Fix regression with -o when compiling C files

[nojb]

A friendly reminder that we shouldn't forget about this regression introduced in #9960 that needs to be fixed before 4.13 is released.

There are two issues with this patch:

1. It broke previous behaviour of ocaml when you don't pass -o:

(before) ocamlc foo.c --> calls "$cc -c foo.c"
(after) ocamlc foo.c --> calls "$cc -o foo.o -c foo.c"

This causes problems when people use:

ocamlc -ccopt '-o whatever' foo.c

which was necessary before this PR.

2. The -o flag to ocaml{c,opt} is ignored if -c is not passed as well when compiling C files.

Originally posted by @nojb in #9960 (comment)

cc @shindere

[shindere]

I didn't forget, thanks!

[shindere]

I checked that #10619 does fix the reported problems. May main concern is tomake sure it does not introduce others. I made a mistake in the PR title, writing #9660 rather than #9960. Would somebody please be kind enough and fix that?

[nojb]

Would somebody please be kind enough and fix that?

Fixed.

[Octachron]

@nojb concerning your second point, as far as I can see, this behavior has been constant since at least OCaml 4.04? In particular,

ocamlopt -o exe a.c b.ml

generates an executable named exe.

[nojb]

@nojb concerning your second point, as far as I can see, this behavior has been constant since at least OCaml 4.04? In particular,

Indeed, that seems to be the case...

[shindere]

Nicolás Ojeda Bär (2021/09/08 06:10 -0700):

> @nojb concerning your second point, as far as I can see, this behavior has been constant since at least OCaml 4.04? In particular, Indeed, that seems to be the case...

How about just ``` ocamlc -o hello hello.c ``` In my tests this was generating an object file, not an executable. And my understanding was that my patch in #9960 did change this, and htat I needed to fix that, too. Taht's why I added the "else" part. But I forgot to re-run the testsuite after that. Before I added the "else" part, with just the other part of hte patch, the testsuite was happy.

[Octachron]

The issue is that without -c the object file is not the final target of the compilation command, and thus

ocamlc -o anything hello.c

generates hello.o since at least OCaml 4.00.

[shindere]

Everything seems clear to me now, thanks. I pushed a version of the code that does not try to fix a problem that does actually not exist, many thanks to you, @Octachron, for having spotted that.

[gasche]

Would it be possible to have tests in the testsuite for the behaviors that we have decided to preserve?

[Octachron]

It does not seem trivial to add test for the C compilation command, but it might it be a good idea on the medium term.

Since #10619 fixed the regression with out-of-thin-air -o flag, I am closing this specific issue. @nojb , do you think that the not obvious behavior with ocamlc -o anything hello.c warrants an issue?

[nojb]

@nojb , do you think that the not obvious behavior with ocamlc -o anything hello.c warrants an issue?

Done! #10625