Conference notes for 2022 05 03

Agenda for Lmod Zoom Mtg 2022-05-03

• Print message when unloading a module that wasn't loaded (Issue #576)
• How TCL break and LmodBreak is handled in Lmod
• Q&A

Attendees

• Robert McLay (TACC, US)
• Kenneth Hoste (HPC-UGent, Belgium)
• Kurt Lust (Univ. of Antwerp, Belgium - LUMI User Support Team)
• Matthew Cawood (TACC, US)
• Ward Poelmans (Vrije Universiteit Brussel, Belgium)
• Mikael Öhman (Chalmers University of Technology, Sweden)
• Bennet Fauber (Univ. of Michigan, US)

Notes

Print message when unloading module that wasn't loaded

• see https://github.com/TACC/Lmod/issues/576
• "warning" message will be printed, but exit code will remain zero
• was implemented on request by Spack developers
• not included in an Lmod release yet
• Kenneth will try to run EasyBuild test suite on top of Lmod that includes this change
• EasyBuild already sets $LMOD_QUIET, so won't be affected by this
  • and it probably doesn't unload modules anyway (especially none that are not loaded)

How Tcl break and LmodBreak is handled in Lmod

• slides available at https://github.com/TACC/Lmod/blob/master/my_docs/22/lmod_zoom_mtg_2022_05_03/presentation.pdf
• break statement outside of a loop in Tcl is an error (and we'll take advantage of that)
• when a break statement is hit, none of the changes specified by that module file get applied (and only that module will not be loaded)
• top of FrameStk is (in some sense) buffer of changes that are made by module file which is being loaded
  • if break happens, all those changes are ignored by kicking out the top of the stack (see M.LmodBreak function)

Q&A

• Robert has not found time yet to look into problem reported by Bennet via Lmod-users mailing list

  • see https://sourceforge.net/p/lmod/mailman/message/37648391/
  • Lmod is producing traceback 3/5 times?!
  • traceback are even not consistent
  • using Lmod 8.6.x on top of Lua 5.3 (RHEL8)
  • looping over hash tables in Lua 5.3 happens in random order of keys (way more than Lua 5.1 did), which could be relevant here...
  • problem is happening on bare module avail or module spider, also with module spider foo
  • Bennet should try the check_module_tree_syntax tool (module file linter)
  • traceback suggests towards loop being triggered
  • could be related to having a module file that is changing $MODULEPATH with location of module file being loaded

• Mikael's issue (https://github.com/TACC/Lmod/issues/575)

  • no answer yet on this...
  • looks like Lmod will need to have AI capabilities?

• LmodError vs LmodWarning

  • LmodError: totally cancels whole module command
  • LmodWarning prints a warning and does non-zero exit
  • LmodBreak: stops processing module being loaded, doesn't apply changes, cancels rest of module command

Next meeting

• Tue June 7th 2022 at 09:30 US Central (14:30 UTC)
• topic: explaining how module load is processed by Lmod, stepping through the codebase
• future topics: how module name is converted into an internal object in Lmod codebase