Re: [Fwd: ORBit2 cross compiling.] - glib cross compiling

[Sven Neumann <sven gimp org>]

Hi,

Daniel Egger <degger fhm edu> writes:

> The issues here are that it's impossible for a build machine to
> figure out what requirements have to be met for the target by running
> compiled programs because the host cannot execute them in most cases.
> Thus to crosscompile the sources the programmer has to create a 
> config.cache which answers the open questions but this can be really
> a royal PITA. 

this might be a PITA but there's no sane way around using a
config.cache file for cross-compiling. If you can't reliably determine
the needed values at compile-time, you need to provide them in one way
or another. Putting this burden on the glib team is IMO the wrong
approach. Especially since cross-compiling environments usually
already have a config.cache file for the target platform.

> Sensible approaches IMHO are:
> - Reducing the amount of runtime tests to a minimum by replacing them
>   with compile tests. This is possible in quite a few cases and eases
>   some pain.

if the values can be determined reliably at compile-time, then this
approach is fine. You won't be able to do this for all values in
questions.

> - Carefully defining defaults which match most machines with a big
>   message that they may not suit for this particular machine when 
>   being used.

As I've said above, I don't think that the glib developers should have
to care about this.

> - Extending the configuration to predefine correct answers for popular
>   systems identified by triplets. Depending on how widespread some of
>   the tests are this might also go into a general macrofile which could
>   be used from other applications also. Disadvantage: bloat.

the same applies here, it shouldn't be the glib developers job to
create and maintain such defaults.


Salut, Sven