Improve Guix support #115
Improve Guix support #115
Conversation
With this change, ‘make check’ runs in 230s on x86_64-linux-gnu (4-core CPU) instead of 880s. * tests/speedup.scm <top level>: Exit with 77 when ‘FIBERS_EXPENSIVE_TESTS’ is unset.
With this change, tests run in 46s instead of 230s on my x86_64 laptop when using ‘guix build -f guix.scm’. * build-aux/guile.am (.scm.go): Set GUILE_AUTO_COMPILE=0. * env.in: Set ‘XDG_CACHE_HOME’ when it’s undefined. * guix.scm (guile-fibers)[arguments]: Remove.
This follows the guidelines in <https://guix.gnu.org/cookbook/en/html_node/Software-Development.html>. * .guix/modules/fibers-package.scm: New file, based on… * guix.scm: … this. Turn into a symlink to the above. * Makefile.am (EXTRA_DIST): Add both files.
This is clearer and less ambiguous: if the person has additional channels providing same-named packages, it will still refer to the canonical packages under (gnu packages …). * .guix/modules/fibers-package.scm (S): Remove. (guile-fibers)[native-inputs, inputs]: Directly refer to the right (gnu packages …) variables. [synopsis]: Remove outdated sentence.
* .guix/modules/fibers-package.scm (guile2.2-fibers) (guile-fibers/libevent): New variables.
* .guix/modules/fibers-package.scm (guile-fibers)[native-inputs]: Add Guile.
* .guix/manifest.scm: New file. * Makefile.am (EXTRA_DIST): Add it.
| # interpreted) so make sure they can be auto-compiled upon 'make | ||
| # check'. | ||
| if test "$XDG_CACHE_HOME" = ""; then | ||
| XDG_CACHE_HOME="${TMPDIR:-/tmp}/guile-fibers-cache" |
There was a problem hiding this comment.
This is terrible w.r.t. security.
If they are meant to be compiled, just compile them (with .scm->.go rules), or put them in a subdirectory of the build directory.
There was a problem hiding this comment.
Also, if you put them in a subdirectory of the build directory, you might as well do this unconditionally.
| '("i586-pc-gnu" | ||
| "aarch64-linux-gnu")))) | ||
|
|
||
| (concatenate-manifests (list native-builds cross-builds)) |
There was a problem hiding this comment.
Also, there probably exists a target-linux? predicate.
You could also unify 'native-builds' with cross-builds':
(map (lambda (system target) [package->manifest stuff]) '(("system1" . "target1") ...))
This is also rather i*86/x86_64 centric.
Also, I think this should be changed into a list of packages instead of a manifest, because the file collisions would make the manifest itself meaningless, whereas the individual builds would be meaningful.
There was a problem hiding this comment.
Yes, it's very x86-centric, but it's already an improvement over the GitHub action that does x86_64-only. (And for now it's only meant for use by contributors on their machines, which, let's face it, is x86_64.)
I think using manifests is more convenient and generally necessary since there's more info than mere packages can carry.
There was a problem hiding this comment.
IME, it's also nice that you can change the last line when testing things locally, for instance to cross-builds if that's all you want to test.
| @@ -53,6 +53,12 @@ | |||
| (lp (1+ i) (make-vector (- words 2) #f)) | |||
| x))) | |||
|
|
|||
| (unless (getenv "FIBERS_EXPENSIVE_TESTS") | |||
There was a problem hiding this comment.
#104 is relevant.
I consider this name misleading. Proposal: "FIBERS_SKIP_BENCHMARKS".
There was a problem hiding this comment.
Right, tests/speedup.scm is not a test per se, it's a benchmark, but I didn't mean to revisit that classification here. I chose a variable that's similar to what other projects do, such as Coreutils (it's called RUN_EXPENSIVE_TESTS there; maybe we can just stick to that name?).
Also, my goal was to skip tests/speedup.scm by default.
These patches improves Guix support to make it easier to test a variety of configurations.
With the last patch, runnin:
... will perform several native builds (x86_64-linux, i686-linux, Guile 3.0 and 2.2, with and without libevent) and cross-compilation builds (currently failing; see #111).