Re: template test with subdirs failure for HEAD on darwin [WAS support standalone libltdl [libtool--gary--1.0--patch-23]]

[Ralf Wildenhues]

* Gary V. Vaughan wrote on Sun, Aug 21, 2005 at 04:59:14PM CEST:
> Moin Ralf!

Good afternoon, Gary!

> On 21 Aug 2005, at 08:21, Ralf Wildenhues wrote:
> >* Gary V. Vaughan wrote on Fri, Aug 19, 2005 at 07:38:31PM CEST:
> >>The c++ template with subdir-objects autotest is failing for me right
> >>now (I think this is because I need to backport another patch to my
> >>local automake-1.9.6 installation -- any clue as to which one?),  

> >Erm, tests/template.at does not use Automake.
> 
> That'll teach me to look before I jump to conclusions.  D'oh!

 :)

> >Please run with -d -v, post testsuite output, to see what's happening.
> 
> Attached.  Looks like the compiler isn't getting all the switches it
> would like.  I don't know anything much about darwin compiler
> namespace stuff... I  treat my Macs like linux machines that work out
> of the box, with a sexy gui :-)

Can you get a demangled name of the symbol __Z2a2c?

I think this is a genuine failure with template libraries on darwin.
IOW: I don't think this ever worked.  Our test is doing the right thing
and exposes the failure.

I guess I'll leave that to our Darwin expert to fix, though.  I'd have
to probably read a lot in order to know how to do something along the
lines of:
- put template instantiations in shared C++ library
- use instantiations from deplibs if possible
- either use some working sort of weak symbols or one instantiation per
  object

We have prelink_cmds as new variable in CVS HEAD to do work before
linking a program, other outputs should be done in already-established
variables.  The linux/pgCC case serves as a good example.

Thanks,
Ralf