GHCi and C libraries

12 Feb 2001 18:05:32 GMT

Mon, 12 Feb 2001 09:37:44 -0800, Simon Marlow <simonmar@microsoft.com> pisze:

> we shouldn't generate the HS*.o files for profiling (if we do,
> it's a bug.  I haven't checked).

We do. What is strange is that Interpreter.c does not compile for
profiling (grep '??' to see the reason), so you and other GHCi hackers
must have some configuration changes applied that you don't generate
profiled libraries at all.

> Julian is in the process of looking into the external libraries
> issue. The deal is that anything which can be dlopen()ed, or is a
> .o file, can be loaded by GHCi. That just excludes plain old static
> .a libraries, but most of these have shared versions anyway.

Let me repeat: what about .a libraries generated by ghc itself from
Haskell packages? Should every external package (qforeign, gtk+hs etc.)
link and install its compiled code twice only because GHCi does not
understand the .a archive format?

We can probably live with only .so support for C libraries, but if
.a would be supported anyway for Haskell packages, then surely it
would not make much difference to support it for native C libraries
as well, just in case a library somebody wants to use is only static
(as was the case with gmp some time ago)?

-- 
 __("<  Marcin Kowalczyk * qrczak@knm.org.pl http://qrczak.ids.net.pl/
 \__/
  ^^                      SYGNATURA ZASTĘPCZA
QRCZAK