[csw-maintainers] [GAR] RFD: Custom assertions for packages

[Dagobert Michelsen]

Hi Maciej,

Am 03.03.2010 um 12:22 schrieb Maciej (Matchek) Blizinski:
> There are certain packages which are susceptible for certain kind of
> problems.  For instance, we might make sure that isaexec is not used,
> or that a /opt/csw/bin/sparcv9/foo binary is not present, or that a
> /opt/csw/lib/libfoo.so library is present in a certain package, or
> that file foo is present in CSWfoo_devel and not CSWfoo.
>
> Such issues could be checked either during build, a part of GAR-driven
> build process, or later on by checkpkg.  Since the tests I'm thinking
> about are package-specific, it's not possible to write generic
> checkpkg rules for them.
>
> One kind of checks I can think of is about testing the prototype.  The
> GAR interface could look something like this:
>
> ASSERT_PROTOTYPE_CONTAINS_CSWfoo = $(bindir)/foo
> ASSERT_PROTOTYPE_ABSENT_CSWfoo = $(bindir)/sparcv9/foo
> ASSERT_PROTOTYPE_ABSENT_CSWfoo-devel = .*\.py
> ASSERT_PROTOTYPE_CONTAINS_CSWpy-foo = .*\.py
>
> The idea is that the algorithm/code is the same, you only change the
> parameters such as package and file names.
>
> I can see two general potential solutions for this kind of checks:
>
> - it gets checked by GAR right after the application of the PKGFILES_*
> rules (good because it catches errors early)
> - information gets packaged up in the package and then is later used
> by checkpkg (good because it's done together with all the other
> checks,  the flip side is that it requires an additional communication
> layer which is itself prone to errors and must be tested)
>
> Here are some questions:
> - Which solution looks better to you and why? How would you  
> implement it?
> - What kind of package-specific checks can you think of?

I prefer checking it in GAR after prototype creation. There should be  
assertions
available to all subsystems generated, similar to PROTOTYPE_MODIFIERS.  
For now
I have put the information at
   <http://sourceforge.net/apps/trac/gar/wiki/Assertions>
for further work.


Best regards

   -- Dago