Re: coreutils-7.1.81-9b653 test results

[Eric Blake]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

[oops, I meant to include bug-gnulib on my original send of this message,
in case Bruno has more ideas of what to try]

Matthew Woehlke <mw_triad <at> users.sourceforge.net> writes:

> >>> ==== sparc/Solaris ====

Which version of Solaris?

> ...okay, trying again after some Makefile hacking:

'make clean' followed by 'make CFLAGS=-g' eliminates the need for Makefile
hacking.  Or use ./configure CFLAGS=-g from the get-go.

> 
> #0  0xff31a480 in _libc_kill () from /usr/lib/libc.so.1
> #1  0xff2b95c0 in abort () from /usr/lib/libc.so.1
> #2  0x00011984 in compare (a=0x21348 "«»", la=4, b=0x31ed0 "`'", lb=2) 
> at test-quotearg.c:182
> #3  0x00011ad4 in compare_strings (func=0x11c9c <use_quotearg_buffer>, 
> results=0x31db0) at test-quotearg.c:196
> #4  0x000123c8 in main (argc=1, argv=0xffbef32c) at test-quotearg.c:309
> 
> Need more?

That helps.  I've made some progress, but still don't fully understand the
failure.  The failure happens on the very first attempt to translate a
string containing ` and ', with a message catalog installed that would
translate it to «», and the failure means that the translation did not
occur.  Somehow the message catalog is not being consulted properly, and
it may be related to the fact that your gettext installation is old.

I first tried the m4 testsuite (m4 is simpler than coreutils for creating
a tarball that exercises the same gnulib unit test for quotearg).  But on
a similarly old Solaris platform that I have access to, I didn't see a
failure on my first attempt:

$ uname -a
SunOS perth 5.8 Generic_108528-15 sun4u sparc SUNW,Sun-Blade-100 Solaris
$ gettext --version | head -n1
gettext (GNU gettext-runtime) 0.14

I finally figured out that the reason is that ENABLE_NLS is defined by
default for coreutils (which uses gettext) but not for m4 (which has not
been through the i18n process yet).

A quick run of:
$ make clean
$ make CFLAGS='-g2 -DENABLE_NLS=1' check
...
test-quotearg.c:182: assertion failed
Abort
...

so I can see a similar failure.  However, I'm wondering if there is an
automake bug where specifying TESTS changes the environment passed to the
tests, because I see the failure for a full run but not on an individual run:

$ make -C tests TESTS=test-quotearg CFLAGS='-g2 -DENABLE_NLS=1' check
PASS: test-quotearg
=============
1 test passed
=============

I also noticed this during compilation:
gcc -std=gnu99  -I. -I../lib  -I. -I. -I.. -I./.. -I../lib -I./../lib -
I../intl   -g2 -DENABLE_NLS=1 -MT test-quotearg.o -MD -MP -MF .deps/test-
quotearg.Tpo -c -o test-quotearg.o test-quotearg.c
In file included from test-quotearg.c:33:
../lib/gettext.h: In function `npgettext_aux':
../lib/gettext.h:158: warning: implicit declaration of function `dcngettext'
../lib/gettext.h:158: warning: initialization makes pointer from integer
without a cast
../lib/gettext.h: In function `dcnpgettext_expr':
../lib/gettext.h:259: warning: assignment makes pointer from integer
without a
cast
mv -f .deps/test-quotearg.Tpo .deps/test-quotearg.Po

Maybe this points to evidence that gettext 0.14 may be too old compared to
what gnulib was expecting?

Next, I tried running things in the debugger:

$ cd tests
$ ldd test-quotearg
        libm.so.1 =>     /usr/lib/libm.so.1
        libc.so.1 =>     /usr/lib/libc.so.1
        libdl.so.1 =>    /usr/lib/libdl.so.1
        /usr/platform/SUNW,Sun-Blade-100/lib/libc_psr.so.1
$ env LOCALE=fr_FR.UTF-8 LOCALEDIR=locale gdb test-quotearg
...
(gdb) p unibyte_locale
$1 = false
(gdb) n
207       switch (quoting_style)
(gdb)
249             char const *left = gettext_quote (N_("`"), quoting_style);
(gdb) s
gettext_quote (msgid=0x17078 "`", s=locale_quoting_style) at quotearg.c:165
165       char const *translation = _(msgid);
(gdb) s
166       if (translation == msgid && s == clocale_quoting_style)
(gdb) p translation
$2 = 0x17078 "`"
(gdb) bt
#0  gettext_quote (msgid=0x17078 "`", s=locale_quoting_style) at
quotearg.c:166
#1  0x00013220 in quotearg_buffer_restyled (
    buffer=0x279b8 '¥' <repeats 100 times>, buffersize=100, arg=0x16d38 "",
    argsize=0, quoting_style=locale_quoting_style, flags=0,
    quote_these_too=0x27e38) at quotearg.c:249
#2  0x00014528 in quotearg_buffer (buffer=0x279b8 '¥' <repeats 100 times>,
    buffersize=100, arg=0x16d38 "", argsize=0, o=0x0) at quotearg.c:585
#3  0x000119d8 in use_quotearg_buffer (str=0x16d38 "", len=0xffbeed34)
    at test-quotearg.c:225
#4  0x0001174c in compare_strings (func=0x11990 <use_quotearg_buffer>,
    results=0x278bc) at test-quotearg.c:195
#5  0x00012094 in main (argc=1, argv=0xffbeee24) at test-quotearg.c:309
(gdb) p gettext
$3 = {<text variable, no debug info>} 0xff2fa49c <gettext>

so the call to gettext didn't return any translation.  But I guess that's
kind of expected, since I don't see libintl.so in the list of libraries
linked into the image, and Solaris native gettext is not as full-featured
as GNU gettext.

And that's where I'm stuck.  Maybe Bruno has some better ideas at what to
look at next, since he maintains GNU gettext, and rewrote that portion of
test-quotearg to force the use of a .mo file with locale-specific quotes?

- --
Don't work too hard, make some time for fun as well!

Eric Blake             address@hidden
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (Cygwin)
Comment: Public key at home.comcast.net/~ericblake/eblake.gpg
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iEYEARECAAYFAknNaG8ACgkQ84KuGfSFAYAA/ACgjqpihxOeFfC9PRco/WWNJp05
5ToAn1EMQ/4bAJNBW63VIJf+iwhCH8/B
=8Woq
-----END PGP SIGNATURE-----