[Rd] make check for 2.5.1 RC fails on mbcsToSbcs in graphics

[Prof Brian Ripley]

On Wed, 27 Jun 2007, Peter Dalgaard wrote:

> Prof Brian Ripley wrote:
>> But R 2.5.0 does the same, and as far as I know R has done so for many
>> past versions.
>>
>> You are asking the impossible: outputting Latin-1 in a Latin-2
>> environment.  Remember that postscript() does not handle UTF-8, and so
>> uses whatever it thinks the local 8-bit encoding is, in your case
>> either Latin-2 or ASCII.  We can probably work around this, but
>> definitely not at this late stage for 2.5.1.

> Yes. Longer term we probably should. I don't think we want  "make check"
> to be locale-dependent, and we do enforce C locale elsewhere.

But you can't really avoid it if you want to test non-ASCII
features: for example, you cannot test Latin-1 graphics in a C locale.

It was intended that the postscript device opened for testing graphics was 
opened with encoding="ISOLatin1".  That should certainly have helped, at 
least in a UTF-8 locale, but unfortunately there were a couple of typos in 
tests/Examples/Makefile.  Fixing that file makes 'make check' work in 
cs_CZ.utf8 at least on my machine.

It still remains that (for examples) you cannot successfully run make 
check in a MBCS locale without iconv, nor will zh_CN.gbk nor zh_TW.big5 
work.  I've added a note to R-admin to say that you may need to try a 
suitable locale.

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595