search

GEOS-ESM / GEOSana_GridComp

Repository containing code for data analysis for the GEOS Earth System Model data assimilation
Apache License 2.0
1 stars 3 forks source link

Fix for mksi build race condition #74

Closed mathomp4 closed 2 years ago

mathomp4 commented 2 years ago

So this is a fun one. In testing the ADAS a while back, @bena-nasa every so often encountered a what looked like a race condition in the build. You'd build it once, it'd fail; build again serially, it worked.

Then, last night, my nightly build of GEOSadas saw this:

/discover/nobackup/mathomp4/SystemTests/builds/ADAS/CURRENT/GEOSadas/src/Components/@GEOSana_GridComp/GEOSaana_GridComp/GSI_GridComp/mksi/make_ozinfo.F90(209): error #7013: This module file
was not generated by any release of this compiler.   [SATINFO_UTIL]
  use satinfo_util,only : assert_,warn,tell
------^

This was odd since there is only one satinfo_util module in the GEOSadas! After consulting with @tclune we found the issue. The make_pcscaling.x add_executable is asking for satinfo_util.F90 as a source:

ecbuild_add_executable(TARGET make_pcscaling.x SOURCES pcscaling_main.F90 m_pcscaling.F90 satinfo_util.F90)

and it is a source, so nothing wrong there! The problem is that if you juuuust happen to build building that module at the same time it's building, say, make_ozinfo.o which also uses satinfo_util then it could see an "incomplete" modulefile.

It turns out @tclune and @bena-nasa actually figured this out a while back (see https://github.com/GEOS-ESM/GEOSana_GridComp/pull/36/commits/e16759e77b0b1b973a4e502a8de514743e981e43) and created the "libmksi" library to handle it.

So, we just apply the same fix here and make make_pcscaling.x depend on libmksi to get satinfo_util.

mathomp4 commented 2 years ago

Note that I did test this in a build and it...built. I don't know how to run this executable, but since all that changed was how CMake tries to build it, I can't see that it wouldn't be zero-diff. So I'm labeling it as so.