[bug#57540] [PATCH] Please rebase (was: Add ocaml-elpi (a dependency of coq-mathcomp-analysis))

[Garek Dyszel]

Hi again,

It looks like Coq has been updated to 8.16 now, which means the two
packages required by coq-mathcomp-hierarchy-builder in this patchset are
now out of date. The build processes have completely changed for
ocaml-elpi and coq-elpi.

The new ocaml-elpi build system got rather confusing and will likely
take me much longer than I originally expected. Maybe we could close
this issue? I think it might be easier if I were to send in ocaml-elpi
on its own, for example.

Let me know.

Thanks!
Garek

At 09:04 2022-09-27 UTC-0400, Garek Dyszel <garekdyszel@disroot.org> wrote:
> Hi Julien and simon,
>
> I planned to write back yesterday but had to run out the door
> unexpectedly.
>
> At 18:52 2022-09-26 UTC+0200, zimoun <zimon.toutoune@gmail.com> wrote:
>> For instance the series reads, ... where the logic about the order is
>> not obvious
>
> The logic was essentially adding dependencies in reverse order. I
> started with the package that I wanted to build (coq-mathcomp-analysis),
> and added the dependencies as I found they were needed.
>
> I'll stick with committing dependencies in forward order (committing
> dependencies before packages) from now on.
>
>> I have tried to clean the mess but I give up for now. :-) It would be
>> much easier if the series provides,
>>
>>  1. the Git commit against which revision these patches apply (see the
>>     option --base of git-format-patch)
>>
>>  2. the correct dependency order of the patches
>
> Probably it would also be easier to start over from the new master
> branch and recommit the remaining packages in the proper order.
>
> If I don't have another major interruption, I will send out a new set of
> commits, in the correct order, formatted with --base, before or by
> Friday. Excluding those packages which were already pushed to master, of
> course :)
>
> At 20:39 2022-09-24 UTC+0200, Julien Lepiller <julien@lepiller.eu> wrote:
>> No need to repeat the license here. Also, this means that the license
>> should be lgpl2.1+, instead of plain lgpl2.1.
>
> Ah, seems like I was getting lost in package-ception there and didn't
> check over the descriptions too rigorously. I'll keep this in mind when
> preparing the next patchset for this thread.
>
>> For python packages, I see you add python to the inputs. Why is that?
>> The python-build-system already provides python.
>
> I had been getting errors of the form "python3 was not found on the
> PATH" during the 'configure' phase of some python packages, even though
> the python-build-system was being used. I added python to everything to
> avoid such errors, but forgot to remove it for packages where it was not
> really needed.
>
> If I can find the first package that produced that error, I'll submit a
> bug report for it with the precise error quoted. 
>
>> It looks like python-jsonschema-next (4.5.1) does not have any
>> dependent, so updating this package instead might be better than
>> adding a new one, wdyt?
>
> I found evidence to the contrary, I think. With graphviz installed, I
> ran
>
> $ guix graph python-jsonschema > /tmp/py-js-deps.dot
> $ gc -n /tmp/py-js-deps.dot
>
> which says that there are 186 nodes, or (186 - 1) = 185 packages
> dependent on python-jsonschema-next. If you prefer viewing it as an
> image,
>
> $ dot /tmp/py-js-deps.dot > /tmp/py-js-deps.png
> $ feh /tmp/py-js-deps.png
>
> shows all 185 packages originating from the node named
> "python-jsonschema@4.5.1".
>
> Maybe for now we could add this transitional python-jsonschema-4.15 to
> build coq-mathcomp-analysis, and remove it in a subsequent patchset? I
> don't want to tie this patch up unnecessarily.
>
> If I have malformed patches now with only 20 packages,... well, let's
> just say I don't know if I want to see the results just yet, if I'll
> need to rebuild 185 :)
>
> -- Garek