Re: [Tinycc-devel] Allow configuration of tcc libraries search path

[Thomas Preud'homme]

Le vendredi 8 juillet 2011 18:42:01, Thomas Preud'homme a écrit :
[SNIP]
> 
> === Sketch for a (hopefully) better solution ===
> 
> One thing to remember though is that almost all files (headers, object,
> archive, but not ld.so) are included via
> tcc_add_file_internal(s1,filename,flags). So if tcc_add_file_internal was
> modified to check for dirname(filename)/<triplet> and filename it would
> solve the problem. Only one option of configure and one macro would then
> be enough: --with-multilib-subdir=triplet and CONFIG_TCC_MULTILIB_SUBDIR
> would do the job.
> 
> Of course we don't want to search all directories for file by trying with
> multilib subdir, only system directories. But fortunetely the difference is
> pretty easy to do. Everything from the command line is user specify and
> should be searched strictly, all the rest can be tried with multilib
> subdir. So all what would be needed is to split paths in two variables and
> change the functions using it, eg. library_paths would become
> system_library_paths and user_library_paths and tcc_add_library would then
> use both these dynarray differently. Same for other search paths. The
> distinction is even already done for includes.
tcc_add_file_internal which I mentionned above also need to be modified for 
system files which are linked, like crt*.o files. The default action would be 
to 
not search in multilib subdir but if flags & multilib it would be searched.

Also I forgot about ld.so which is not searched currently. However, instead of 
using blindly the tcc_interp variable whenever there is no LD_SO key in the 
environment, we could tcc_open both the multiarch subdir and the normal dir to 
detect where it is.

I'll write a patch implementing this solution this sunday and post it here for 
review before committing it, unless you tell me this design is crapy before 
then.
> 
> So, what do you think?
> 
> Remains that tcc use its own system search path which are different from
> the one used by references tools (cpp, ld and ld.so). But this is
> completely orthogonal to multilib support and could be done later since we
> lived with this since a long time.
> 
> > Thanks,
> > 
> > --- grischka
> 
> Best regards,
> 
> Thomas Preud'homme
Best regards,

Thomas Preud'homme

signature.asc
Description: This is a digitally signed message part.