Re: GFortran can’t find system headers

[Ludovic Courtès]

Nikita Karetnikov <address@hidden> skribis:

>> Instead, what should do is preserve the arguments; the value associated
>> with #:configure-flags should be changed to replace any
>> --enable-languages=.* flag with yours.  See ‘gcc-boot0’ in base.scm for
>> how to do that.
>
> Are you talking about the ‘substitute-keyword-arguments’ part?

Yes.

>> Alternately, you could turn the current ‘gcc-4.8’ definition into a
>> ‘make-gcc-4.8’ procedure like this:
>
>>   (define* (make-gcc-4.8 #:key languages)
>>     (package
>>        ...
>
>>        #:configure-flags ... ,(string-join languages ",")
>
>>        ...))
>
>>   (define gcc-4.8
>>     (make-gcc-4.8 #:languages '("c" "c++")))
>
>> That would probably be easier to work with.
>
> I don’t like the above version because it doesn’t allow to choose the
> version of GCC.

Just add a ‘gcc’ parameter.

> Also, ‘/nix/store/nxpzxlvg5z5qq10wzxrzk9hjyhxyicxq-gfortran-4.8.1/bin’
> contains these programs:
>
> c++     gcc-ranlib             i686-pc-linux-gnu-gcc-4.8.1
> cpp     gcov                   i686-pc-linux-gnu-gcc-ar
> g++     gfortran               i686-pc-linux-gnu-gcc-nm
> gcc     i686-pc-linux-gnu-c++  i686-pc-linux-gnu-gcc-ranlib
> gcc-ar  i686-pc-linux-gnu-g++  i686-pc-linux-gnu-gfortran
> gcc-nm  i686-pc-linux-gnu-gcc
>
> Are C++ and C-related programs supposed to be there?

Yes, the C and C++ front-ends are always compiled.

> -(define-public gcc-4.7
> +(define* (inherit-gcc name* version* source* #:key (languages '("c" "c++")))

I would really ‘make-gcc’, or maybe ‘custom-gcc’.

Good news: the stars in these parameter names are no longer needed
(commit 59fbeb8).

>    (let* ((stripped? #t)                           ; TODO: make this a 
> parameter
>           (maybe-target-tools
>            (lambda ()
> @@ -68,13 +68,14 @@ where the OS part is overloaded to denote a specific 
> ABI---into GCC
>              ;; contents of (maybe-target-tools).
>              (list 'quasiquote
>                    (append
> -                   '("--enable-plugin"
> -                     "--enable-languages=c,c++"
> +                   `("--enable-plugin"
> +                     ,(string-append "--enable-languages="
> +                                     (string-join languages ","))
>                       "--disable-multilib"
>  
> -                     "--with-local-prefix=/no-gcc-local-prefix"
> +                     "--with-local-prefix=/no-gcc-local-prefix")
>  
> -                     ,(let ((libc (assoc-ref %build-inputs "libc")))
> +                   '(,(let ((libc (assoc-ref %build-inputs "libc")))

I guess this line shouldn’t have changed.

> @@ -200,6 +195,17 @@ Go.  It also includes standard libraries for these 
> languages.")
>        (license gpl3+)
>        (home-page "http://gcc.gnu.org/";))))
>  
> +(define-public gcc-4.7
> +  (let ((version "4.7.3"))
> +    (inherit-gcc "gcc" version
> +                 (origin
> +                  (method url-fetch)
> +                  (uri (string-append "mirror://gnu/gcc/gcc-"
> +                                      version "/gcc-" version ".tar.bz2"))
> +                  (sha256
> +                   (base32
> +                    
> "1hx9h64ivarlzi4hxvq42as5m9vlr5cyzaaq4gzj4i619zmkfz1g"))))))

I guess this triggers a complete rebuild, right?  (That means it would
go in ‘core-updates’.)

Using the ‘substitute-keyword-arguments’ hack as in base.scm would allow
you to avoid that.

Also, the ‘origin’ thing should be factorized:

  (define (gcc-source version)
    (origin ...))

It’s good that you’re tackling this!  Then we can also build the
Objective-C front-end, and then start GNUstepping.

Thanks,
Ludo’.