Description

Currently for every test and application the BOARD_INSUFFICIENT_MEMORY variable needs to be manually maintained. This makes adding new boards with low RAM/FLASH a nightmare. Also, boards will practically never be removed from BOARD_INSUFFCIENT_MEMORY, even though newer toolchains and improvements in code could result in lower RAM/ROM requirements.

Additionally, the BOARD_INSUFFICIENT_MEMORY approach reduces compilation test coverage. E.g. a test will not be compiled at all when blacklisted via BOARD_INSUFFCIENT_MEMORY, but only the linking stage will fail because of insufficient RAM/flash. Any possible issue that the compilation stage would uncover remain unrevealed.
Update: Only the linking step is skipped when blacklisted via BOARD_INSUFFICIENT_MEMORY

Brainstorming of Possible Alternatives

1. Let make fail with a custom exit code upon linking stage. The CI can handle treat this return code not as an error
   • Pros:
     • Relatively straight forward
   • Cons:
     • No obvious disadvantages
2. Add a special Make target that overrides link time checks
   • Pros:
     • Relatively straight forward
     • No changes in the CI required except for using that specific target
   • Cons:
     • A really ugly hack
     • A user might not read the doc, notice that the make target "ci_build" magically makes the error go away and will try to flash the result
3. Add something like RAM_PROVIDED and FLASH_PROVIDED to every MCU, let boards override those in case of bootloaders. Add RAM_REQUIRED and FLASH_REQUIRED to every test and example
   • Pros:
     • No obvious advantages
   • Cons:
     • A lot of effort to add those to every board
     • A maintenance burden to keep them up to date
     • RAM/flash requirements depend on the toolchain, the CPU/board specific code, the name of the git branch baked into the "hello message" upon boot, and the alignment of the stars