@Dennis: No problem. I'll be on vacation next week, so if it doesn't work out for tomorrow, let me know and I'll have someone else on the team check in here while I'm away.
@Dennis: Give a shout if you have the time and inclination to try and get the multi-locale distribution of Chapel built and available today. I'll be online until 4:30-5:00 PT.
@Dennis Possible errata in the "Why TIO" section: " use any of the contact options in the contact options listed below." Not sure if the repetition is intentional. Definitely not golfy :-P
@Brad Building now. If I understood your comments on the tryitonline repo correctly, all I should add to the wrapper are QT_HWPAR and the three GASNET_ variables, yes?
@Dennis: Got interrupted by a call there. Thanks! I'll kick the tires a bit and see what I find.
@Dennis: Hmm, not so good results on my first attempt. The linker complains: 'ld terminated with signal 25 [File size limit exceeded].' I don't think I've seen this before -- is the system set up to limit the size of binary files in some way?
I'm now getting mv errors that I think you may have seen once the other day, and which I've seen from time to time. They seem much more consistent this time around, though: mv: failed to set default file creation context to ‘unconfined_u:object_r:sandbox_file_t:s0’: Permission denied
I'm digging a bit to see where they're coming from (best guess: we do do moves from TMPDIR to the cwd IIRC).
That's nothing to worry about. mv tries to set the default context, but everything in the sandbox has a MCS range specific to that sandbox, which differs from the default.
@Dennis: It appears that that guess was correct. Our compiler's Makefiles do two mv commands which seem to correspond to the two mv errors that I'm getting:
Oh wait, I think I see the problem. I think the mv errors were a false alarm.
Unfortunately, it looks like another 'make install' issue (sigh). This is a new feature for us that hasn't been in an official release yet, so this is turning out to be the acid test that's identifying bugs. Sorry about that.
On the plus side, all the mods to your scripts look correct and to be working properly. We just need to fix a path in our make install step (it refers back to the build location and in our testing, we didn't clean up the build so the path was still valid).
@Dennis: Bottom line: Ball's back in our court, but I think you could commit these script changes if you haven't for once the bug is fixed.
Will do. Not sure if I'll be able to fix this before leaving for vacation this coming week. If not, I'll recruit someone else to take it over or else pause this effort until I'm back.
@Brad Building finished and I'm currently syncing to the arena servers. make installsays Detecting CHPL_HOME as /opt/chapel/build among other things, but that directory gets removed to save space. Is that a problem?
@Dennis: That does sound problematic, but I don't think anything w.r.t. that should've changed there since last time around. Let me give it a whirl and see.
It looks like that was not a problem.
I'll check the message afterwards.
We got further, but no joy. And in retrospect, maybe I was being naive.
I was thinking that, when run in oversubscribed mode, GASNet would not require things like sockets, which I think you said were intentionally disabled. But I think it was wrong (at least in our current build configuration).
Specifically, I'm getting a: socket() failed while creating a listener socket (13:Permission denied)
@Dennis: If you could confirm that this is expected (sockets should not be available to processes within a compute node), I'll research whether we have other options available to us, like communicating through shared memory.
(But probably not until after vacation -- or else I'll have someone else on the team look into it).
Yes, the sandbox disables all network access, including sockets. This is a problem for a number of languages, but I haven't found a (safe) way around it yet.
@Dennis: It does appear that there is a shared memory option and that I was just forgetting that it isn't used by default by GASNet or at least in our use of it. We'll look into what it would take to enable this for the TIO build and get back to you.
Thanks for all the help this week and sorry for the false start or two along the way.