compiling ruby on HP-UX / PA-RISC

[Alex Harvey]

Hi all,

Not sure if anyone out there would be using HP-UX on PA-RISC but just in case we've run into a real brick wall on this one.

We are apparently encountering exactly the same issue as this person did -
http://www.ruby-forum.com/topic/191987

We can get around the issue using the same workaround mentioned in the above thread but it's not really a maintainable solution.

Does anyone have any further ideas of what could be stopping us from compiling? 

We're trying to compile ruby 1.8.7p352 on HP-UX 11.11 PA-RISC.

Best regards,
Alex Harvey

[Michael Stahnke]

1. You may need to file a bug with ruby-lang.org
2. Would static linking help at all with being maintainable?


>
> --
> You received this message because you are subscribed to the Google Groups
> "Puppet Users" group.
> To view this discussion on the web visit
> https://groups.google.com/d/msg/puppet-users/-/_gb66_Nv3ukJ.
> To post to this group, send email to puppet...@googlegroups.com.
> To unsubscribe from this group, send email to
> puppet-users...@googlegroups.com.
> For more options, visit this group at
> http://groups.google.com/group/puppet-users?hl=en.

[Alex Harvey]

On Monday, November 5, 2012 4:27:54 PM UTC+11, Michael Stanhke wrote:

1.  You may need to file a bug with ruby-lang.org
2.  Would static linking help at all with being maintainable?

Thanks Michael.  Actually static linking should be fine - I didn't think of it.  I also raised a bug as you suggested, https://bugs.ruby-lang.org/issues/7279.  I'll see how it goes.

[Alex Harvey]

It turns out static linking didn't work either but for the benefit of people reading this in the archives I found a workaround -

There is some documentation of HP-UX PA-RISC compiler here http://h21007.www2.hp.com/portal/site/dspp/menuitem.863c3e4cbcdc3f3515b49c108973a801?ciid=4727276391695110VgnVCM100000275d6e10RCRD

(There's probably a better one somewhere but that's the one I used.)

The +b option is used by the linker to embed a library path list in the executable for use at run time. However, if passing these options via CC or GCC then the option should be -Wl,+b. The mkmf.log file showed, however, that an unknown option +b was being passed directly to gcc.  I found this was coming from configuration in /usr/local/lib/ruby/1.8/hppa2.0w-hpux11.11/rbconfig.rb.

After running configure I made the following change in config.status -

mv config.status config.status.orig
sed -e 's/^.*RPATHFLAG.*$/S["RPATHFLAG"]=" -Wl,+b%1$-s"/' config.status.orig >config.status
chmod +x config.status
./config.status

This resulted in /usr/local/lib/ruby/1.8/hppa2.0w-hpux11.11/rbconfig.rb having

# grep RPATHFLAG /usr/local/lib/ruby/1.8/hppa2.0w-hpux11.11/rbconfig.rb
CONFIG["RPATHFLAG"] = " -Wl,+b%1$-s"

which is what I wanted.

However, the make step still didn't run properly because -Wl,+b was now passed to ld, which is also wrong.

Thus after the make step finished I made another change -

cd ext/zlib
mv Makefile Makefile.orig
sed -e 's#^LIBPATH.*$#LIBPATH = -L. -L$(topdir) -L/usr/local/lib +b/usr/local/lib#' Makefile.orig >Makefile
make

That works fine.

Then cd ../.. && make install and the zlib extension was installed.

I also updated the redmine ticket https://bugs.ruby-lang.org/issues/7279