Re: libsvn_client/commit.c:remove_tmpfiles()

kfogel@collab.net writes:

> Well, this is interesting. I simply commented out the
> svn_io_check_path() and the conditional guard, so that we'd
> unconditionally run svn_io_file_remove() on the tempfile.
>
> To my surprise, that caused a bunch of test failures (see below).
> Maybe there's some code path by which tmpfiles are removed early?

My guess is that most of the time the files are removed before
remove_tmpfiles is called. I suspect that only after certain commit
failures, a commit race leading to an fs merge failure say, will there
ever be files for remove_tmpfiles to delete.

You could probably reinstate the conditional guard, replace the
svn_io_file_remove with an abort, and still have the regression tests
pass--however stress.pl may well fail.

> Next thing to try is the APR_EEXIST error check; perhaps all of the
> errors below were simply due to that.

That should be OK.

-- 
Philip Martin
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org