treewide: reduce usage of shell in Makefiles

[basilfx]

Contribution description

Using $(shell ..) in Makefiles spawns a subprocess, which is less performant than built-in alternatives. Recalling #13174 from before, I let Claude analyze where this may improve on current master.

An overview:

• $(shell basename $(CURDIR)) works the same as $(notdir $(CURDIR)) (unless $(CURDIR) equals /)
• uppercase_and_underscore can be used instead of tr a-z- A-Z_
• Use memoization for flags re-evaluated multiple times

The impact will be unnoticeable for a single build and will probably be in the noise when actual compilation takes the overhand. But (hopefully) it will be more noticeable for CI builds with warmer caches. I don't expect miracles, and it started as an experiment.

To give an idea of the improvement (I let Claude do the benchmarking) on my server (i5 14600K with 64 GiB of memory):

For make -j1 (rebuild, so warm cache):

App / board	base	head	saving	execve base → head
hello-world / nucleo-f767zi	2.30 s	0.66 s	−71%	651 → 207 (−68%)
gnrc networking / nucleo-f767zi	3.23 s	1.10 s	−66%	807 → 311 (−61%)
hello-world / samr21-xpro	1.19 s	0.65 s	−46%	314 → 209 (−33%)
hello-world / native	0.54 s	0.41 s	−23%	197 → 174 (−12%)

For make -j8 (rebuild, so warm cache):

App / board	base	head	saving
hello-world / nucleo-f767zi	2.50 s	1.72 s	−31%
gnrc networking / nucleo-f767zi	4.83 s	3.68 s	−24%
hello-world / samr21-xpro	1.52 s	1.26 s	−17%
hello-world / native	0.52 s	0.56 s	~0 (noise)

Testing procedure

This should not impact the builds. The end result should be the same.

To count subprocessed, one can use for example strace -f -qq -e trace=execve -o trace.txt make BOARD=nucleo-f767zi all. Then count the number of execve in the output file.

Issues/PRs references

None

Declaration of AI-Tools / LLMs usage:

AI-Tools / LLMs that were used are:

• Claude Code Opus 5 for coming up with alternatives (especially strings.mk) and performing benchmarks. I have read an understand every line changed, but I am not a Makefile export, and it can probably be better.

[riot-ci]

Murdock results

✔️ PASSED

ccc1ea2 tests/build_system/utils: add tests

Success	Failures	Total	Runtime
11285	0	11286	15m:38s

Artifacts

• Documentation preview