Previously, when a chg server is exiting, it does not handle connected
clients so clients may get ECONNRESET and crash:
1. client connect() # success
2. server shouldexit = True and exit
3. client recv() # ECONNRESET
297d89f2789e makes this race condition easier to reproduce if a lot of short
chg commands are started in parallel.
This patch fixes the above issue by unlinking the socket path to stop
queuing new connections and processing all pending connections before exit.
This patch changes unixforkingservice so it only calls
`self._servicehandler.unlinksocket(self.address)` at most once.
This is needed by the next patch.
Previously, checkportisavailable only checks ports on the IPv4 address. This
patch makes it check IPv6 as well. It'll be useful if "localhost" does not
have an IPv4 address, or its IPv4 address does not exist somehow.
zeroconf only knows how to deal with IPv4; I develop on a system where the only
IPv4 address is 127.0.0.1.
Teach zeroconf to ignore IPv6 addresses when looking for plausible IPv4
connectivity.
According to the specification [1], $XDG_RUNTIME_DIR should be ignored
unless:
The directory MUST be owned by the user, and he MUST be the only one
having read and write access to it. Its Unix access mode MUST be 0700.
This patch adds a check and ignores it if it does not meet part of the
criteria.
[1]: https://standards.freedesktop.org/basedir-spec/basedir-spec-latest.html
Previously, the code was similar to what revlog._chunks() was doing,
which took a raw data segment and delta chain, obtained buffers for
the raw revlog chunks within, and decompressed them.
This commit splits the "get raw chunks" action from "decompress." The
goal of this change is to more accurately measurely decompression
performance.
On a ~50k deltachain for a manifest in mozilla-central:
! full
! wall 0.430548 comb 0.440000 user 0.410000 sys 0.030000 (best of 24)
! deltachain
! wall 0.016053 comb 0.010000 user 0.010000 sys 0.000000 (best of 181)
! read
! wall 0.008078 comb 0.010000 user 0.000000 sys 0.010000 (best of 362)
! rawchunks
! wall 0.033785 comb 0.040000 user 0.040000 sys 0.000000 (best of 100)
! decompress
! wall 0.327126 comb 0.320000 user 0.320000 sys 0.000000 (best of 31)
! patch
! wall 0.032391 comb 0.030000 user 0.030000 sys 0.000000 (best of 100)
! hash
! wall 0.012587 comb 0.010000 user 0.010000 sys 0.000000 (best of 233)
These classes are pretty large and independent from revset computation.
2961 mercurial/revset.py
973 mercurial/smartset.py
3934 total
revset.prettyformatset() is renamed to smartset.prettyformat(). Smartset
classes are aliased since they are quite common in revset.py.
pager replaced stdout with a line buffered version to work around glibc
deciding on a buffering strategy on the first write to stdout. This is going
to make my next patch hard, as replacing stdout will make tracking time
spent blocked on it more challenging.
Move the line buffering requirement to util.py, and remove it from pager.
This means that the abuse of ui.formatted=True and pager set to cat or equivalent
no longer results in a line-buffered output to a pipe, hence (BC), although
I don't expect anyone to be affected
changelog.heads() first calls headrevs then converts them to nodes.
localrepo.heads() then sorts them using self.changelog.rev function and makes
useless conversion back to revs. Instead let's call changelog.headrevs() from
localrepo.heads(), sort the output and then convert to nodes. Because headrevs
does not support start parameter this optimization only works if start is None.
This is a small style update for clarity. The previous situation was:
if foo:
50 lines
else:
2 lines
In such case I tend to invert these to get the simpler branch out of the way
earlier:
if not foo:
2 lines
else:
50 lines
This makes the conditional and various alternatives fit on the same screen,
simpler to read overall.
Bundle2 has its own mechanisms to check for heads (and other) changes, so push
using bundle2 is relying on the "check:heads" bundle part of unbundle and the
'check_heads' call is not checking anything. We add a small comment to make
this clearer.
The default SIGPIPE handler causes Mercurial to exit immediately, without
running any Python cleanup code (except and finally blocks, atexit handlers
etc). This creates problems if you want to do something at exit.
If we need a different exit code for broken pipe from pager, then we should
code that ourselves in Python; this appears to have been cargo-culted from
the fork implementation of pager that's no longer used, where it was needed
to stop Broken Pipe errors appearing on the user's terminal.
The verifier calls out to _validpath() to check if it should verify
that path and the narrowhg extension overrides _validpath() to tell
the verifier to skip that path. In treemanifest repos, the verifier
calls the same method to check if it should visit a
directory. However, the decision to visit a directory is different
from the condition that it's a matching path, and narrowhg was working
around it by returning True from its _validpath() override if *either*
was true.
Similar to how one can do "hg files -I foo/bar/ -X foo/" (making the
include pointless), narrowhg can be configured to track the same
paths. In that case match("foo/bar/baz") would be false, but
match.visitdir("foo/bar/baz") turns out to be true, causing verify to
fail. This may seem like a bug in visitdir(), but it's explicitly
documented to be undefined for subdirectories of excluded
directories. When using treemanifests, the walk would not descend into
foo/, so verification would pass. However, when using flat manifests,
there is no recursive directory walk and the file path "foo/bar/baz"
would be passed to _validpath() without "foo/" (actually without the
slash) being passed first. As explained above, _validpath() would
return true for the file path and "hg verify" would fail.
Replacing the _validpath() method by a matcher seems like the obvious
fix. Narrowhg can then pass in its own matcher and not have to
conflate the two matching functions (for dirs and files). I think it
also makes the code clearer.
The comment was introduced in 0a14c8556910 (rebase: ensure rebase
revision remains visible (issue4504), 2015-01-27), which mentions the
right issue in the description.
0b5f1f2efc77 introduced handling of a crash in this case. A review comment
suggested that it was not entirely obvious that a 'dm' always would have a 'r'
for the source file.
To mitigate that risk, make the code more conservative and make less
assumptions.
When this test was introduced, it used the short-form of all the flags
on this update invocation. I suspect, based on the "start with clean
dirstates" comment and the fact that the no-exec branch of the #if
guard leaves dirstate clean, that this should have been 'update -qCr'
instead of 'update -qcr', but that a bug in largefiles --check
handling left this problem unnoticed.
I'll leave a breadcrumb further up about the current failure mode in
the hopes that we can fix this some day.
This was previously discussed in [0] but the trail in that thread goes
cold after a few replies. Given that this is still a flaky test, that
appears to only be passing by bad fortune, I think it's worth
correcting the code of the test to make a correct assertion, and to
keep track of the suspected bug with some other mechanism than an
invalid test (if we had support for "expected failure" blocks this
might be a worthwhile use of them?).
0: https://www.mercurial-scm.org/pipermail/mercurial-devel/2016-October/089501.html
I spent some time confused by this test. I'm pretty sure that this
line intends to be cleaning the dirstate, not checking that it's clean
before updating: the preceding #if block leaves the dirstate clean in
the noexec case, and dirty in the exec case, so we can't expect
consistent behavior across that platform variation. A subsequent patch
will modify this command to use --clean instead of --check.
I'll elaborate in that patch about the hypothetical bug here.
Work around that 'dm' in the data model only can have one operation for the
target file, but still can have multiple and conflicting operations on the
source file where the other operation is a 'rm'. The move would thus fail with
'abort: No such file or directory'.
In this case it is "obvious" that the file should be removed, either before or
after moving it. We thus keep the 'rm' of the source file but drop the 'dm'.
This is not a pretty fix but quite "obviously" safe (famous last words...) as
it only touches a rare code path that used to crash. It is possible that it
would be better to swap the files for 'dm' as suggested on
https://bz.mercurial-scm.org/show_bug.cgi?id=5020#c13 but it is not entirely
obvious that it not just would create conflicts on the other file. That can be
revisited later.
dict.keys() is documented to return a copy, so it's surprising that
sortdict.keys() did not. I noticed this because we have an extension
that calls readlocaltags(). That method tries to remove any tags that
point to non-existent revisions (most likely stripped). However, since
it's unintentionally working on the instance it's modifying, it
sometimes fails to remove tags when there are multiple bad tags in a
row. This was not caught because localrepo.tags() does an additional
layer of filtering.
sortdict is also used in other places, but I have not checked whether
its keys() and/or __delitem__() methods are used there.
GNU grep, when emitting a matching line that doesn't have a terminating
newline, will add an extra newline. Solaris grep passes the original line
through without the newline. This causes differences in test output when
looking at the last line of the output of get-with-headers.py, which
doesn't usually emit (and certainly doesn't guarantee) a terminating
newline.
Both grep implementations succeed in matching the requested pattern,
though, so rely on specifying the full pattern on grep's commandline
instead of expecting it in the output, and send the output to /dev/null.
I've seen this in a (misconfigured) FreeBSD jail which has ::1 as an
entry for localhost, but IPv6 support is disabled in the jail. It took
me months to figure out what was going on (and I only figured it out
when tinyproxy.py got confused by similar IPv4-level misconfiguration
of the localhost domain in /etc/hosts.)
I don't feel strongly about this patch: on the one hand, it's papering
over a host-level misconfiguration, but on the other it avoids some
weird and hard to diagnose problems that can occur in weirdly
restricted environments.
outgoing() and remote() may stall for long due to network I/O, which seems
unsafe per definition, "whether a predicate is safe for DoS attack." But I'm
not 100% sure about this. If our concern isn't elapsed time but CPU resource,
these predicates are considered safe. Perhaps that would be up to the
web/application server configuration?
Anyway, outgoing() and remote() wouldn't be useful in hgweb, so I think
it's okay to ban them.
Only FreeBSD seems to be this picky. Note that this explicit
absolute-path `cd` exposes a defect in the test, in that we end up
still inside the cwd-vanish repository, but that's not a regression in
this change. Since we're in a code freeze, I'm doing the smallest
thing possible to try and fix bugs on FreeBSD, rather than cleaning up
the entire problem. I'll follow up with a more complete fix after the
freeze.
We've made chg utilize the common code path implemented in pager.py (by
e8fb65f5e551 and e97133c7a9dc), but the chg server does not always start
with a tty. Because of this, uisetup() of the pager extension could be
skipped on the chg server.
Kudos given to Sean Farley for dogfooding new chg and spotting this problem.
This patch makes it possible to unshelve while having missing files
in your repo as long as shelved changes don't touch those missing files.
It also makes error message better otherwise.
statprof has a __main__ handler that allows viewing of previously
written data files. As Yuya pointed out during review, 82ee01726a77
broke this. This patch fixes that.
Until callsites are updated, this will have no effect. Once callsites
are updated, specifying experimental.editortmpinhg will create editor
temporary files in a subdirectory of .hg, which will make it easier
for tool integrations to determine what repository is in play when
they're asked to edit an hg-related file.