[QGIS-Developer] SIP coverage test removal and Travis optimization

[Denis Rouzaud]

Hi Nyall,

Thanks for the feedback.

Le mer. 31 mai 2017 à 23:14, Nyall Dawson <nyall.dawson at gmail.com> a écrit :

> On 31 May 2017 at 22:15, Denis Rouzaud <denis.rouzaud at gmail.com> wrote:
> > Dear all,
> >
> > Our automatic SIP files generation is almost there and I would like to
> > remove the SIP coverage test from running on Travis.
>
> > Is this acceptable?
>
> In general, but...
>
> > Is there any reason to keep the test (running on Travis)?
>
> There's 2 extra things the sip coverage test does which the sipify test
> doesn't:
>
> 1. Flags if a class is totally missing from the bindings. The sipify
> test can't detect when someone's added a new header to the project
> without adding a corresponding .sip include to the
> core/gui/server/analysis.sip files. Ideally this would be fixed with
> some cmake magic which avoids the need to even manually build the
> core.sip/gui.sip/etc files...
>

So, we have 3 types of headers in core:
1) automatic SIP file creation
2) manual SIP file creation (might be not needed anymore at some point)
3) no SIP file creation

What I would propose is to use some macro definition in the header for case
2) and 3), like:
#define SIP_MANUALLY_CREATED
or
#define SIP_NOT_GENERATED

By default, the SIP file would be automatically created.

The core/gui/etc.sip could then automatically be generated either by CMake
or by a script.

This would depend if we want to keep the SIP files in the source. I tend to
think they're quite useful there for inspection and that the sipify script
is (still?) too fragile to fully trust it.
If we keep them in source, does it make sense to use CMake to build the
core.sip file? or a script run by prepare-commit.sh as it is now?

What do you think?




>
> 2. Checks that the methods *are* actually accessible in the Python
> bindings. Occasionally there'll be weird things where even though a
> method is correctly listed in the .sip files, it just doesn't exist in
> the Python library. There's no pattern here - it happens differently
> on different platforms. But this is a sip bug, so I don't really care
> if we lose this extra check.
>
> For me 1 would need to be addressed before we can drop the test.
>
> Nyall
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/qgis-developer/attachments/20170602/04e74960/attachment-0001.html>