I haven't seen this one when I opened #2007 but it is worth comparing them. What #2007 does better:

• Use the last option from configopts when multiple are present
• Handles the case when swr-arches was passed but swr is not build. I expect this PR to fail in this case.
• Check added to sanity_check_* (where it actually belongs)

I don't think self.gallium_configopts is actually required. (Same actually for #2007 and self.swr_arches) as IMO setting the config params should be done in the config step to make it easier to understand.

@Flamefire you are right, this easyblock would have failed with a configopts such as "-Dgallium-drivers='swrast' -Dswr-arches=avx2". I've checked your PR and made the following changes:

• swr-arches configuration and sanity checks only considered if swr is in gallium-drivers
• function get_configopt_value() always takes last provided value and warns if multiple options were found
• added additional log messages

I tested these changes on Mesa-20.0.2-GCCcore-9.3.0.eb with the configration options listed in the description of #2006 (comment), plus the following ones:

• configopts += "-Dgallium-drivers='swrast' -Dswr-arches=avx"
  • SWR is not built and no sanity checks are added for SWR libs
• configopts += "-Dswr-arches=avx"
  • SWR will be built and SWR libs checked depending on the architecture
• configopts += "-Dgallium-drivers='swrast,swr' -Dswr-arches=avx -Dgallium-drivers='swrast,swr' -Dswr-arches=avx2"
  • only last option is taken into account

Regarding the use of self.gallium_configopts, it is true that all code sitting in __init__() could be moved to configure_step() and self.gallium_configopts would not be necessary. I do not really mind which option is used, both have their advantages and disadvantages. I'll let this decision to @boegel .

re-reviewed, tested and approved, thanks @lexming and @Flamefire!