Re: Bug#780640: gcc-5: merge gnat back to gcc
On Fri, Apr 3, 2015 at 7:10 AM, Matthias Klose <doko@debian.org> wrote:
> On 03/29/2015 04:54 PM, YunQiang Su wrote:
>> On Thu, Mar 26, 2015 at 10:10 PM, YunQiang Su <wzssyqa@gmail.com> wrote:
>>> On Tue, Mar 24, 2015 at 11:24 PM, Matthias Klose <doko@debian.org> wrote:
>>>> some comments:
>>>>
>>>> - please add appropriate changelog entries
>>>
>>> Added
>>>
>>> * Rewrite patches for libgnat build, and add mips64el support.
>
> YunQiang,
>
> spent some time working on your patch. I replied on general stuff at
> https://lists.debian.org/debian-gcc/2015/03/msg00142.html
>
> the things that were not mentioned in this report:
>
> - dropping of patches without any mentioning and without any
> effort for forward porting these patches.
>
> - how your patches were tested. At least I found that
> building libgnatprj without -DIN_GCC results in a build error.
>
> - please state how you built an existing package with your patch.
> It may build on some mips64 configuration, but please check
> that a native amd64 build still works.
>
> - updated patches in this report add a gnat cross base package,
> again not mentioned, and not part of "merge gnat back to gcc".
>
>>> * Use the same scheme as gcc etc for gnat commands:
>>> aka gnat-5 -> <triplet>-gnat-5.
>
> pretty please send a separate patch, and explain why this would work. At least
> afaicr Ludovic didn't do that, because the tools are calling each self with
> fixed names.
Yes, in the upstream, it calls the commands without version suffix.
While to make it the same way works like other languages,
To have a version suffix may be our better choice,
and then let gcc-default to link them.
>
>>>> - don't rely on autogen during the build. I was just
>>>> happy to get rid off it. I think it's fine
>>>> to keep the auto generated toplevel Makefile.
>>>> It doesn't change that often.
>>>
>>> OK, add a patch named: src-Makefile-autogen.diff
>
> No. just add it to the single patches. Now done.
In fact to add it in to a single patch will complex it.
We have so many patches touch src/Makefile.in,
to do so, we will meet a situation patch over patch, patch over patch ...
So for me, the autogen may be a better choice.
as we can drop changes for this file in all patches, (even for other languages).
>
>>>> I think it makes sense to build gnat out of the gcc-5 source package. I talked
>>>> to Ludovic about this at Fosdem, and he didn't have major concerns. And
>>>> probably he'll update Ada only once during the next release cycle, and that
>>>> likely will be for GCC 6.
>>>
>>> I switch gnat off in gcc by default.
>>> So we can still use gnat-5 packages.
>>> I tested it on amd64, it build well.
>>>
>>
>> I saw you merged gnat back to gcc, then why are you still use the old flavor?
>> aka, the command is gnat not gnat-5.
>
> I don't understand your last comment. Please elaborate.
I meant the scheme converting: /usr/bin/gnat -> /usr/bin/gnat-5.
To have the same behavior with other languages, I perfer to use
the same way, aka with a version suffix for commands.
>
>> This way will make cross build impossible.
>> Any consideration?
>
> yes, any patch which makes a native build fail to build is broken. Please check
> your patches with a native build (which is known to work) first.
>
Sorry for sending a so big patch.
I did test it native, on both mips64el and amd64.
Maybe some upstream changes make it fails to build now.
> I'm closing this issue, please open separate issues for any remaining issues.
OK, I will try to split it.
>
> Matthias
>
--
YunQiang Su
Reply to: