feat(kissfft): add LLAR formula - #174
Conversation
Convert the Conan Center kissfft recipe into an LLAR Formula covering
131.1.0 through 131.2.0. Keep the upstream pkg-config file, rewrite its
prefix to a relocatable ${pcfiledir} form, and verify the Conan
test_package consumer against the published metadata.
![fennoai[bot]](https://avatars.githubusercontent.com/in/1924301?s=60&v=4){kind=small}
There was a problem hiding this comment.
Review: add mborgerding/kissfft recipe
Solid new recipe that closely follows the established sibling pattern (bdwgc/libatomic_ops, google/crc32c): defaults + filter + cmake + relocatable pkg-config. Metadata (id, fromVer, versions.json, directory layout) is internally consistent, and the single prefix= rewrite is correct here because upstream emits libdir/includedir relative to ${prefix} — so the simpler one-line replace (vs libatomic's three) is appropriate, not an omission. The datatype allowlist, CMake option names, and .pc naming all match upstream 131.1.0.
Security, performance, and documentation-accuracy passes found no issues. Two code-quality notes are inline.
| fPIC := slices.contains(target.options["fPIC"], "ON") | ||
| openmp := slices.contains(target.options["openmp"], "ON") | ||
| useAlloca := slices.contains(target.options["use_alloca"], "ON") | ||
| datatype := target.options["datatype"][0] |
There was a problem hiding this comment.
Multi-value options can silently mis-build. filter iterates over all values per option and only rejects when some value is invalid, so it accepts targets that pass multiple values for one option (e.g. datatype: ["float", "double"] or shared: ["ON", "OFF"]). But onBuild collapses those inconsistently: the booleans (lines 56-59) use slices.contains(..., "ON") (ON wins), while datatype here uses [0] (first value wins, rest discarded). A target requesting two datatypes would pass filter, build only the first, and produce a package that doesn't match the requested options.
Simplest fix: reject multi-value options in filter, e.g. if values.len != 1 { return false } inside the loop, since each of these options is logically single-valued. That also removes the mixed slices.contains vs [0] idiom.
| consumer := filepath.join(testDir, "consumer.c") | ||
| os.writeFile(consumer, []byte(consumerSource), 0o644)! | ||
|
|
||
| pcName := "kissfft-" + target.options["datatype"][0] |
There was a problem hiding this comment.
Minor: the pcName derivation ("kissfft-" + datatype plus optional -openmp) is duplicated here from onBuild (lines 81-84), and the two copies read the options differently (onBuild uses the local datatype/openmp vars; here it re-reads target.options[...] inline). They can drift if the upstream naming scheme changes. Likely unavoidable across the build/test handler boundary, but worth keeping the two sites in lock-step.
1 participant
Convert the Conan Center kissfft recipe into an LLAR Formula.
mborgerding/kissfft131.1.0(CMakeLists exists from this tag;v130/v131are Makefile-only)prefixto${pcfiledir}/../..test_package.cCloses #72