, we might move the blackbox logic
to a separate thread.
See also https://gregoryszorc.com/blog/2015/10/22/append-i/o-performance-on-windows/
where Greg moved `close` to background threads on Windows to improve performance.
Reviewed By: sfilipco
Differential Revision: D18452761
fbshipit-source-id: 53901611fd9538e3c5e897431fdfcab3d7e6f909
Summary: Command will be used only in the tests for now.
Reviewed By: markbt
Differential Revision: D18303171
fbshipit-source-id: 4938cca6b0ac0fa1868ab75a64db6d23c201a4f8
Summary:
Previously we returned early on no content changes but in case of metadata-only
changes (like file mode or copy info) we should output some header indicating those.
Reviewed By: markbt
Differential Revision: D18303168
fbshipit-source-id: b715a2254799f76a8e9ca87196865269c53e4830
Summary:
Both simpkins and I encountered cases where `restack` selects too many
commits.
Reviewed By: kulshrax
Differential Revision: D18344037
fbshipit-source-id: 4e0883ee149bcf62ab368ee129e49f3e2eb9c971
Summary:
The test shows that restack rebases some ancient commits to old stacks where
the user probably doesn't care about.
Reviewed By: kulshrax
Differential Revision: D18344038
fbshipit-source-id: 65fa1b809b8515a9546a03eb30b2a6304538cc83
Summary: Previously, whenever EdenFS needed to fetch a single tree node, the import helper would run a prefetch for the entire tree, resulting in an excessively large fetch. Let's limit the fetch to depth 1 to just get the desired node.
Reviewed By: singhsrb
Differential Revision: D18263067
fbshipit-source-id: edd6ac56940a2ade64d95e78154db5c46eedccea
Summary: The top-level prefetchtrees function didn't support a depth parameter, while the underlying implementation does. Let's add an argument to pass on so that callers of the top-level function can configure the depth. The particular use case I have in mind is EdenFS's import helper, which needs to be able to limit the fetch depth to 1.
Reviewed By: quark-zju
Differential Revision: D18263661
fbshipit-source-id: 59991da2c30a5107df86c5a4b9614f113327f468
Summary: There are several places where we need to perform an HTTP tree fetch, and fall back to SSH if it fails. So far, every callsite has had to wrap the fetch function with a try block and manually print an error message and log the exception. Let's put all this functionality into a single wrapper function that catches all exceptions and returns a boolean indicating success or failure.
Reviewed By: quark-zju
Differential Revision: D18262915
fbshipit-source-id: b5c3395509eb90f70311f439129a6d60089bfe6b
Summary:
Throw errors if a read or write is called on the `Overlay` class after `close()` has been called. Adds IO reference counting to know when it is safe to release the overlay lock after calling `close()`
This takes care to catch errors that could be thrown around overlay IO and makes sure to correctly decrement the IO count in the case of errors.
Reviewed By: chadaustin
Differential Revision: D17957360
fbshipit-source-id: af5e3c3f98c77800914d03cd748fb2f1c59ae264
Summary:
Smartlog depends on `pytz` to render times in timezones other than the default
timezone. Add this as a dependency to `libhg`.
Reviewed By: xavierd, farnz
Differential Revision: D18425630
fbshipit-source-id: 16bc147cbea6cdf34dda73eee17946186c9277b6
Summary:
This makes `hg dbsh` work on Windows because IPython can now detect missing
`pyreadline` correctly.
Reviewed By: DurhamG
Differential Revision: D18388302
fbshipit-source-id: 07dcdd8886316f577249ea42b8229de30ae2c96f
Summary:
The push operation has 2 phases: push, then move bookmarks, etc. Make that in
one transaction so it wouldn't end up with an inconsistent state.
Reviewed By: sfilipco
Differential Revision: D18362363
fbshipit-source-id: 338ef1b088975a9d1b043ccef81782e14c77d8e1
Summary: This makes `hg continue` the way to exit the "unfinished merge" state.
Reviewed By: sfilipco
Differential Revision: D18110070
fbshipit-source-id: cf22242a8dc47603d9cc3a92cf544959330636c7
Summary: The current behavior is actually okay. So let's add a test to protect it.
Reviewed By: sfilipco
Differential Revision: D18329288
fbshipit-source-id: 117b02447f371f1c369f897b136c1a97607bb93e
Summary:
We want to parse revsets even when dirstate is broken.
We need this code to work in as many scenarios as possible os that we can start
to fix broken repositories.
Reviewed By: quark-zju
Differential Revision: D18385918
fbshipit-source-id: d161b8e3041f503a4662eddeb1a11e76f730e66d
Summary:
This avoids some buggy cases. For example, the client repetitively marks X as
"landed as X".
Reported By: simpkins
Reviewed By: simpkins
Differential Revision: D18389747
fbshipit-source-id: 2a66730b5df1cb7971bce6c7ace04c55426f77c1
Summary:
Not getting the landed commits is not a fatal error. So let's just ignore them
with some warning messages.
Reviewed By: DurhamG
Differential Revision: D18384062
fbshipit-source-id: bef01fdd3303ab66170175f5f5ed690182464df2
Summary:
The "commit identifier" returned via Phabricator GraphQL can include subversion
numbers that cannot be converted to commit hashes. Ignore them.
Reviewed By: singhsrb
Differential Revision: D18376641
fbshipit-source-id: e62ba41a486fc91910928e4f5d66c5430bf8c142
Summary:
To avoid scenarios where hard reboots prevent the SNAPSHOT page from
getting flushed, atomically write the file upon clone.
Reviewed By: simpkins
Differential Revision: D18382583
fbshipit-source-id: 1c4de46165587635f71744daf4d2a8c71c58cc5f
Summary:
If a checkout's SNAPSHOT file disappeared, the mount error given would
be that a Hash had an invalid size. Instead, throw a file not found
error on mount.
Reviewed By: genevievehelsel
Differential Revision: D18381074
fbshipit-source-id: 35282e0990189d4084a2a64330a5733561cf88c2
Summary:
Looking at a log, it wasn't immediately obvious what might have passed
an invalid hash into the Hash constructor. Improve the error message
to make the cause clearer.
Reviewed By: genevievehelsel
Differential Revision: D18380916
fbshipit-source-id: 620b8fa902a87496b87a5aa0ff304e6991585864
Summary:
We now list some directories with `hg rage` on Windows, but a few fail with:
```
Abort: dir /o-s C:\open\myrepo\.hg/store/packs\ error: Parameter format not correct - "store\packs\".
```
This is due to the forward-slash after .hg, so replace it with a backslash on Windows.
Reviewed By: quark-zju
Differential Revision: D18378509
fbshipit-source-id: 3ff1a8ba9f400a7330fece4dd83e29efa7ffebc1
Summary: The uptime on Windows originally ran `time /T` to get the start time for comparison. By itself this would be insufficient because it doesn't include the date, and there is actually a portable `time.ctime()` step above in the log so this extra command is unnecessary.
Reviewed By: quark-zju
Differential Revision: D18376448
fbshipit-source-id: 06d5b4a87baec2352b6a07ac264161c6e43b5965
