[mapserver-dev] travis-ci msautotest behavior changes

[Even Rouault]

Le lundi 20 juillet 2015 17:20:10, thomas bonfort a écrit :
> Hi devs,
> 
> The current travis-ci/msautotest setup required us to use a submodule
> inside our mapserver git repository, which was somehow a pain to use
> as a submodule tracks a specific commit SHA and not a whole branch:
> 
> - the submodule's sha needed to be maintained/updated/committed each
> time there was a change to the msautotest repository itself
> - it was nearly impossible to have travis-ci run using an msautotest
> branch from a repo outside of the mapserver organisation, thus making
> the creation of pull requests difficult to validate if they required
> an update to the autotests themselves.
> 
> I've updated the travis test script to now behave like this:
> 
> - by default, if there's a commit on mapserver/mapserver 's
> branch-7-0, then the autotests from mapserver/msautotest 's branch-7-0
> will be used
> - you can specify which msautotest repo/version should be used by
> specifying it in the commit message, with a line containing (only) the
> string msautotest=repo at sha, e.g.
> 
> msautotest=git://github.com/mapserver/msautotest.git@master
> 
> or
> 
> msautotest=git://github.com/tbonfort/msautotest.git@my-feature-branch
> 
> or
> 
> msautotest=git://github.com/myuser/msautotest.git@a457de23ea

Hi Thomas,

I agree with your observations on the current situation;

How does that work with a pull request made of several commits ? I suppose 
that only the commit with msautotest= in its message gets tested with the 
specified msautotest version, and others fallback to head of msautotest's 
branch-7-.0 ? So consequence is : add msautotest= to all commits of a PR ?

Not sure if there's a developer guide somewhere, but that would be worth 
adding the above in it.

Question: is there a strong reason for not having msautotest has a normal 
directory of mapserver repo, and docs too ? too big pershaps ? But I'd hope 
that all developers have mapserver, msautotest and docs checked in, so the 
size argument doesn't really seem relevant.
Personnaly, I'd find it great/simpler to have a single commit with related 
code, test and doc changes.

Even


-- 
Spatialys - Geospatial professional services
http://www.spatialys.com