Dr. David Kirkby wrote:
Sorry, I realised the problem - we are discussing two different issues.
On Solaris 10 (SPARC), i.e. t2, this has never been an issue, as python
has built ok. There is no need to do anything fancy - it just works.
It's probably the case the OpenSSL libraries are *not* found, but the
haslib module builds, which is enough to pass the test in spkg-install.
I need to correct this slightly.
* I don't recall ever having an issue on 't2' with haslib failing, though I
will not say for 100% certainty I've never had an issue. I would search trac and
find it, but that is down.
* When I tried to build the Sage binaries, which I later copied to
http://t2nb.math.washington.edu:8000/home/davek/
I did get a failure with python. Then I added /usr/sfw/lib to LD_LIBRARY_PATH,
and python built ok. But I'm pretty sure I've built Sage on that machine without
having LD_LIBRARY_PATH include /usr/sfw/lib.
(One big difference between t2 and the machine I build the binaries on, is that
t2 is running almost the latest Solaris 10 release, whereas the machine I used
to build the binaries had the very first Solaris 10 release.
So I think it is fair to say that building python on Solaris 10 (SPARC) is a bit
hit-and-miss. Sometimes haslib builds, other times it does not. I can't quite
put my finger on what is the problem.
On OpenSolaris, python has been a real pain in the butt.
The only reason for not linking python against OpenSSL would be a license issue.
The GPL allows has this bit.
"However, as a special exception, the source code distributed need not include
anything that is normally distributed (in either source or binary form) with the
major components (compiler, kernel, and so on) of the operating system on which
the executable runs, unless that component itself accompanies the executable. "
I would have thought it was hard to argue OpenSSL is a major component of
Solaris - especially given it does not even ship with OpenSolaris. But this Sun
blog
http://blogs.sun.com/janp/entry/on_openssl_versions_in_solaris
explains that OpenSSL is central to Solaris. For this reason, Sun have not
updated beyond 0.9.7, but only apply security patches. The reason is that for
them to change the OpenSSL version, and properly test it, would take months of
work.
As such, I'm now of the opinion we don't have a problem linking against OpenSSL
on Solaris 10, as OpenSSL is a major component of the operating system.
The same can't be said for OpenSolaris, as OpenSolaris does not even ship with
OpenSSL.
Hence I believe in the long term, we do need to find another solution to
OpenSSL.
But in the short term, adding /usr/sfw/lib to python's setup.py will do no harm,
and might solve a few problems at build time.
I realise my arguments might not be too convincing, but the truth is, like you I
believe haslib should be built with python's code, without OpenSSL.
But in practice, having built Sage multiple times on various Solaris systems, I
can tell you that it is not the case, and the failure of python has been an
issue both on Solaris 10 and OpenSolaris.
However, on OpenSolaris, the issue is more severe. Both Jaap and William will
confirm that the test in python for hashlib has failed.
Dave
--
To post to this group, send an email to sage-devel@googlegroups.com
To unsubscribe from this group, send an email to
sage-devel+unsubscr...@googlegroups.com
For more options, visit this group at http://groups.google.com/group/sage-devel
URL: http://www.sagemath.org
