daily winetest generation

[Paul Millar]

Hi Saulius,

On Friday 06 Jan 2006 15:15, Saulius Krasuckas wrote:
> * On Fri, 6 Jan 2006, Paul Millar wrote:
> > WRT builds winetest.exe anyway (in fact, all the DLLs and EXEs it can),
> > but the results are not registered, so no one sees them.
>
> I didn't got it - when do the results get not registered?  Or when it
> does?

There are two distinct "registrations" here.

The first happens when winetest.exe is built.  The build process registers the 
availability of a new winetest.exe by sending an HTTP GET to some end-point 
at test.winehq.org with the corresponding URL for downloading winetest.exe.

As far as I understand it, this GET triggers some machinery at winehq end that 
causes any future winerash to download the new winetest.exe and execute it. A 
simple HTTP-redirect should work, but I don't know if that's how it works.

The second "registration" is when winetest.exe registers its results to 
winehq, which then makes these results available on the test.winehq.org 
pages.

> > However, dealing with asynchronous updates was thought both too
> > difficult and not necessary so we have the synchronous once-a-day
> > builds.  I'm maintaining a separate cron-triggered build process,
>
> Are its results public? :)

Yes.  Have a look at [4].

> What belongs to WRT and what does to your separate build process?

The files that are wine-<type>-<date><time>.zip are from WRT.  These are 
bundled executables and DLLs along with corresponding signatures.

Files that start "winetest-" are from the cron build process.

All these files have a "latest" (wine-dlls-latest,  winetest-latest, ...) that 
are sym-links to whatever is the latest build.  Downloading the "latest" 
should always work (although, thinking about it, using an HTTP redirect would 
be better).

A word of warning, though.  To get WRT to work, I need to patch the wine tree 
slightly.  The cron-built winetest doesn't include these patches as they seem 
to cause winetest.exe to crash under some (possibly all) circumstances.  I 
started to try and merge these changes into wine, but other things came up 
and I ran out of time :^(

[snip!]
> Well, I may not understand your reply correctly, but why then I see only
> 1 release during 2006 year at the of the page [2]:
>
>  [DIR] 200512271000/           06-Jan-2006 04:05    -
>  [DIR] 200601061000/           06-Jan-2006 05:50    -
>
> when WRT summary page [3] tells us there were several successful builds in
> this year already:
>
>  Thu Jan 5 17:29:03
>  Wed Jan 4 20:51:31
>  Wed Jan 4 15:00:50
>  Tue Jan 3 21:38:32
>  Tue Jan 3 14:38:04
>
> Has that happend due to lack of running Winrash clients during this time?
> My Winrash stayed IDLE when I had launched it several times.

There's two reasons why certain days have no results: either nothing has 
changed or the build process is broken.

The former can happen even when there is code being committed (hence WRT 
activity) but these changes don't affect any of the tests.  The build process 
checks this and if the current winetest.exe matches the previous one then it 
doesn't make the new winetest.exe available.  This increases the likelihood 
of having greater number of testers running with the one winetest.exe, and so 
increase the value of their work (more tests with which to compare).

Sadly the later happens all too often.  MinGW has a poor coverage of Windows 
API (well, poorer than Wine's tests anyway ;^) so when someone adds a new 
test, testing against some additional API, it often breaks the 
cross-compilation (in 2005 this happened 22 times).

Usually its pretty straightforward job to get MinGW working again (thanks to 
help from Stefan Leichter and Hans Leidekker) and rebuild, but it requires me 
to be near a computer with Internet access.  Usually not a problem, but 
occasionally it causes delays -- hence there are some times where there's no 
winetest.exe.

The problem above (at the beginning of this year) was with a missing GUID in 
the compiler, which is now fixed.

Cheers,

Paul.

> [2] http://test.winehq.org/data/
> [3] http://www.astro.gla.ac.uk/users/paulm/WRT/wrt.php
[4]  http://www.astro.gla.ac.uk/users/paulm/WRT/CrossBuilt/
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
Url : http://www.winehq.org/pipermail/wine-devel/attachments/20060107/7b8c8ece/attachment-0001.pgp