gnu: icedtea-8: Hardcode dynamically loaded libraries.

The previous patch failed with icedtea-6, which has uncatched
'decoding-error' in addition to 'encoding-error'.

And it seems that only icedtea-8 needs patch, since in icedtea-7
the USE_SYSTEM_GTK, USE_SYSTEM_FONTCONFIG, etc make it links with
those libraries instead of dlopen them.

So, now only patch icedtea-8:

iyzsong@member.fsf.org (宋文武) skribis:

> From 1d0dce6f0c40fa35162ecb441b53b32f77dd5b8d Mon Sep 17 00:00:00 2001
> From: =?UTF-8?q?=E5=AE=8B=E6=96=87=E6=AD=A6?= <iyzsong@gmail.com>
> Date: Thu, 8 Sep 2016 23:25:23 +0800
> Subject: [PATCH] gnu: icedtea-8: Hardcode dynamically loaded libraries.
>
> Fixes <https://bugs.gnu.org/24327>.
>
> * gnu/packages/java.scm (icedtea-8)[arguments]: Add 'patch-jni-libs' phase.

I’m a bit late, but thanks for fixing it!

> +                    (lambda (file)
> +                      (catch 'encoding-error
> +                        (lambda ()
> +                          (substitute* file
> +                            (("VERSIONED_JNI_LIB_NAME\\(\"(.*)\", \"(.*)\"\\)"
> +                              _ name version)
> +                             (format #f "\"~a\""  (find-library name)))
> +                            (("JNI_LIB_NAME\\(\"(.*)\"\\)" _ name)
> +                             (format #f "\"~a\"" (find-library name)))))
> +                        (lambda _
> +                          ;; Those are safe to skip.
> +                          (format (current-error-port)
> +                                  "warning: failed to substitute: ~a~%"
> +                                  file))))

What often works in such cases is to force ISO-8859-1 encoding
(“Latin-1”), which is a “catch-all” encoding (it’s an 8-bit encoding
that covers the 256 values):

  (with-fluids ((%default-port-encoding "ISO-8859-1"))
    (substitute* file-in-arbitrary-ascii-compatible-encoding
      …))

Ludo’.

ludo@gnu.org (Ludovic Courtès) writes:

> [...]
>
>> +                    (lambda (file)
>> +                      (catch 'encoding-error
>> +                        (lambda ()
>> +                          (substitute* file
>> +                            (("VERSIONED_JNI_LIB_NAME\\(\"(.*)\", \"(.*)\"\\)"
>> +                              _ name version)
>> +                             (format #f "\"~a\""  (find-library name)))
>> +                            (("JNI_LIB_NAME\\(\"(.*)\"\\)" _ name)
>> +                             (format #f "\"~a\"" (find-library name)))))
>> +                        (lambda _
>> +                          ;; Those are safe to skip.
>> +                          (format (current-error-port)
>> +                                  "warning: failed to substitute: ~a~%"
>> +                                  file))))
>
> What often works in such cases is to force ISO-8859-1 encoding
> (“Latin-1”), which is a “catch-all” encoding (it’s an 8-bit encoding
> that covers the 256 values):
>
>   (with-fluids ((%default-port-encoding "ISO-8859-1"))
>     (substitute* file-in-arbitrary-ascii-compatible-encoding
>       …))
>

Yeah, I tried that, but it was also producing 'encoding-error' in the
builder, so I gave up it.

It seems that's because the locale is "C" when calling `substitute*',
and the files have UTF-8 copyright sign (©).  But out of the builder,
the `substitute*' works fine even with '(setlocale LC_ALL "C")'.

Here is an example:
--8<---------------cut here---------------start------------->8---
(use-modules (guix packages)
             (guix build-system gnu))

(package
  (name "test")
  (version "0")
  (source #f)
  (build-system gnu-build-system)
  (arguments
   '(#:phases
     (modify-phases %standard-phases
       (replace 'unpack
         (lambda _
           (setlocale LC_ALL "en_US.utf8")
           (call-with-output-file "test"
             (lambda (port)
               (display "©" port)))

           (setlocale LC_ALL "C")
           (with-fluids ((%default-port-encoding #f))
             (substitute* "test"
               (("t") ""))))))))
  (home-page #f)
  (synopsis #f)
  (description #f)
  (license #f))
--8<---------------cut here---------------end--------------->8---

iyzsong@member.fsf.org (宋文武) skribis:

> ludo@gnu.org (Ludovic Courtès) writes:

[...]

>> What often works in such cases is to force ISO-8859-1 encoding
>> (“Latin-1”), which is a “catch-all” encoding (it’s an 8-bit encoding
>> that covers the 256 values):
>>
>>   (with-fluids ((%default-port-encoding "ISO-8859-1"))
>>     (substitute* file-in-arbitrary-ascii-compatible-encoding
>>       …))
>>
>
> Yeah, I tried that, but it was also producing 'encoding-error' in the
> builder, so I gave up it.
>
> It seems that's because the locale is "C" when calling `substitute*',

Weird; the ‘install-locale’ phase normally installs a UTF-8 locale, and
the locale shouldn’t matter once we’re overridden
‘%default-port-encoding’.

> and the files have UTF-8 copyright sign (©).  But out of the builder,
> the `substitute*' works fine even with '(setlocale LC_ALL "C")'.

Hmm, OK.  Subtle things going on.  :-)

Thanks,
Ludo’.