Closed gasche closed 5 years ago
I tried to deal with a race in mkdir_p
and another in rm_f
, and now the reproduction script works reliably on my machine. I'm thinking of merging this, getting a minor release out, and waiting for more user reports to do more if needed.
I'm creating this PR to track my progress on fixing #300 and related race-conditions issue. The current reproduction testcase is as follows:
running
make setup; make -j all
exposes races in ocamlbuild's parallel accesses to the filesystem.I don't think we will ever try to guarantee that concurrent ocamlbuild invocations work robustly, but we should at least fix the issues that users encounter in the wild, when reasonably possible.