[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: texinfo-6.8.90 pretest

From: Eli Zaretskii
Subject: Re: texinfo-6.8.90 pretest
Date: Sun, 23 Oct 2022 10:15:30 +0300

> From: Gavin Smith <gavinsmith0123@gmail.com>
> Date: Sat, 22 Oct 2022 21:36:20 +0100
> Cc: pertusus@free.fr, bug-texinfo@gnu.org
> > In any case, both methods produce just a number, not an encoding
> > identifier.  We will need to prepend something to it, like "CP" maybe,
> > depending on how the codeset is used by texi2any.
> Could you try the following?

This works, except that I used

> +      $locale_encoding = 'cp'.$CP;

(lower-case "cp") instead of "CP", since that's what
https://perldoc.perl.org/Encode seemed to imply.  texinfo.info is now
successfully built.

The next problem is all the non-ASCII file names in tp/tests/ and its
subdirectories: they cause failures like this one:

  make[2]: Entering directory `/d/gnu/texinfo-6.8.90/tp/tests'
  Making check in .
  make[3]: Entering directory `/d/gnu/texinfo-6.8.90/tp/tests'
  make  input_file_names_recoded_stamp.txt
  make[4]: Entering directory `/d/gnu/texinfo-6.8.90/tp/tests'
  make[4]: *** No rule to make target `input/included_lat?rn1.texi', needed by 
`input_file_names_recoded_stamp.txt'.  Stop.
  make[4]: Leaving directory `/d/gnu/texinfo-6.8.90/tp/tests'
  Makefile:3105: recipe for target `check-am' failed
  make[3]: *** [check-am] Error 2

That "?" there stands for a non-ASCII character in the file's name.
These file names are encoded in UTF-8 in the tarball, and are
referenced in UTF-8 encoding in the corresponding Makefile's.  This
causes trouble on Windows due to several issues, which are not really
predictable up front:

  . Windows doesn't natively support UTF-8 encoded file names.
  . Various programs used for building and testing Texinfo may or may
    not support UTF-8 file names on Windows.  The relevant programs

    - the 'tar' program used to unpack the tarball
    - the 'make' program used to run the tests
    - the Unixy shell used in the test scripts
    - various Coreutils programs used by the tests

    I use a native port of bsdtar (from libarchive) to unpack the
    tarball, which produces file names where UTF-8 byte sequences are
    literally present in the file system with each byte as a separate
    character (Windows attempts to interpret each byte as a character
    encoded in the system codepage).

    My Make and Bash are from MSYS, and they don't support UTF-8 in
    file names.  Likewise with Coreutils.

    People who use tools from Cygwin or MSYS2 (which is a fork of
    Cygwin) might have the UTF-8 file names supported, but only if
    they consistently use all of the tools from those packages, and
    run on a recent enough version of Windows (because Cygwin and
    MSYS2 dropped support for Windows before 7).

  . Finally, the texi2any itself: if it uses a native Windows port of
    Perl (as it does in my case), it won't support UTF-8 file names
    and neither will it reliably accept UTF-8 encoded command-line

So this is a mess on MS-Windows, and I really am at a loss how to
solve it reliably.  I don't want to lose the ability of running the
texi2any test suite with the native MS-Windows port, so perhaps some
way of skipping the files with non-ASCII file names could be added,
either automatically on MS-Windows or by some Make variable (as in
"make check NONASCII=no")?  I presume that those files test
capabilities that are not relevant for Windows anyway, since native
Windows programs cannot reliably accept UTF-8 encoded command-line


reply via email to

[Prev in Thread] Current Thread [Next in Thread]