push: rework the computation of fallbackheads to be correct
The previous computation tried to be smart but ended up being wrong. This was
caught by phase movement test while reworking the phase discovery logic to be
faster.
The previous logic was failing to catch case where the pushed set was not based
on a common heads (i.e. when the discovery seemed to have "over discovered"
content, outside the pushed set)
In the following graph, `e` is a common head and we `hg push -r f`. We need to
detect `c` as a fallback heads and we previous failed to do so::
e
|
d f
|/
c
|
b
|
a
The performance impact of the change seems minimal. On the most impacted
repository at hand (mozilla-try), the slowdown seems mostly mixed in the
overall noise `hg push` but seems to be in the hundred of milliseconds order of
magnitude. When using rust, we seems to be a bit faster, probably because we
leverage more accelaratd internals.
I added a couple of performance related common for further investigation later
on.
Test that, when an hg push is interrupted and the remote side receives SIGPIPE,
the remote hg is able to successfully roll back the transaction.
$ hg init -q remote
$ hg clone -q ssh://user@dummy/`pwd`/remote local
$ SIGPIPE_REMOTE_DEBUG_FILE="$TESTTMP/DEBUGFILE"
$ SYNCFILE1="$TESTTMP/SYNCFILE1"
$ SYNCFILE2="$TESTTMP/SYNCFILE2"
$ export SIGPIPE_REMOTE_DEBUG_FILE
$ export SYNCFILE1
$ export SYNCFILE2
$ PYTHONUNBUFFERED=1
$ export PYTHONUNBUFFERED
On the remote end, run hg, piping stdout and stderr through processes that we
know the PIDs of. We will later kill these to simulate an ssh client
disconnecting.
$ remotecmd="$RUNTESTDIR/testlib/sigpipe-remote.py"
In the pretxnchangegroup hook, kill the PIDs recorded above to simulate ssh
disconnecting. Then exit nonzero, to force a transaction rollback.
$ cat >remote/.hg/hgrc <<EOF
> [hooks]
> pretxnchangegroup.00-break-things=sh "$RUNTESTDIR/testlib/wait-on-file" 10 "$SYNCFILE2" "$SYNCFILE1"
> pretxnchangegroup.01-output-things=echo "some remote output to be forward to the closed pipe"
> pretxnchangegroup.02-output-things=echo "some more remote output"
> EOF
$ hg --cwd ./remote tip -T '{node|short}\n'
000000000000
$ cd local
$ echo foo > foo ; hg commit -qAm "commit"
(use quiet to avoid flacky output from the server)
$ hg push --quiet --remotecmd "$remotecmd"
abort: stream ended unexpectedly (got 0 bytes, expected 4)
[255]
$ cat $SIGPIPE_REMOTE_DEBUG_FILE
SIGPIPE-HELPER: Starting
SIGPIPE-HELPER: Redirection in place
SIGPIPE-HELPER: pipes closed in main
SIGPIPE-HELPER: SYNCFILE1 detected
SIGPIPE-HELPER: worker killed
SIGPIPE-HELPER: creating SYNCFILE2
SIGPIPE-HELPER: Shutting down
SIGPIPE-HELPER: Server process terminated with status 255 (no-windows !)
SIGPIPE-HELPER: Server process terminated with status 1 (windows !)
SIGPIPE-HELPER: Shut down
The remote should be left in a good state
$ hg --cwd ../remote tip -T '{node|short}\n'
000000000000
#if windows
XXX-Windows Broken behavior to be fixed
Behavior on Windows is broken and should be fixed. However this is a fairly
corner case situation and no data are being corrupted. This would affect
central repository being hosted on a Windows machine and accessed using ssh.
This was catch as we setup new CI for Windows. Making the test pass on Windows
was enough of a pain that fixing the behavior set aside for now. Dear and
honorable reader, feel free to fix it.
$ hg --cwd ../remote recover
rolling back interrupted transaction
(verify step skipped, run `hg verify` to check your repository content)
#else
$ hg --cwd ../remote recover
no interrupted transaction available
[1]
#endif