Pierre-Yves David <pierre-yves.david@octobus.net> [Thu, 05 Aug 2021 12:53:36 +0200] rev 47808
subrepo: compare normalised vfs path
Otherwise the realpath call can turn `/` into `\` on windows confusing the
check.
(We probably needs this in more location)
Differential Revision: https://phab.mercurial-scm.org/D11259
Raphaël Gomès <rgomes@octobus.net> [Thu, 05 Aug 2021 18:25:35 +0200] rev 47807
pager: account for flakiness in Windows output
This test case is cursed and probably not worth losing more time over. This
makes apparent what the intended behavior is while still removing the flakiness
from the CI.
Differential Revision: https://phab.mercurial-scm.org/D11257
Raphaël Gomès <rgomes@octobus.net> [Fri, 23 Jul 2021 10:45:08 +0200] rev 47806
windows-ci: clean up the Heptapod CI file now that the baseline is solid
Enough work has been done one the CI side, this now works with little effort
on our side. The next patch will remove the manual switch.
Differential Revision: https://phab.mercurial-scm.org/D11254
Pierre-Yves David <pierre-yves.david@octobus.net> [Tue, 03 Aug 2021 21:22:02 +0200] rev 47805
test-nointerrupt: make "sure" the handler "might" trigger (
issue6558)
We are sure that the signal got sent in the right time frame, however, we still
have race, so either the code is actually buggy or we need some security to make
sure the signal get processed.
We might be affected by https://bugs.python.org/
issue43406 ?
Differential Revision: https://phab.mercurial-scm.org/D11251
Pierre-Yves David <pierre-yves.david@octobus.net> [Tue, 03 Aug 2021 19:26:26 +0200] rev 47804
testing: make sure write_file is "atomic"
This make sure viewer cannot see the new file with partial content.
This was likely the cause of some flakiness in `test-nointerrupt.t`
Differential Revision: https://phab.mercurial-scm.org/D11250
Pierre-Yves David <pierre-yves.david@octobus.net> [Wed, 04 Aug 2021 19:45:13 +0200] rev 47803
test: disable test-subrepo-git.t in python2 + chg
I am a couple of days in try to debug that at it seems minor enough with enough
other priority to simply disable it for now.
Differential Revision: https://phab.mercurial-scm.org/D11253
Georges Racinet <georges.racinet@octobus.net> [Tue, 20 Jul 2021 17:20:19 +0200] rev 47802
hgwebdir: avoid systematic full garbage collection
Forcing a systematic full garbage collection upon each request
can serioulsy harm performance. This is reported as
https://bz.mercurial-scm.org/show_bug.cgi?id=6075
With this change we're performing the full collection according
to a new setting, `experimental.web.full-garbage-collection-rate`.
The default value is 1, which doesn't change the behavior and will
allow us to test on real use cases. If the value is 0, no full garbage
collection occurs.
Regardless of the value of the setting, a partial garbage collection
still occurs upon each request (not attempting to collect objects from
the oldest generation). This should be enough to take care of
reference cycles that have been created by the last request
(assessment of this requires changing the setting, not to be 1).
In my experience chasing memory leaks in Mercurial servers,
the full collection never reclaimed any memory, but this is with
Python 3 and biased towards small repositories.
On the other hand, as explained in the Python developer docs [1],
frequent full collections are very harmful in terms of performance if
lots of objects survive the collection, and hence stay in the
oldest generation. Note that `gc.collect()` is indeed trying to
collect the oldest generation [2]. This happens usually in two cases:
- unwanted lingering objects (i.e., an actual memory leak that
the GC cannot do anything about). Sadly, we have lots of those
these days.
- desireable long-term objects, typically in caches (not inner caches
carried by repositories, which should be collected with them). This
is a subject of interest for the Heptapod project.
In short, the flat rate that this change still permits is
probably a bad idea in most cases, and the default value can
be tweaked later on (or even be set to 0) according to experiments
in the wild.
The test is inspired from test-hgwebdir-paths.py
[1] https://devguide.python.org/garbage_collector/#collecting-the-oldest-generation
[2] https://docs.python.org/3/library/gc.html#gc.collect
Differential Revision: https://phab.mercurial-scm.org/D11204
Pierre-Yves David <pierre-yves.david@octobus.net> [Tue, 03 Aug 2021 18:29:31 +0200] rev 47801
check-module-imports: ignore non-stdlib module installed by distribution
Previously, the check script would detect breezy as part of the stdlib if
installed using the debian package manager.
This silence the following complains:
tests/test-convert-bzr.t:117: imports not lexically sorted: breezy.bzr.bzrdir < sys
tests/test-convert-bzr.t:117: stdlib import "breezy.bzr.bzrdir" follows local import: breezy
tests/test-convert-bzr-ghosts.t:7: imports not lexically sorted: breezy.bzr.bzrdir < sys
tests/test-convert-bzr-ghosts.t:7: stdlib import "breezy.bzr.bzrdir" follows local import: breezy
tests/test-convert-bzr-treeroot.t:7: imports not lexically sorted: breezy.bzr.bzrdir < sys
tests/test-convert-bzr-treeroot.t:7: stdlib import "breezy.bzr.bzrdir" follows local import: breezy
Differential Revision: https://phab.mercurial-scm.org/D11249
Valentin Gatien-Baron <valentin.gatienbaron@gmail.com> [Mon, 02 Aug 2021 08:06:27 -0400] rev 47800
remotefilelog: fix what looks like a wrong refactoring
when various store functions started returning a revlog type as the
first element of the tuple.
Differential Revision: https://phab.mercurial-scm.org/D11243
Georges Racinet <georges.racinet@octobus.net> [Sun, 01 Aug 2021 14:39:38 +0200] rev 47799
rust-nodemap: falling back to C impl as mitigation
This is a mitigation for https://bz.mercurial-scm.org/show_bug.cgi?id=6554
We see sometimes almost all data except the most recent revisions
removed from the persistent nodemap, but we don't know how to
reproduce yet.
This has sadly repercussions beyond just needing to reconstruct the
persistent nodemap: for instance, this automatically filters out
all bookmarks pointing to revisions that the nodemap cannot resolve.
If such filtering happens in a transaction, the update of the
bookmarks file that happens at the end of transaction loses all
bookmarks that have been affected. There may be similar consequences
for other data.
So this is a data loss, something that we have to prevent as soon as
possible.
As a mitigation measure, we will now fallback to the C implementation
in case nodemap lookups failed. This will add some latency, e.g., in
discovery, yet less than disabling the persistent nodemap entirely.
We considered implementing the fallback directly on the Python
side, but `revlog.get_rev()` is not systematically used, there are
also several direct calls to the index method (`self.index.rev()` for
a `revlog` instance). It is therefore more direct to implement the
mitigation in the rust-cpython wrapper.
Differential Revision: https://phab.mercurial-scm.org/D11238