BUG: devel/gmp should build dylib; breaks lang/ghc and all dependents

[scooter.phd@…]

devel/gmp no longer builds a dylib, breaking lang/ghc and all port depending on lang/ghc, such as darcs.

[danielluke (Daniel J. Luke)]

This also breaks clamav.

[mww@…]

my fault - adding "--enable-shared --enable-cxx" to configure.args fixes this problem; unfortunately the cvs server is not working for me - if someone would commit this change together with a revision inc. asap, that would be cool. Thanks!

[danielluke (Daniel J. Luke)]

configure.env ABI=32 might also be a good idea.

(If not specified, a 64 bit version will be built on a G5 machine, while the old port version was 32 bit)

[mww@…]

o.k. - commited changes to configure.env & .args

[gwright@…]

I removed the --enable-cxx configuration flag. It generated a conflict with the gmp-cxx-wrappers port. The gmp-cxx-wrappers port was was introduced because gmp's c++ build failed for some OS/compiler versions. (And no other port depended on the c++ support, it was really a convenience for people developing their own apps.) It can probably be retired eventually, but we should verify that c+ support work on Panther and mac/intel before we do.

-Greg

[mww@…]

so what about just adding --enable-cxx and then deleting the c++ files in post-destroot?

[gwright@…]

Hi Markus,

Maybe we should do that for now. I guess it's a configuration bug that don't get the shared libs without specifying --enable-cxx. I'll make the change and send a note to the person who originally requested the gmp-cxx-wrappers port. He seemed to know about some of the subtle problems with the c++ support, especially on Intel.

-Greg

[gwright@…]

I put back the --enable-cxx arg and deleted the c++ libraries so they don't conflict with the gmp-cxx-wrappers port.

I will try to find out if the gmp-cxx-wrappers port can be retired.

-Greg

[scooter.phd@…]

I'm not sure --enabled-shared does what it's really supposed to do. Under CA law, it looks like configure and libtool are guilty of deceptive advertising.

I've done the manual port process (port -v fetch gmp; port -v extract gmp; port -v patch gmp; port -v configure gmp; port -v build gmp). I've even cd-ed into the work dir and manually configured and gmake'd. No dylib is ever created or installed.

I've got the latest "port -d selfupdate" that includes the --enable-shared option to configure (heck, I even used that in the manual configure process too).

It occurs to me that I might need to update the auto* tools, but if there were any dependencies, of course, port would have updated that too... :-)

[gwright@…]

Hi Scott,

Yes, it looks like configure is fooling around. The ABI= environment variable is also not consistent on ppc and intel. ABI=32 is not recognized; the configure script wants ABI=long or ABI=longlong.

Ugh.

Some patching is in order, it seems.

Best Wishes, Greg

[gwright@…]

I have updated the portfile to not use the ABI= environment variable on intel processors. According to my reading of the configuration file, there is only one ABI available on OS X/intel.

If this fixes the build on intel I will close this bug. Could someone with an intel mac gice it a try? Thanks.

-Greg

[pguyot (Paul Guyot)]

It's been nearly two months. Closing.