| Message ID | 20220212214532.23087-2-ludo@gnu.org |
|---|---|
| State | Accepted |
| Headers | show |
| Series | Build nhc98 and ghc@4 with regular dependencies | expand |
| Context | Check | Description |
|---|---|---|
| cbaines/comparison | success | View comparision |
| cbaines/git branch | success | View Git branch |
| cbaines/applying patch | success | View Laminar job |
| cbaines/issue | success | View issue |
| cbaines/comparison | success | View comparision |
| cbaines/git branch | success | View Git branch |
| cbaines/applying patch | success | View Laminar job |
| cbaines/issue | success | View issue |
Ludovic Courtès schreef op za 12-02-2022 om 22:45 [+0100]: > + (replace 'configure > + (lambda* (#:key outputs build configure-flags > + #:allow-other-keys) > + ;; It's an old 'configure' script so it needs some help. > + (let ((out (assoc-ref outputs "out"))) Input labels (well, ‘output labels’) can be removed here to a degree: (lambda* (#:key build configure-flags #:allow-other-keys) ;; [...] (define out #$output) (setenv "CONFIG_SHELL" [...]) [...]) Greetings, Maxime.
Ludovic Courtès schreef op za 12-02-2022 om 22:45 [+0100]: > + (apply invoke "./configure" > + (string-append "--prefix=" out) > + (string-append "--build=" build) > + (string-append "--host=" build) > + configure-flags)))) Seems incorrect when gcc is being cross-compiled, "--build" or "--host" (I always forget which) should be set to 'target' (in the Guix sense, not the autools sense). Or perhaps it doesn't matter since #:system is always "i686-linux" ... Greetings, Maxime.
Ludovic Courtès schreef op za 12-02-2022 om 22:45 [+0100]: > + (add-before 'configure 'remove-bundled-texinfo > + (lambda _ > + ;; Do not build the bundled Texinfo. > + (delete-file-recursively "texinfo") Unbundling seems more something for a source snippet to me, since it's ‘cleaning up’ the source code. ‘(guix)Snipppets versus Phases’ also says: The boundary between using an origin snippet versus a build phase to modify the sources of a package can be elusive. Origin snippets are typically used to remove unwanted files such as bundled libraries, [...] Greetings, Maxime.
Ludovic Courtès schreef op za 12-02-2022 om 22:45 [+0100]: > + (native-inputs (list texinfo)) > [...] > + ;; Go ahead despite the many warnings. > + (substitute* '("Makefile.in" "gcc/Makefile.in") > + (("^MAKEINFOFLAGS =.*") > + "MAKEINFOFLAGS = --force\n"))))))) FWIW, since gcc@2.95 is an ancient version of gcc only used for bootstrapping, personally I would not bother with building the documentation. OTOH, there have been quite some changes in gcc, and the documentation isn't getting in the way of anyone not interested in it ... Greetings, Maxime.
Ludovic Courtès schreef op za 12-02-2022 om 22:45 [+0100]: > + (native-search-paths > + (list (search-path-specification > + (variable "C_INCLUDE_PATH") > + (files '("include"))) > + (search-path-specification > + (variable "LIBRARY_PATH") > + (files '("lib"))))))) Perhaps a comment ;; version 2.95 of gcc@ only supports C, so no CPLUS_INCLUDE_PATH would be useful to answer the question of why the search paths are overridden? Greetings, Maxime
Ludovic Courtès schreef op za 12-02-2022 om 22:45 [+0100]:
> + #:tests? #f
Why are tests disabled here?
Greetings,
Maxime.
Maxime Devos <maximedevos@telenet.be> skribis: > Ludovic Courtès schreef op za 12-02-2022 om 22:45 [+0100]: >> + (apply invoke "./configure" >> + (string-append "--prefix=" out) >> + (string-append "--build=" build) >> + (string-append "--host=" build) >> + configure-flags)))) > > Seems incorrect when gcc is being cross-compiled, "--build" or "--host" > (I always forget which) should be set to 'target' (in the Guix sense, > not the autools sense). > > Or perhaps it doesn't matter since #:system is always "i686-linux" ... Right, it probably cannot be cross-compiled and that’s fine. (For some reason this old ‘configure’ script needed both ‘--build’ and ‘--host’.) Ludo’.
Maxime Devos <maximedevos@telenet.be> skribis: > Ludovic Courtès schreef op za 12-02-2022 om 22:45 [+0100]: >> + (add-before 'configure 'remove-bundled-texinfo >> + (lambda _ >> + ;; Do not build the bundled Texinfo. >> + (delete-file-recursively "texinfo") > > Unbundling seems more something for a source snippet to me, > since it's ‘cleaning up’ the source code. Right I’ll move it there. (The bundled texinfo was a surprise to me; it actually builds but seems to be broken in some way.) Ludo’.
Maxime Devos <maximedevos@telenet.be> skribis: > Ludovic Courtès schreef op za 12-02-2022 om 22:45 [+0100]: >> + (native-search-paths >> + (list (search-path-specification >> + (variable "C_INCLUDE_PATH") >> + (files '("include"))) >> + (search-path-specification >> + (variable "LIBRARY_PATH") >> + (files '("lib"))))))) > > Perhaps a comment > > ;; version 2.95 of gcc@ only supports C, so no CPLUS_INCLUDE_PATH > > would be useful to answer the question of why the search paths are > overridden? Good idea, will do!
Maxime Devos <maximedevos@telenet.be> skribis: > Ludovic Courtès schreef op za 12-02-2022 om 22:45 [+0100]: >> + #:tests? #f > > Why are tests disabled here? That’s the same for all the GCC variants, including ‘gcc-core-mesboot0’, which was the inspiration for this one. I think the main reason is that “it’s difficult” (there’s a dependency on DejaGNU, the fact that you cannot expect all tests to pass, and the fact that many tests would probably need to be adjusted.) In this case I didn’t even try, but since it’s an ancient package, it’s even more likely to be problematic, and since it’s only used for bootstrapping, I would not bother as long as it’s good enough to build ghc@4. Thoughts? Ludo’.
Ludovic Courtès <ludo@gnu.org> skribis: > Maxime Devos <maximedevos@telenet.be> skribis: > >> Ludovic Courtès schreef op za 12-02-2022 om 22:45 [+0100]: >>> + #:tests? #f >> >> Why are tests disabled here? > > That’s the same for all the GCC variants, including ‘gcc-core-mesboot0’, > which was the inspiration for this one. I think the main reason is that > “it’s difficult” (there’s a dependency on DejaGNU, the fact that you > cannot expect all tests to pass, and the fact that many tests would > probably need to be adjusted.) So I tried out of curiosity and was surprised to see that ‘make check’ succeeds! It’s not entirely clear how many tests were run though since DejaGNU isn’t very talkative: --8<---------------cut here---------------start------------->8--- Test run by nixbld on Mon Feb 14 11:21:04 2022 Native configuration is i686-unknown-linux-gnu Test run by nixbld on Mon Feb 14 11:21:04 2022 === gcc tests === Native configuration is i686-unknown-linux-gnu === g77 tests === Schedule of variations: WARNING: Couldn't find tool init file unix [...] === g++ Summary === make[1]: Leaving directory '/tmp/guix-build-gcc-2.95.3.drv-0/gcc-2.95.3/gcc' /gnu/store/7s2zlc4r9wqvlh9x2gqmzm4lnmbsdzmn-bash-minimal-5.1.8/bin/sh ./regress-demangle ./demangle-expected All 616 tests passed make[2]: Leaving directory '/tmp/guix-build-gcc-2.95.3.drv-0/gcc-2.95.3/libiberty/testsuite' make[1]: Leaving directory '/tmp/guix-build-gcc-2.95.3.drv-0/gcc-2.95.3/libiberty' --8<---------------cut here---------------end--------------->8--- Anyway, it can’t hurt. I also found that building in parallel is reliable enough. Ludo’.
diff --git a/gnu/packages/gcc.scm b/gnu/packages/gcc.scm index efa0baeaa1..2f947e4322 100644 --- a/gnu/packages/gcc.scm +++ b/gnu/packages/gcc.scm @@ -1,5 +1,5 @@ ;;; GNU Guix --- Functional package management for GNU -;;; Copyright © 2012, 2013, 2014, 2015, 2016, 2017, 2018, 2019, 2020, 2021 Ludovic Courtès <ludo@gnu.org> +;;; Copyright © 2012-2022 Ludovic Courtès <ludo@gnu.org> ;;; Copyright © 2014, 2015, 2018 Mark H Weaver <mhw@netris.org> ;;; Copyright © 2014, 2015, 2016, 2017, 2019, 2021 Ricardo Wurmus <rekado@elephly.net> ;;; Copyright © 2015 Andreas Enge <andreas@enge.fr> @@ -641,6 +641,75 @@ (define-public gcc-11 ;; the gcc-toolchain-* definitions. (define-public gcc gcc-10) + +;;; +;;; Historical version. +;;; + +(define-public gcc-2.95 + ;; Note: 'gcc-core-mesboot0' in commencement.scm provides 2.95 as well, but + ;; with additional tricks to support compilation with TinyCC and Mes-libc. + (package + (inherit gcc) + (version "2.95.3") + (source (origin + (method url-fetch) + (uri (string-append "mirror://gnu/gcc/gcc-2.95.3/gcc-core-" + version ".tar.gz")) + (sha256 + (base32 + "1xvfy4pqhrd5v2cv8lzf63iqg92k09g6z9n2ah6ndd4h17k1x0an")))) + (supported-systems '("i686-linux" "x86_64-linux")) + (native-inputs (list texinfo)) + (inputs '()) + (propagated-inputs '()) + (outputs '("out")) + (arguments + (list #:system "i686-linux" ;x86_64 didn't exist back then + #:tests? #f + #:parallel-build? #f + #:configure-flags #~'("--disable-werror") + + #:phases + #~(modify-phases %standard-phases + (add-before 'configure 'set-dynamic-linker-file-name + (lambda* (#:key inputs #:allow-other-keys) + ;; Tell GCC what the real loader file name is. + (substitute* "gcc/config/i386/linux.h" + (("/lib/ld-linux\\.so\\.[12]") + (search-input-file inputs "/lib/ld-linux.so.2"))))) + (replace 'configure + (lambda* (#:key outputs build configure-flags + #:allow-other-keys) + ;; It's an old 'configure' script so it needs some help. + (let ((out (assoc-ref outputs "out"))) + (setenv "CONFIG_SHELL" (which "sh")) + (apply invoke "./configure" + (string-append "--prefix=" out) + (string-append "--build=" build) + (string-append "--host=" build) + configure-flags)))) + (add-before 'configure 'remove-bundled-texinfo + (lambda _ + ;; Do not build the bundled Texinfo. + (delete-file-recursively "texinfo") + (substitute* "configure" + (("host_tools=(.*)texinfo" _ before) + (string-append "host_tools=" before))) + + ;; Go ahead despite the many warnings. + (substitute* '("Makefile.in" "gcc/Makefile.in") + (("^MAKEINFOFLAGS =.*") + "MAKEINFOFLAGS = --force\n"))))))) + (native-search-paths + (list (search-path-specification + (variable "C_INCLUDE_PATH") + (files '("include"))) + (search-path-specification + (variable "LIBRARY_PATH") + (files '("lib"))))))) + + (define-public (make-libstdc++ gcc) "Return a libstdc++ package based on GCC. The primary use case is when using compilers other than GCC."