Failure building on HP-UX with more than 2 threads (internal bug #4308)

[Mișu Moldovan]

On 06/09/18 15:06, Torbjörn Granlund wrote:
> Mișu Moldovan <misu.moldovan at proatria.com> writes:
> 
>    Just wanted to let you know I found a workaround for this particular
>    issue.  We are able to build GMP on HP-UX with more than 2 threads
>    using the HP-UX C++ compiler, when invoked as an ANSI C compiler, for
>    example with CC="/opt/aCC/bin/aCC -Ae".  Maybe worth documenting on
>    your side?
> 
> Are you saying that a parallel build fails with one compiler and works
> with another, and that the "make" command is somehow of less importance?

Both are important as on HP-UX with the proprietary compilers:

   * only with the HP's make (using -P) I was able to compile GMP / 
Python / etc. with more than one thread, and

   * only with the HP's C++ compiler (in C-compatible mode) I was able 
to build GMP with more than two threads.


> The only somewhat realistic way the compiler could cause parallel builds
> failures is if it uses hardwired temp filenames.  But I'd be extremely
> surprised if that were the case.

Hard to say as these are proprietary compilers, so our results are from 
a black box sort of testing.

However, we frequently got errors with both HP's compilers when building 
in parallel (more frequent for GMP, less for Python).  They look like 
race conditions.

So we switched back to using GNU's make to get consistent results 
(albeit with noticeably slower builds).  More on this at 
https://github.com/chevah/python-package/pull/107.

Thank you once more for GMP,

-- 
Mișu Moldovan - misu.moldovan at proatria.com
Pro:Atria Ltd - http://www.proatria.com
Tel: +44 (0) 1963 441311

The Old Exchange, South Cadbury, YEOVIL, BA22 7ET, UK
Registered in England - Company No: 4213930.