test, subprocess: Improve coverage report correctness

[hebasto]

As a child process uses the execvp() call, an explicit dumping of the collected profile information is required.

Coverage:

• on the master branch:
  

• with this PR:
  

This PR mostly follows gcc/libgcc/libgcov-interface.c#L320-L332:

/* A wrapper for the execvp function.  Flushes the accumulated
   profiling data, so that they are not lost.  */


int
__gcov_execvp (const char *path, char *const argv[])
{
  /* Dump counters only, they will be lost after exec.  */
  __gcov_dump ();
  int ret = execvp (path, argv);
  /* We reach this code only when execv fails, reset counter then here.  */
  __gcov_reset ();
  return ret;
}

[DrahtBot]

The following sections might be updated with supplementary metadata relevant to reviewers and maintainers.

Code Coverage

For detailed information about the code coverage, see the test coverage report.

Reviews

See the guideline for information on the review process.

Type	Reviewers
Approach ACK	ajtowns

If your review is incorrectly listed, please react with 👎 to this comment and the bot will ignore it on the next update.

[theuni]

Are there some docs somewhere that suggest this?

[hebasto]

Are there some docs somewhere that suggest this?

https://gcc.gnu.org/onlinedocs/gcc/Gcov-and-Optimization.html

UPD. The code has been taken from flux-framework/flux-core@a9bf417.

[ajtowns]

Approach ACK

[ajtowns]

These lines are indented too much compared to surrounding code.

Are they necessary at all? [EDIT: Probably yes, because they precede an _exit() call which presumably skips the cleanup/dump code that is only included in regular exit()]

[hebasto]

Well, reset seem unneeded here. I'm going to drop it.

[ajtowns]

The code in https://inbox.sourceware.org/gcc-cvs/20200505141638.8B852388F43D@sourceware.org/T/ suggests that __gcov_dump() should be called before exec() because otherwise a successful exec will cause the information to be discarded, and __gcov_reset() should be called after exec() returns, because that means exec failed and the information wasn't discarded, so coverage stats should continue to be gathered without duplicating the info that's already been dumped. Doing the reset unconditionally prior to the exec function this way seems like a reasonable alternative approach to that to me.

[hebasto]

Doing the reset unconditionally prior to the exec function this way seems like a reasonable alternative approach to that to me.

It looks better to me because it counts the failed execvp(), which returns.

[hebasto]

Addressed @ajtowns's feedback. Added comments. Updated the PR description.