[Commented On] D9952: revlog: add a mechanism to verify expected file position before appending

[marmoute (Pierre-Yves David)]

marmoute added a comment.


  In D9952#159118 <https://phab.mercurial-scm.org/D9952#159118>, @spectral wrote:
  
  > I'm unable to reproduce. I've run the test over 10,000 times (I added a `#testcases a b c d e f g h i j k l m n o p q r s t u v w x y z` so it ran 26 times each run, and I've run over 300 instances of that like `run-tests.py -j26 -l --chg test-racy-mutations.t`, over 100 with -j108, and other combinations (with and without the added testcases, with and without `--chg`, etc.)
  > I'm on a rather beefy linux machine, based on Debian testing, using Python3.9, on a commit descended from 856820b4 <https://phab.mercurial-scm.org/rHG856820b497fcf31b7904e0b160c318437581d0c0>.
  > It hasn't failed a single time, let alone like that. I also don't know what would cause that kind of failure. :(
  
  The error seems to hint at at attempted cleanup to a not fully initialized `dirstateguard`. So it might be the symptom of another failure that probably have a narrow windows for happening.
  
  The test does not use a python extensions so it looks like that racy windows exists in the main code. @spectral, What is the range of code you want to ensure a race with ? and how is the synchronization happening to reach it ?

REPOSITORY
  rHG Mercurial

CHANGES SINCE LAST ACTION
  https://phab.mercurial-scm.org/D9952/new/

REVISION DETAIL
  https://phab.mercurial-scm.org/D9952

To: spectral, indygreg, #hg-reviewers
Cc: marmoute, Alphare, martinvonz, joerg.sonnenberger, mercurial-patches
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mercurial-scm.org/pipermail/mercurial-patches/attachments/20210420/20a19b5f/attachment-0002.html>