[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: 04/05: gnu: autoconf: Support cross-build.
From: |
Ludovic Courtès |
Subject: |
Re: 04/05: gnu: autoconf: Support cross-build. |
Date: |
Wed, 22 Apr 2020 23:00:07 +0200 |
User-agent: |
Gnus/5.13 (Gnus v5.13) Emacs/26.3 (gnu/linux) |
Hullo!
Jan Nieuwenhuizen <address@hidden> skribis:
>>> + (arguments
>>> + `(#:tests? #f
>>
>> Nope. :-)
>
> Ahum, yes, well...this made me blush for a moment ("I should not rely on
> a review by Ludovic for catching such a development hack") but look at
> the three preceeding lines:
>
> ;; XXX: testsuite: 209 and 279 failed. The latter is an impurity. It
> ;; should use our own "cpp" instead of "/lib/cpp".
> - (arguments `(#:tests? #f))
>
> Tests are disabled currently (even on master!), so I guess that
> inserting this line break is fine :-)
Damn it, this is terrible! But not your fault, for sure. :-)
>>> + ,@(if (%current-target-system)
>>> + `(#:phases
>>> + (modify-phases %standard-phases
>>> + ;; Autoconf cannot be cross-built properly: it lacks the
>>> + ;; concept of <tool>-for-build. It even runs the host
>>> + ;; `autom4te' (a perl script) during build.
>>> + (add-after 'install 'fake-cross-build
>>> + (lambda* (#:key build inputs outputs #:allow-other-keys)
>>> + (let ((m4 (assoc-ref inputs "m4"))
>>> + (perl (assoc-ref inputs "perl"))
>>> + (out (assoc-ref outputs "out")))
>>> + (substitute* (find-files (string-append out "/bin"))
>>> + (("/gnu/store/[^/]*-m4-[^/]*") m4)
>>> + (("/gnu/store/[^/]*-perl-[^/]*") perl))
>>> + #t)))))
>>
>> Why is this needed? The ‘patch-shebangs’ phase normally takes the
>> inputs, not the native inputs, when changing shebangs.
>
> Because it's not only the shebangs...but
>
> Good question...especially because it teaches me about patch-shebangs!
> Also, I failed to determine exactly what went wrong. Without m4, perl
> in INPUTS, the shebangs are wrong (obviously).
>
> After adding m4,perl in INPUTS, the shebangs are indeed correct. I
> hadn't noticed that before, because look:
>
> $ head $(./pre-inst-env guix build --target=i586-pc-gnu
> autoconf)/bin/autoheader
> #!/gnu/store/ drz7805gcsrqkgr8v43r1f7zydlsxh05-perl-5.30.2/bin/perl
> # -*- Perl -*-
> # Generated from autoheader.in; do not edit by hand.
>
> eval 'case $# in 0) exec
> /gnu/store/8zvc5mvk0xm3ygrxsgpyy5ilxb5rzjry-perl-5.30.2/bin/perl -S "$0";; *)
> exec /gnu/store/8zvc5mvk0xm3ygrxsgpyy5ilxb5rzjry-perl-5.30.2/bin/perl -S "$0"
> "$@";; esac'
> if 0;
>
> shebangs correct, re-execute EVALs: wrong.
Oh, got it.
> From 3d776e0077d62000f20d23a1b42b32fef718a503 Mon Sep 17 00:00:00 2001
> From: "Jan (janneke) Nieuwenhuizen" <address@hidden>
> Date: Sat, 18 Apr 2020 19:49:54 +0200
> Subject: [PATCH] gnu: autoconf: Support cross-build.
>
> Autoconf cannot be cross-built properly: it lacks the concept of
> <tool>-for-build. It runs the host `autom4te' (a perl script) during build.
>
> * gnu/packages/autotools.scm (autoconf)[inputs]: When cross-building, add perl
> and m4.
> [native-inputs]: when cross-building, use -for-build names.
> [arguments]: When cross-building, add `fake-cross-build' phase to substitute
> m4 and perl.
> ---
> gnu/packages/autotools.scm | 28 +++++++++++++++++++++++++++-
> 1 file changed, 27 insertions(+), 1 deletion(-)
>
> diff --git a/gnu/packages/autotools.scm b/gnu/packages/autotools.scm
> index 99ca52730e..fa3ced182f 100644
> --- a/gnu/packages/autotools.scm
> +++ b/gnu/packages/autotools.scm
> @@ -55,12 +55,38 @@
> (base32
> "113nlmidxy9kjr45kg9x3ngar4951mvag1js2a3j8nxcz34wxsv4"))))
> (build-system gnu-build-system)
> + (inputs
> + ;; TODO: remove `if' in the next rebuild cycle.
> + (if (%current-target-system)
> + `(("perl" ,perl)
> + ("m4" ,m4))
> + '()))
> (native-inputs
> `(("perl" ,perl)
> ("m4" ,m4)))
> ;; XXX: testsuite: 209 and 279 failed. The latter is an impurity. It
> ;; should use our own "cpp" instead of "/lib/cpp".
> - (arguments `(#:tests? #f))
> + (arguments
> + `(#:tests? #f
I’m really nitpicking, but could you move the XXX comment right after
#:tests? #f?
> + ,@(if (%current-target-system)
> + `(#:phases
> + (modify-phases %standard-phases
> + ;; `patch-shebangs' patches shebangs only, and the Perl
> + ;; scripts use a re-exec feature that references the build
> + ;; hosts' perl. Also, M4 store references hide in the
> + ;; scripts.
> + (add-after 'install 'patch-non-shebang-references
> + (lambda* (#:key build inputs outputs #:allow-other-keys)
> + (let ((m4 (assoc-ref inputs "m4"))
> + (perl (assoc-ref inputs "perl"))
> + (out (assoc-ref outputs "out"))
> + (store-directory (%store-directory)))
> + (substitute* (find-files (string-append out "/bin"))
> + (((string-append store-directory "/[^/]*-m4-[^/]*"))
> m4)
> + (((string-append store-directory
> "/[^/]*-perl-[^/]*"))
> + perl))
I’m very much in favor of keeping regexps literal. What about writing
them as:
(substitute* …
(("/[[:graph:]]/bin/m4")
(string-append m4 "/bin/m4"))
…)
?
Also please move the comment right below ‘lambda*’. :-)
And then I think you can push it to ‘core-updates’.
Thank you!
Ludo’.