[Scilab-Dev] Packaging issues

[michael.baudin at contrib.scilab.org]

Hi,

I just released a new version of a module this week-end [1], and it was 
(wrongly) packaged this morning. Actually, I noticed two problems with 
this packaging, which are making the process a pain : I think they are 
worth to share.

* The module contains C source code. But the packaging was (wrongly) 
done as if it was a macro-based module. This is a human error which can 
happen because the developper has no way to give the information to the 
packager.

What we may need is a button "My module contains a C source" on the 
atoms web interface. This way, the packager knows what exactly must be 
done with the sources of the module (i.e. compile the code on several 
machines).

* Many tests fail (as indicated in the distfun_0.8_test.txt logfile). 
The typical log looks like :

[...]
    001/160 - [SCI/contrib/distfun/0.8] betacdf..................failed: 
premature end of the test script
    002/160 - [SCI/contrib/distfun/0.8] betainv..................passed
[...]

I have the same behavior on my machine. Notice that if you run the tests 
twice, you will not get the same failed/passed status. What happens is 
that, sometimes (looks like random), some module (say apifun, for 
example) is not loaded. But the betacdf uses apifun, and this makes the 
test fail. This happens if the module is loaded either manuall after an 
exec statement in the .scilab script, or automatically, after an 
atomsInstall statement.

I think that there might be an issue with the way the modules are loaded 
at startup. Perhaps some function does not wait until all the packages 
are started up. I have never been able to see this precisely. But this 
makes the testing process really horrible when it involves dependent 
modules.

Regards,

Michaël

PS
[1] http://atoms.scilab.org/toolboxes/distfun/0.8