Re: [fortran, patch] Implement EXECUTE_COMMAND_LINE

[FX <fxcoudert at gmail dot com>]

> Please add a @ref to SYSTEM, pointing to the new intrinsic; maybe one should also recommend there to use the F2008 intrinsics (similarly to GETARG/IARGC).

OK.

> +The @code{COMMAND} argument is passed to the shell and executed, using
> +the @code{system()} call.  (The shell is @code{sh} on Unix systems, and
> 
> I wonder whether one should add a qualifier before "system", e.g. "the POSIX's system call" or the C library's system call" to make clear that it is not Fortran's SYSTEM call.

I changed to "the C library's system() call".

> +execute_command_line (const char *command, bool wait, int *exitstat,
> 
> Using 'bool wait' instead of "bool *wait" looks wrong.

No call to execute_command_line is emitted directly by the front-end. The library's entry points are execute_command_line_i4 and execute_command_line_i8, which in turn call  execute_command_line(). They take care of this and other issues as absent optional arguments (CMDSTAT and EXITSTAT).

>  set_attr_value (5, false, true, false, false, false);
> (or shorter: set_attr_value (2, false, true))

Whow, that's new, and that's neat!

> +#if defined(HAVE_FORK)
> +  if (!wait)
> +      *cmdstat = 0;
> +      if ((pid = fork())<  0)
> +	*cmdstat = EXEC_SYSTEMFAILED;
> 
> That won't work if cmdstat is not available. As F2008 states:
> 
> "If a condition occurs that would assign a nonzero value to CMDSTAT but the CMDSTAT variable is not present, error termination is initiated."

I missed that sentence, thanks for pointing it out.

> +      /* Synchronous execution.  */
> +      int res = system (cmd);
> +
> +      if (!wait)
> +	*cmdstat = -2;
> 
> 
> Ditto for the cmdstat/exitstat. Additionally, I read the standard such that the intrinsic immediately returns with cmdstat = -2 if HAVE_FORK is false - not after the synchronous execution. However, I think the standard is not unambiguous.

I agree with you that it's ambiguous, but I think the answer is in the definition of the WAIT argument: "If WAIT is present with the value false, and the processor supports asynchronous execution of the command, the command is executed asynchronously; otherwise it is executed synchronously."

I think that part is unambiguous: WAIT = .true. and !HAVE_FORK falls into the "otherwise": the command is actually executed.



I have two testsuite failures on x86_64-darwin: trim_optimize_1.f90 and widechar_intrinsics_5.f90. I don't know how they could possibly be related, but my baseline testsuite run is still running.
OK to commit if regtest completes OK?

FX

Attachment: exec_2.ChangeLog
Description: Binary data

Attachment: exec_2.diff
Description: Binary data