Compilation error on 4.02.3 on Mac OSX

[dannywillems]

I didn't investigate. See https://travis-ci.org/ocsigen/ocsigen-start/jobs/174512894.

[vasilisp]

Does ln -s -f work on Mac OS X?

[dannywillems]

@vasilisp I think so.

[dannywillems]

I close. I can't reproduce.

[vasilisp]

@dannywillems, re-opening because I wouldn't like that to happen to users, during opam-repository CI jobs, etc.

Looks like a race condition in our Makefile. I find these two continuous lines from the Travis log suspicious:

# ln -s -f ../ocsipersist.mli .
# ln -s -f ../ocsipersist.mli .

Speculation: different make threads try to make the link at the same time, and the link is created between the check implied by -f and the actual action?

No idea what to do!

[vasilisp]

We could remove the -j%{jobs}%, I think the trouble started around the time we added it. It shouldn't be too big of a deal. OPAM will still do adequate parallelization (different packages in parallel), it will just be coarser-grained.

[dannywillems]

@vasilisp with GNU ln, ln -sf doesn't exit with error when the file exists. I will try for BSD ln.

EDIT: It's also working on BSD. No idea for the moment.

[vasilisp]

@dannywillems I know about GNU, that's why I added the -f. Judging by the manual, it should be the same for OS X, and that's exactly what makes the file exists error hard to understand :).

Anyway, it is about parallel builds, and I have disabled them, so don't worry too much :).