Hi
I've received a request to package a version of scotch with 64bit
integers (as opposed to 32bit). I suppose the details are less
important, the bottom line is
scotch 32bit: typedef int32_t SCOTCH_Num;
scotch 64bit: typedef int64_t SCOTCH_Num;
where SCOTCH_Num affects the public ABI and is used by third parties
which use scotch.
Upstream allows selecting the integer size at compile-time (i.e. passing
-DINTSIZE64 for int64_t). However, this choice has no effect on the
library name, so vanilla upstream will build a library named
libscotch.so regardless of how you configure it.
I'm skeptical whether introducing a downstream scotch64 package with
i.e. libscotch64.so is a good solution, given that possibly no build
system of third-party software using scotch knows about libscotch64.so
and would need to be carefully patched (i.e. to not mix parts using
libscotch and those using libscotch64). Also, introducing downstream
specific suffixes is never a good idea.
The alternative would be to just switch the main scotch package to 64bit
integers, but this may be undesirable for memory-bound applications
which rely on the smaller memory-usage of 32bit integers.
I'm not really sure whether there is a good solution, happy to hear
opinions.
Thanks
Sandro
_______________________________________________
devel mailing list -- devel@lists.fedoraproject.org
To unsubscribe send an email to devel-le...@lists.fedoraproject.org
