Re: CVS commit: pkgsrc/lang/gcc8

[Roland Illig <roland.illig%gmx.de@localhost>]

On 27.04.2020 11:51, Jonathan Perkin wrote:
> * On 2020-04-27 at 04:38 BST, Roland Illig wrote:
>
> > The file gcc/Makefile.in does not contain GCC_TARGET_MACHINE, and there
> > is no patch that adds this placeholder. The related words that actually
> > appear in Makefile.in are -DTARGET_MACHINE, $(target_noncanonical),
> > $(target), $(real_target_noncanonical).
>
> FWIW, this is a quick indicator that the gcc8 package is pretty
> broken.  This variable was previously used as an alias for
> MACHINE_GNU_PLATFORM and used to substitute the correct value into
> patch-gcc_Makefile.in.
>
> Given that it's been broken since gcc8 was imported (just compare that
> patch to gcc7) I guess either not many people use gcc8 or they're just
> running into issues that they haven't diagnosed back to this yet.
>
> Not saying this commit is wrong, just pointing out that sometimes when
> something appears to do nothing it might be pointing to other
> breakage.

I know.

In this case I didn't research further where all this came from. Instead
I wrote that detailed comment hoping that someone would make sense of
it. You did. :)

For several other packages I think I found out the real reason behind
SUBST being a no-op and fixed these, but for GCC I just didn't have any
clue.

It would be really cool if every important patch came with a test case
that would fail without the patch. But that's probably too much asked.
Or maybe it's not? We already have some testing infrastructure ready in
regress/. It would be easy to add a "see regress/gcc-rpath-patch" note
to the patch. The more difficult part would be to write the actual test,
but that should be doable. I'll try that the next time.