Pierre-Yves David <pierre-yves.david@octobus.net> [Mon, 09 Dec 2019 10:09:40 +0100] rev 43851
remotefilelog: remove most sleep in tests
Now that we removed most of the races, we can remove the infamous sleeps.
Differential Revision: https://phab.mercurial-scm.org/D7587
Pierre-Yves David <pierre-yves.david@octobus.net> [Mon, 09 Dec 2019 09:57:34 +0100] rev 43850
remotefilelog: have command wait for background task in the tests
We leverage the feature introduced in the two previous commit to have the main
command wait for the background task to be finished before returning.
All test still pass and instability seems to have gone away.
Most of the sleep and explicit wait are now useless and will be cleaned up in
later changesets.
Differential Revision: https://phab.mercurial-scm.org/D7586
Pierre-Yves David <pierre-yves.david@octobus.net> [Mon, 09 Dec 2019 09:54:27 +0100] rev 43849
remotefilelog: add a developer option to wait for background processes
In order to block the main command on the subprocess exiting, we ensure the
repo's ui object will call the subprocess.wait() method to ensure the top-level
hg process doesn't exit until all background processes have also done so.
Currently, in the tests, most operation spawning background process as followed
by commands waiting for these operations to complete. However this waiting is
racy. First because it seems like we can start waiting before the background
operation actually start, in which case it is prematurely detected as "done".
Second, because some commands may spawn multiple background operation for the
same operation (eg: rebase can apparently trigger multiple prefetch). The
current approach could be updated to maybe handle the first issue, but the
second one will never be properly handled.
In most case, we do not care that the bg process keep running after the command
end. (Since we explicitly wait for them to end before doing anything else). So
we add an option to wait on the background process before exiting the command.
We'll put it in use in the next changeset.
Differential Revision: https://phab.mercurial-scm.org/D7585
Pierre-Yves David <pierre-yves.david@octobus.net> [Mon, 09 Dec 2019 09:53:43 +0100] rev 43848
procutil: add a option to not fully detach background process
When writing test, it is useful to make code pretending to spawn a fully
detached process while still actually waiting for it to finish. We add the
option to `procutil.runbgcommand`, to do so. We will use this in remote-filelog
tests.
Differential Revision: https://phab.mercurial-scm.org/D7584
Georges Racinet <georges.racinet@octobus.net> [Thu, 05 Dec 2019 20:41:23 +0100] rev 43847
cext-revlog: fixed __delitem__ for uninitialized nodetree
This is a bug in a code path that's seldom used, because in practice
(at least in the whole test suite), calls to `del index[i:j]` currently
just don't happen before the nodetree has been initialized.
However, in our current work to replace the nodetree by a Rust implementation,
this is of course systematic.
In `index_slice_del()`, if the slice start is smaller than `self->length`,
the whole of `self->added` has to be cleared.
Before this change, the clearing was done only by the call to
`index_invalidate_added(self, 0)`, that happens only for initialized
nodetrees. Hence the removal was effective only from `start` to `self->length`.
The consequence is index corruption, with bogus results in subsequent calls,
and in particular errors such as `ValueError("parent out of range")`, due to
the fact that parents of entries in `self->added` are now just invalid.
This is detected by the rebase tests, under conditions that the nodetree
of revlog.c is never initialized. The provided specific test is more direct.
Differential Revision: https://phab.mercurial-scm.org/D7603
Matt Harbison <matt_harbison@yahoo.com> [Wed, 20 Nov 2019 13:03:22 -0500] rev 43846
filemerge: fix a missing attribute usage
Flagged by both pytype and VSCode.
Differential Revision: https://phab.mercurial-scm.org/D7465
Matt Harbison <matt_harbison@yahoo.com> [Wed, 20 Nov 2019 13:01:56 -0500] rev 43845
filemerge: drop a default argument to appease pytype
The function slices and takes the length of this argument without internally
setting it if not provided. There was no bug here because both callers passed
the argument.
Differential Revision: https://phab.mercurial-scm.org/D7464
Augie Fackler <augie@google.com> [Tue, 10 Dec 2019 19:04:53 -0500] rev 43844
fuzz: add a seed corpus for the dirs fuzzer
I was hoping to trigger an asan violation under Python 3 that some internal
tests at Google found, but for some reason that's beyond me I can't seem to
manage.
Differential Revision: https://phab.mercurial-scm.org/D7600