Bug#903694: maybe a proper fix

[Matthias Klose <doko@debian.org>]

On 18.07.2018 17:41, Nicolas Boulenguez wrote:
> Please consider the attached changes.
> 
> * apply the attached ada-verbose.diff
>   I fail to understand why anyone ever silents errors or command lines
>   nowadays.
> 
> * append the attached howto-test-path-with-cross-build to the README
>   of gcc-*-cross, or more experienced advices for beginners like me.

Thanks, your method only works if you have an upstream patch, not a patch for
the packaging. I have now changed the README in gcc-X-cross to:

"""
The gcc-X-cross package should carry no patches for the GCC build, however
when an upload of a gcc-X-cross package is required without doing a gcc-X
upload, then

 - prepare a new gcc-N package
 - save the debdiff of the existing and the new gcc-N package as
   updates.diff
 - Apply the updates.diff in the gcc-X-cross package (search debian/rules
   for updates.diff).
"""

> * replace debian/patches/ada-gcc-name.diff with the attached version.
>   It ensures in Osint.Program_Name that:
>   * arguments breaking implicit assertions trigger an immediate failure.
>   * if present in the program name, the suffix repeats the linked libgnat version.
>   * the executed subcommand *always* carries exactly one suffix.
>   It will hopefully close 903694 and friends.
> 
>   With gcc-7 in unstable, the reproducer_in_tree script called after a
>   patched build works as expected for
>   * 903694: all *gnatmake* recursive subcommands carry exactly one -7 suffix.
>   * 856274: all *gnatchop* succeed.
> 
> I suggest that we leave the bug open as a reminder to forward the
> changes once the dust has settled. Also, it would be nice to suggest
> that gnatchop should now behaves like the other tools.

I verified that I can build a gcc 8 cross compiler using this method, after
rebuilding gcc-8.