[Bug fortran/66528] [6 Regression] unbalanced IF/ENDIF with -fmax-errors=1 causes invalid free

[tkoenig at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66528

--- Comment #5 from Thomas Koenig <tkoenig at gcc dot gnu.org> ---
(In reply to Manuel López-Ibáñez from comment #4)
> diagnostic_finish tries to free the output_buffer,
> but the error_buffer is statically allocated. I think this should be enough:
> 
> --- error.c     (revision 223651)
> +++ error.c     (working copy)
> @@ -1379,12 +1379,12 @@ gfc_error_check (void)
>        output_buffer *tmp_buffer = pp->buffer;
>        pp->buffer = pp_error_buffer;
>        pp_really_flush (pp);
>        ++errorcount;
>        gcc_assert (gfc_output_buffer_empty_p (pp_error_buffer));
> -      diagnostic_action_after_output (global_dc, DK_ERROR);
>        pp->buffer = tmp_buffer;
> +      diagnostic_action_after_output (global_dc, DK_ERROR);
>        return true;
>      }
> 
>    return false;
>  }
> 
> However, a better fix may be to make the error_buffer also dynamically
> allocated like the warning_buffer. Not sure why I did the change.

I think either version would be OK, but I agree that making
the error_buffer dynamic would be better.

> (It would be nice to have a testcase testing this in the regression
> testsuite.)

Easy enough to generate from the test case in comment #2.