Sam James <s...@gentoo.org> writes:
> From: Eli Schwartz <eschwa...@gentoo.org>
>
> The existing check makes an intimidating value proposition: that all
> software being checked was installed using distutils-r1.eclass, hence
> moving the check from there as-is to a new home is sufficient. This
> includes the use of functions specific to the distutils-r1 eclass
> inheritance chain. In particular, get_modname is part of
> multilib.eclass, which distutils-r1 inherits, but python-single-r1 does
> not inherit.
>
> This results in the following QA warning:
>
> ```
> /var/db/repos/gentoo/metadata/install-qa-check.d/60python-site: line 53:
> get_modname: command not found
> /var/db/repos/gentoo/metadata/install-qa-check.d/60python-site: line 53:
> get_modname: command not found
> /var/db/repos/gentoo/metadata/install-qa-check.d/60python-site: line 53:
> get_modname: command not found
> * Verifying compiled files for python3.12
> *
> * QA Notice: Extensions found compiled for the wrong Python version
> * (likely broken build isolation):
> *
> *
> /usr/lib/python3.12/site-packages/__pycache__/init_calibre.cpython-312.pyc
> *
> /usr/lib/python3.12/site-packages/__pycache__/init_calibre.cpython-312.opt-1.pyc
> *
> /usr/lib/python3.12/site-packages/__pycache__/init_calibre.cpython-312.opt-2.pyc
> ```
>
> because we are matching all files matching "*.cpython*" that are also
> named "*" instead of all files named "*$(get_modname)".
>
> Instead of using multilib.eclass, go directly to the preferred canonical
> source, and query cpython what *it* thinks a module extension should be.
> This is also more flexible since cpython itself doesn't really guarantee
> that extension modules are named anything like get_modname, but
> generally equals -- for backwards compatibility -- the final value from
> `_PyImport_DynLoadFiletab` / `_imp.extension_suffixes()` (and on
> Windows, that is .pyd instead of .dll, though admittedly,
> sysconfig.get_config_vars is pretty empty there; on the other hand, PyPy
> doesn't recognize unadorned .so because it doesn't need the
> compatibility, so we see that it's not really a guarantee, and might as
> well go for the sysconfig variable which is unambiguous where present).
>
> Signed-off-by: Eli Schwartz <eschwa...@gentoo.org>
> Signed-off-by: Sam James <s...@gentoo.org>
> ---
> This fixes the case where we install some .so without distutils-r1
> because of multilib.eclass not being inherited. It's also a bit
> faster hopefully (cost of Python startup once vs repeated subshells).
>
> I'll commit this later today if no objections.
I didn't mean to push this yet but forgot I had it staged, so it's in
now. I don't see much point in reverting it though.
>
> metadata/install-qa-check.d/60python-site | 13 ++++++++++---
> 1 file changed, 10 insertions(+), 3 deletions(-)
>
> diff --git a/metadata/install-qa-check.d/60python-site
> b/metadata/install-qa-check.d/60python-site
> index 3791fa80ecddb..49d4b3a1a8303 100644
> --- a/metadata/install-qa-check.d/60python-site
> +++ b/metadata/install-qa-check.d/60python-site
> @@ -75,6 +75,13 @@ python_site_check() {
> local sitedir=( "${pydir}"/site-packages )
> [[ -d ${sitedir} ]] || continue
>
> + local modname=$(
> + "${impl}" - <<-EOF
> + import sysconfig
> + print(sysconfig.get_config_var("SHLIB_SUFFIX"))
> + EOF
> + )
> +
> # check for bad package versions
> while IFS= read -d $'\0' -r f; do
> bad_versions+=( "${f#${ED}}" )
> @@ -107,7 +114,7 @@ python_site_check() {
> -name '*.pth' -o \
> -name '*.py' -o \
> -name '*.pyi' -o \
> - -name "*$(get_modname)" -o \
> + -name "*${modname}" -o \
> -name 'README.txt' \
> ')' -print0
> )
> @@ -130,8 +137,8 @@ python_site_check() {
> wrong_ext+=( "${f#${ED}}" )
> done < <(
> find "${sitedir}" '(' \
> - -name "*.pypy*$(get_modname)" -o \
> - -name "*.cpython*$(get_modname)" \
> + -name "*.pypy*${modname}" -o \
> + -name "*.cpython*${modname}" \
> ')' -a '!' -name "*${ext_suffix}" -print0
> )
>
>
> base-commit: 8a7efac6a20aafc403fcf36c88156ac6b388fb9b
