I wrote:
> Tatsuo Ishii <t-ishii@sra.co.jp> writes:
>> Here is a problem report from a HP-UX/11.0/cc user.
>> src/makefiles/Makefile.hpux seems to lack:
>> %.sl: %.o
>> $(LD) -b -o $@ $<
>> This makes regression test failed. Please note that in 6.4
>> Makefile.hpux has above.
> I removed that rule from Makefile.hpux because I believed it to be
> dead code --- src/Makefile.shlib is now responsible for causing
> shared libs to be built correctly for each platform.
> I have not seen any problems building on HPUX 9 or 10. Can you
> provide more info about the failure on HPUX 11?
Sigh. Tatsuo is quite right: I screwed up. The 6.4.2 fileset
*builds* OK on HPUX, but it will not run the regression test.
The regression test includes building a shared library (two of 'em
in fact) ... and those makefiles do not use Makefile.shlib.
They rely on the port makefile to supply a default build rule for
shared libraries. So that rule is not quite as unused as I thought.
I don't know how I missed this --- I must not have tried to run the
regression tests after that set of changes, figuring that it was
purely a build-time issue and if the system built then things were OK.
Mea maxima culpa.
For the moment I will re-insert the .sl rule into Makefile.hpux in
both current and REL6_4 trees. A better fix would be to make the
various makefiles in src/test and contrib use Makefile.shlib ...
but currently I'm thinking that where we really want to go is to
start using GNU libtool to control shared library construction,
so as to not have to maintain all this platform-specific build
knowledge ourselves. So I'd rather spend time on that.
regards, tom lane
