bug#21124: 'invalid-mb-seq-UMR' test fails

[Assaf Gordon]

Hello,

On Mon, Jul 27, 2015 at 10:31:21PM -0700, Jim Meyering wrote:
> On Thu, Jul 23, 2015 at 2:47 PM, Assaf Gordon <address@hidden> wrote:
> > With sed from git (4.2.2.89-e387), I get one test failure in
> > 'invalid-mb-seq-UMR' (log attached).
> >
> > The error seem to be:
> > ===
> > +valgrind --quiet --error-exitcode=1 sed -e 's/a/b\U\xb2c/' in
>
> What versions of valgrind and gcc are you using -- and what type of system?

I'm using:

    $ uname -a
    Linux XXX 3.13.0-58-generic #97-Ubuntu SMP Wed Jul 8 02:56:15 UTC 2015 
x86_64 GNU/Linux
    $ gcc --version
    gcc (GCC) 5.2.0
    $ valgrind --version
    valgrind-3.10.0.SVN

But, looking deeper I've noticed the test uses a locale which is not 
available on my system (ja_JP.eucJP).
Here are couple of related tests:

    # (the extra commit .90 is the ununsed-code removal
    #  http://lists.gnu.org/archive/html/bug-sed/2015-07/msg00005.html)
    $ ./sed/sed --version | head -n1
    ./sed/sed (GNU sed) 4.2.2.90-9f3b

    # The locale I don't have produces 'ff'
    $ echo a | LC_ALL=ja_JP.eucJP ./sed/sed -e 's/a/b\U\xb2c/' | od -tx1a
    0000000  62  ff  43  0a
              b del   C  nl
    0000004

    # A locale I do have works as expected
    $ echo a | LC_ALL=ja_JP.utf8 ./sed/sed -e 's/a/b\U\xb2c/' | od -tx1a
    0000000  62  43  0a
              b   C  nl
    0000003

For comparison, an older 'sed':

    $ sed --version
    sed (GNU sed) 4.2.2

    # Existing locale
    $ echo a | LC_ALL=ja_JP.utf8 sed -e 's/a/b\U\xb2c/' | od -tx1a
    0000000  62  00  43  0a
              b nul   C  nl
    0000004

    # non-existing locale
    $ echo a | LC_ALL=ja_JP.eucJP sed -e 's/a/b\U\xb2c/' | od -tx1a
    0000000  62  ff  43  0a
              b del   C  nl
    0000004

So perhaps the fix is simply to skip the test if the locale doesn't 
exist.

HTH,
-assaf