On 29-07-2022 17:26, Artyom V. Poptsov wrote:
> Hello Maxime,
>
> thanks you for the patch review!
>
>> Often GUIs have problems with finding icons, could you check whether
>> qucs-s works in a pure environment?
> I tried to run Qucs-S installed through Guix, like the follows:
>
> --8<---------------cut here---------------start------------->8---
> guix shell -D guix
> ./pre-inst-env guix shell -D --pure qucs-s
> qucs-s
> --8<---------------cut here---------------end--------------->8---
>
> And all the icons seem to be fine
>
> Although it can be that I don't fully understand what you mean by "pure
> environment" in that case.


That's what I meant, thanks.  I guess the icons issue is GTK-specific 
and doesn't happen for Qt.

>> Propagation is not very robust (e.g. if run directly from the store with $(guix
>> build qucs-s)/bin/the-binary), would it be possible to not propagate this? E.g.,
>> if you are propagating it because of not finding a binary, you could use
>> substitute* to insert an appropriate absolute file name.
> Okay, that's a good point.  I checked the sources for the places where
> Qucs-S uses executable paths and added substitutions for that.
These substitutions look fine ...

> + (("QucsSettings.NgspiceExecutable = ngsp_exe")
> +                      (string-append "QucsSettings.NgspiceExecutable 
> = \""
> +                                     (assoc-ref inputs "ngspice")
> +                                     "/bin/ngspice"
> +                                     "\""))
> +                     (("QucsSettings.OctaveExecutable = \"octave\" + 
> QString(executableSuffix)")
> +                      (string-append "QucsSettings.OctaveExecutable = \""
> +                                     (assoc-ref inputs "octave")
> +                                     "/bin/octave"
> +                                     "\""))))))))
... but they can be improved, by replacing the assoc-ref with 
search-input-file: (search-input-file inputs "/bin/ngspice"). That way, 
it doesn't depend on the package name anymore, which is preferred by 
<https://guix.gnu.org/blog/2021/the-big-change/> (*) and makes in some 
cases --with-input more usable. That blog post also has en example.

(*) Don't follow the first 'hello' example though -- it showcases 
possibilities, not what should be done, and interferes with --with-input.

> I kept NGSpice in "propagated-inputs" because Qucs-S requires it to do
> the simulations in runtime.
>
> Also I've added GNU Octave to the "propagated-inputs" and substituted
> the path to Octave too.
By that logic, since qtbase and qtsvg are used at runtime too, they 
should be propagated as well, but ...
> I tried to run simulations from the examples provided with the Qucs-S
> and it seems to me that Qucs-S mostly works as it should.

... as you have observed, things work even when they aren't propagated 
(at least for qtbase etc., ngspice and octave have not yet been tested).

Less experimentally, the propagated- in propagated-inputs means only 
that: propagation -- the dependencies are still installed even if they 
aren't propagated, because the substitute* inserts a reference so the GC 
knows to not delete it, the only difference is that the dependency isn't 
automatically put in the profile when not propagating.

In theory, the propagation shouldn't be required because you added a 
'substitute*', so in principle qucs-s should know where to find it.

Also, I noticed these substitutions modify configuration, could you 
verify they aren't saved in wherever qucs-s' configuration file is 
located? Because if they are, then even after an update of octave etc. 
it would seem that qucs-s would still use the old octave.

Greetings,
Maxime.