Error creating Cray toolchain modulefiles with Lua syntax

[lucamar]

I report here as a placeholder for further discussion the issue with the swap function mentioned in #3575.
The error is raised at the sanity check when the Lua syntax is used to create the Cray toolchain modulefiles:

== 2021-02-18 10:41:00,170 build_log.py:169 ERROR EasyBuild crashed with an error (at easybuild/base/exceptions.py:124 in __init__): Module command '/usr/share/lmod/lmod/libexec/lmod python load cpeGNU/21.02' failed with exit code 1; stderr: Lmod has detected the following error: Unable to load module because of error when evaluating modulefile:
     /tmp/eb-l61bgz4f/tmpk9aiog8r/all/Core/cpeGNU/21.02.lua: [string "help([==[..."]:29: attempt to call a nil value (global
'swap')
     Please check the modulefile and especially if there is a the line number specified in the above message 
While processing the following module(s):
    Module fullname  Module Filename
    ---------------  ---------------
    cpeGNU/21.02     /tmp/eb-l61bgz4f/tmpk9aiog8r/all/Core/cpeGNU/21.02.lua

; stdout: _mlstatus = False
 (at easybuild/tools/modules.py:814 in run_module)

The framework scripts involved are easyblocks/generic/craytoolchain.py and tools/module_generator.py. The workaround is to list Cray external modules using module unlod/load commands in modluafooter instead of dependencies, to avoid the use of the function swap in the Lua modulefile created by EasyBuild for the Cray toolchains.
In the example below I make use of the custom easyblock cpeToolchain adapted from craytoolchain.py:

# Compiler toolchain for Cray EX Programming Environment GNU compiler (cpe-gnu)
easyblock = 'cpeToolchain'

name = 'cpeGNU'
version = "21.02"

homepage = 'https://pubs.cray.com'
description = """Toolchain using Cray compiler wrapper with cpe-gnu module (CPE release: %s).\n""" % version

toolchain = SYSTEM

dependencies = [
   ('cpe-gnu', EXTERNAL_MODULE)
]

modluafooter = '''
if not ( isloaded("cray-dsmml/0.1.3") ) then
    load("cray-dsmml/0.1.3")
end
if not ( isloaded("cray-libsci/20.12.1.2") ) then
    load("cray-libsci/20.12.1.2")
end
if not ( isloaded("cray-mpich/8.1.2") ) then
    load("cray-mpich/8.1.2")
end
if not ( isloaded("craype/2.7.5") ) then
    load("craype/2.7.5")
end
if not ( isloaded("craype-x86-rome") ) then
    load("craype-x86-rome")
end
if not ( isloaded("craype-network-ofi") ) then
    load("craype-network-ofi")
end
if not ( isloaded("gcc/10.2.0") ) then
    load("gcc/10.2.0")
end
if not ( isloaded("libfabric/1.11.0.0.233") ) then
    load("libfabric/1.11.0.0.233")
end
if not ( isloaded("perftools-base/21.02.0") ) then
    load("perftools-base/21.02.0")
end
if not ( isloaded("xpmem/2.2.35-7.0.1.0_1.9__gd50fabf.shasta") ) then
    load("xpmem/2.2.35-7.0.1.0_1.9__gd50fabf.shasta")
end
prepend_path("MODULEPATH",root:gsub("software","modules/all/Toolchain"))
'''
moduleclass = 'toolchain'

[boegel]

I took a quick look at this, and it's easy to reproduce with a simple test module:

$ find /tmp/$USER/testmods -type f
/tmp/testmods/swapit/1.0.lua
/tmp/testmods/foo/1.2.lua
/tmp/testmods/foo/1.0.lua

$ cat /tmp/testmods/swapit/1.0.lua
swap("foo", "foo/1.2")

This triggers the same error:

$ module list

Currently Loaded Modules:
  1) foo/1.0

$ module load swapit
Lmod has detected the following error:  Unable to load module because of error when evaluating modulefile:
     /tmp/testmods/swapit/1.0.lua: [string "swap("foo", "foo/1.2")..."]:1: attempt to call a nil value (global 'swap')
     Please check the modulefile and especially if there is a the line number specified in the above message
While processing the following module(s):
    Module fullname  Module Filename
    ---------------  ---------------
    swapit/1.0       /tmp/testmods/swapit/1.0.lua

So apparently the swap statement that EasyBuild has been generating since Feb'17 in Lua module files is simply not supported by Lmod at all, and we never noticed?!

@rtmclay Can you confirm that swap is not supported in Lua module files (while module swap in Tcl module files is)?

[rtmclay]

The module swap command in TCL modulefiles just does an unload() followed by a load().

[boegel]

OK, so this is fairly easy to fix: we just do an unload + load to emulate a swap operation when using Lua module files.

Fixed in #3685

[lucamar]

Thanks a lot @boegel!