OpenModelica users wanting to have rpms?

[Kevin Kofler]

Christoph Höger wrote:
> I have started converting MetaModelica to autootols to boot-bootstrap
> the omc build process.

May I suggest using CMake instead? It's easier to use, its new versions are 
more backwards-compatible with older ones, it's more portable to other 
(inferior ;-) ) operating systems, it doesn't require or expect you to ship 
generated files in source tarballs, it supports nice features like progress 
percentages, it's faster and it's successfully used in more and more 
upstream projects, including all of KDE.

On the other hand, CMake would probably be less than helpful for the SML 
parts, which comprise a significant portion of the codebase as far as I can 
see, you'd have to work with add_custom_command which isn't that wonderful. 
(For common languages like C/C++ and a few others, CMake does a lot of stuff 
for you, but less common ones aren't really supported and you end up having 
to write CMake commands equivalent to makefile rules.)

So each tool has its advantages and drawbacks.

> Building the rml compiler and the related libraries was easy, but now I
> could need some help and advice on how to build the testcases.
> In the original buildsystem from
> https://openmodelica.ida.liu.se/svn/MetaModelica there are some examples
> with own makefiles, but those refer to the buildsystem itself.
> I am not sure how to handle this with automake (obviously it would
> require to build the compiler before the tests).
> So currently I am wondering if the examples should have a build system
> that requires the compiler to be installed, any thoughts?

Normally testsuites can use the just-built compiler directly from the source 
tree. Look at existing projects and how they handle this. As you're using 
autotools, I guess GCC would be a good place to look.

> On the other hand, there are some "style" questions, I'd like to be
> answered:
> 
> This package builds three slightly different libraries in three differen
> flavors: called (librml_plain|librml_mask|librml_diff)(_g|_p|).so
> Those flavors only differ by the CFLAGS set upon compilation (_p means
> -p, _g -g).
> Upstream told me, they require them all, but would this be acceptable?

Sure, I don't see why not. You just need to be careful when building (you 
need to build the object files to different places so they don't conflict).

> Is the name rml ok for a library in /usr/lib or shall I
> use /usr/lib/rml/ by default? (Same for headers)

Hmmm, that's a bit at the limit, 3 letters are a bit short for a unique 
name. :-( But there's no librml.so in Fedora yet as far as repoquery tells 
me, so at least there's no current conflict. Let's see what others think.

> What with the name? Is MetaModelica even a good name, if the main binary
> is rmlc?

If that's the upstream project name (used in things like tarballs), it's 
fine. (But is the MixedCase really necessary? :-( Usually things like 
tarball and package names are all lowercase, but sometimes MixedCase is used 
by upstream and the Fedora packages usually match that. Probably something 
to discuss with upstream.)

> The package builds a compiler driver, essentially a shell script, by
> copying some configuration variables into a shell template (mainly how
> to invoke cc). Would this be fine as a /usr/bin script?

Yes, but beware of multilib conflicts: if that script is in the same package 
as some libraries, that package will end up multilibbed due to the libraries 
and if the script is not identical for 32-bit and 64-bit, there will be a 
conflict between the 2 multilibbed packages. (Splitting out the libraries 
into a -libs package is a way to work around that.)

        Kevin Kofler