Bug#1050432: rpy2: FTBFS on mips64el

[Paul Gevers]

Hi Dirk, mips-porters,

On 24-08-2023 17:25, Dirk Eddelbuettel wrote:
> Paul: While I have you here, and as all those failures are on _complex_, is
> there a known issue with mips64el or a specific compiler switch that may be
> needed? rpy2 is fairly widely used but also 'glue' around other Python
> libraries, are any others affected?

I don't know. I guess we should ask the mips* porters (done so by
putting them in TO).

Paul

[Dirk Eddelbuettel]

On 27 August 2023 at 14:09, YunQiang Su wrote:
| Dirk Eddelbuettel <e...@debian.org> 于2023年8月27日周日 00:15写道：
| >
| >
| > Hi all,
| >
| > As the test failures for complex valued variables appeared to be systemic on
| > the 'mips64el' platform, I buckled down, taught myself some Python ==:-) and
| > conditioned the number of failing tests away via
| >
| > @pytest.mark.skipif(platform.machine() == 'mips64' and sys.byteorder == 'little',
| > reason="Complex tests fails for 'mips64el'.")
| >
| > Maybe the porters team can shed some light on why we needed it, and if this
| > worked (the autobuilders will tell us soon enough) I can pass the patch on to
| > Laurent for a possible inclusion upstream.
| >
|
| Sorry for the late reply. I can work on it.

Appreciate that!

(While my fix corrected the build, it is a stop-gap as it avoids the issue. A
real fix would be good.)

| Do you knwo any way to run a single testcase?

Can you start from the unit tests I conditioned out? Each of those has
simple expression with a complex in it. Those should be executable directly
in the Python interpreter. You probably need all the build-deps (python, R,
rpy2, numpy, ...) installed.

Dirk

--
dirk.eddelbuettel.com | @eddelbuettel | e...@debian.org

[Dirk Eddelbuettel]

On 27 August 2023 at 18:44, YunQiang Su wrote:
| Maybe there is something wrong with ffi. (In fact the complex support of mips
| was added by me. ;)

Hah.

| I am looking for a way to use debug to debug the extensions.
| If you have any documents, can you point it to me.

I can help you with R, I think here the issue is more on the Python side.

But to the best of my knowledge, the applications languages generally just
call into the C library 'and assume things work'. But maybe here we need
another compile-time / link-time option to turn ffi on?