Move pthreads to ExternalLibraries

[Erik Schnetter]

We should move pthreads support to ExternalLibraries (or to the flesh).

I have asked CCT to create the respective repository.

Keyword:

[Roland Haas]

I have a version that does this here: https://github.com/rhaas80/Pthreads

[Frank Löffler]

Most (all) of the logic of this thorn is its detect.sh. What about creating a patch for doing this in the flesh? I think that would probably be better than a separate thorn, or is there a reason why a thorn would be better?

[Roland Haas]

I put it in a thorn since every other external library is also in a thorn. Specifically I also have a thorn to link against Intel's trace analyzer. Currently the feature is provided by the flesh (via the extras directory in lib/make) using PTHREADS=yes which is the single last remaining user of those features. There is no code since we cannot build our own pthreads since they required kernel support and there seems to be no userland library that could be used in multiple OS in the case we encounter an system (workstation?) where the kernel provides thread support but the userland libraries are missing (like PAPI).

[Frank Löffler]

Makes sense. There is already https://svn.cactuscode.org/projects/ExternalLibraries/pthreads. Can you use that?

[Roland Haas]

Yes, I will commit the changes from the github repo to the subversion on you pointed me to.

[Roland Haas]

Replying to [comment:5 rhaas]:

Yes, I will commit the changes from the github repo to the subversion on you pointed me to.

Done.

[Roland Haas]

Would it be ok to add this thorn to the ET (and to adjust it to use Frank's shell scripts once they are ready)?

[Erik Schnetter]

Yes, pleasae.

[Roland Haas]

Erik: I added the thorn to the manifest (current DISABLED) but have trouble running your distribute script to test them on a significant number of machines.

Would you mind updaing manifest.git and enabling PThreads then switching simfactory to remotes/origin/rhaas/pthreads then testing, please? The alternative is for me to just push the changes and wait for things to fail.

Unfortunately adding PThreads to the thornlist causes the build to abort as long as PTHREADS=yes is set.

[Erik Schnetter]

<bikeshedding>Shouldn't the name be Pthreads?</bikeshedding>

[Roland Haas]

This is not fully clear. The standards document http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/pthread.h.html#tag_13_35 does not mention any name at all for the module. I tend to find more Pthreads in a Google search than otherwise.

Mostly the repo is now called pthreads because CCT had already created one using all lowercase spelling (the one on GitHub was mixed case). When we move to bitbucket they will force lowercase repo names anyway. So this will become academic. Either one is fine with me.

Shedding my bike now (and misusing the bikeshedding definition) :-)

[Erik Schnetter]

All lower case is fine. I was just surprised by PThreads (two upper case letters), which is a spelling I haven't seen anywhere.

[Roland Haas]

Applied in git hash 7adbf95b36108a345a8c65b15feb035d9a43b6a1 of simfactory.