From: Vladimir Prus (ghost_at_[hidden])
Date: 2004-01-16 04:22:05
> Andre Hentz wrote:
> > The results are attached. I cleaned up the file a bit. If I can make
> > some time later, I'll investigate further the failures that are not
> > related to "attempted multiple inclusion of file", including tag.py,
> > which never finished.
Thanks for doing this!
> The first problem is that on solaris, one has to compile with -fPIC if
> the object file will ever be used in shared libraries. I added that in
> manually for the tests but some logic is required in gcc.jam.
Maybe, it's possible to use:
flags gcc.compile OPTIONS <link>shared : -fPIC ;
> After that we have:
> - searched_lib: "...attempted multiple inclusions...", just like other
Okay. That's the problem to be addressed first.
> - tag: the profile/static build never finishes. TestCmd.py is stuck at
> line 435:
It's probably deadlock caused by catching both stderr and stdout. Command
wrote too much to stderr and got blocked. After that, TestCmd.py is blocked
reading stdout -- but command is blocked on stderr so won't write anything to
I'll try to find a way to check if that's the cause of hang.
> - library_chain: FAILED:
> ld: warning: file a.so: required by b/bin/gcc/debug/b.so, not found
> Undefined first referenced
> symbol in file
> geek() b/bin/gcc/debug/b.so
I believe it's because on Linux, this requires -rpath-link to work. But on
Solaris, you've removed it. I think we need
> - conditionals: GNU ar version 2.11.2 creates invalid empty object.
> Should work with recent versions but I can't test. It works with Sun's ar.
Hmm... let's postpone this test until others are fine.
So, to summarize, we've got this problems:
- library order
- -L vs -rpath-link
- tag hangup
I hope we'll be able to solve -fPIC soon, and then I'll have to take care of
the first ones. Not yet sure when I'll come to that, though.
Boost-Build list run by bdawes at acm.org, david.abrahams at rcn.com, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk