On 09/14/2018 12:10 AM, William A Rowe Jr wrote:
Although 1.6.5 has flown, please do follow up with your observations.

The last few days have been a tad busy and I did not mean to drop the
ball on this. With the OpenSSL 1.1.1 release I had to get that rolled
out and into some pre-prod systems as well as various other things
that pop up.

There aren't a lot of Sparc machines between the committers, and
I'd even thrown my x86-solaris VM's to archive, but I can restore
them if we can narrow down what you are seeing.

Well, I have not seen x86 Solaris in a while. I certainly have full
access and can spin that up neatly if needed. Trivial to do so in
fact. Merely time consuming.

It sounds from Yann's feedback that repeated tests are necessary
to decide if the pass or failure is a fluke.

Right and I agree. To that end I have build 1.6.5 on a few systems and
have seen very consistent results from the tests.  After a busy few
days I felt it would be best to isolate the build in a new server with
the Oracle 12.6 dev tools and try in my usual sequence :

1) use c99 and very strict CFLAGS

2) move on to cc and more relaxed CFLAGS

3) try with gcc 8.1.0

I am never surprised when code that builds clean and fine in the Linux
world gets torn to shreds by the brutal Oracle/Sun c99 compiler.

I usually just move on to step 2 but some code out there like OpenSSL
is wonderfully clean. Some of the obvious variables in play for a build
and test of $name_your_project is the presence or absence of modern GNU
toolchains and tools like automake and libtool. These things don't exist
in the dry wasteland of POSIX Solaris 10. Modern Perl doesn't either. I
have a full 64-bit stack and have had it running for years in production
finance situations. So test (1) above usually starts with a clean dry
bone empty Solaris 10 server and then I see what happens.

The real question should be "what is the value here?" and to that I can
only say "if it builds clean here then it builds anywhere."


Reply via email to