Dear list, A problem with creating pdf or postscript plots from an embedded R, and in an environment where a locale is defined, was recently reported on the rpy (Python->R interface) mailing list. The problem appears to be that whenever the locale defines ',' as a separator for the decimal part of a numerical value, the pdf/ps file generated still has the ',' instead of a '.' causing problem when open the resulting file with a postscript/pdf viewer. Of course, an obvious workaround would be to force the locale for numbers to C ( LC_NUMERIC = "C" ), but that could be overriding internationalization features otherwise in R. Any advice ? Thanks, Laurent
problem with postscript/pdf plots, locales and embedded R
5 messages · Laurent Gautier, Brian Ripley
On Fri, 14 Nov 2008, laurent wrote:
Dear list, A problem with creating pdf or postscript plots from an embedded R, and in an environment where a locale is defined, was recently reported on the rpy (Python->R interface) mailing list. The problem appears to be that whenever the locale defines ',' as a separator for the decimal part of a numerical value, the pdf/ps file generated still has the ',' instead of a '.' causing problem when open the resulting file with a postscript/pdf viewer. Of course, an obvious workaround would be to force the locale for numbers to C ( LC_NUMERIC = "C" ), but that could be overriding internationalization features otherwise in R. Any advice ?
Don't set LC_NUMERIC with R: ?Sys.setlocale did warn you about that. It used to cause far more problems than this.
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
Thanks for the quick answer. The documentation for Sys.setlocale mentions the possibility of warning when LC_NUMERIC is set. After checking it is looking like setting LC_NUMERIC from Sys.setlocale does issue a warning, but there is does not seem to be a warning when LC_NUMERIC is set in the parent shell. In the case the later is likely to cause trouble, would it be appropriate to issue a warning when R is initialized ? Laurent
On Fri, 2008-11-14 at 19:37 +0000, Prof Brian Ripley wrote:
On Fri, 14 Nov 2008, laurent wrote:
Dear list, A problem with creating pdf or postscript plots from an embedded R, and in an environment where a locale is defined, was recently reported on the rpy (Python->R interface) mailing list. The problem appears to be that whenever the locale defines ',' as a separator for the decimal part of a numerical value, the pdf/ps file generated still has the ',' instead of a '.' causing problem when open the resulting file with a postscript/pdf viewer. Of course, an obvious workaround would be to force the locale for numbers to C ( LC_NUMERIC = "C" ), but that could be overriding internationalization features otherwise in R. Any advice ?
Don't set LC_NUMERIC with R: ?Sys.setlocale did warn you about that. It used to cause far more problems than this.
You start from a false premise, and it seems that you are reporting a
serious bug in Rpy to the wrong list:
Note that R expects to be run with the locale category `LC_NUMERIC' set
to its default value of C, and so should not be embedded into an
application which changes that.
('Writing R Extensions').
On Sat, 15 Nov 2008, laurent wrote:
Thanks for the quick answer. The documentation for Sys.setlocale mentions the possibility of warning when LC_NUMERIC is set. After checking it is looking like setting LC_NUMERIC from Sys.setlocale does issue a warning, but there is does not seem to be a warning when LC_NUMERIC is set in the parent shell. In the case the later is likely to cause trouble, would it be appropriate to issue a warning when R is initialized ?
Of course R does not violate its own documentation. None of the R executables reset the LC_NUMERIC locale category, and all C programs start in the 'C' locale (both facts you should have known from the homework you did before posting). BDR
Laurent On Fri, 2008-11-14 at 19:37 +0000, Prof Brian Ripley wrote:
On Fri, 14 Nov 2008, laurent wrote:
Dear list, A problem with creating pdf or postscript plots from an embedded R, and in an environment where a locale is defined, was recently reported on the rpy (Python->R interface) mailing list. The problem appears to be that whenever the locale defines ',' as a separator for the decimal part of a numerical value, the pdf/ps file generated still has the ',' instead of a '.' causing problem when open the resulting file with a postscript/pdf viewer. Of course, an obvious workaround would be to force the locale for numbers to C ( LC_NUMERIC = "C" ), but that could be overriding internationalization features otherwise in R. Any advice ?
Don't set LC_NUMERIC with R: ?Sys.setlocale did warn you about that. It used to cause far more problems than this.
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
That might also be reporting something that can be felt like a brittle aspect of R when embedded. I am very tempted to ask why enforcing LC_NUMERIC to "C" for the whole process while this only concerns numerical values written into "data" files. I am understanding that at least postscript/pdf files are affected, and so might files representing R serialized objects. Functions like glib's 'g_ascii_dtostr()/g_ascii_strtod()' appear to have been developed for such purposes, for example. L.
On Sun, 2008-11-16 at 12:40 +0000, Prof Brian Ripley wrote:
You start from a false premise, and it seems that you are reporting a
serious bug in Rpy to the wrong list:
Note that R expects to be run with the locale category `LC_NUMERIC' set
to its default value of C, and so should not be embedded into an
application which changes that.
('Writing R Extensions').
On Sat, 15 Nov 2008, laurent wrote:
Thanks for the quick answer. The documentation for Sys.setlocale mentions the possibility of warning when LC_NUMERIC is set. After checking it is looking like setting LC_NUMERIC from Sys.setlocale does issue a warning, but there is does not seem to be a warning when LC_NUMERIC is set in the parent shell. In the case the later is likely to cause trouble, would it be appropriate to issue a warning when R is initialized ?
Of course R does not violate its own documentation. None of the R executables reset the LC_NUMERIC locale category, and all C programs start in the 'C' locale (both facts you should have known from the homework you did before posting). BDR
Laurent On Fri, 2008-11-14 at 19:37 +0000, Prof Brian Ripley wrote:
On Fri, 14 Nov 2008, laurent wrote:
Dear list, A problem with creating pdf or postscript plots from an embedded R, and in an environment where a locale is defined, was recently reported on the rpy (Python->R interface) mailing list. The problem appears to be that whenever the locale defines ',' as a separator for the decimal part of a numerical value, the pdf/ps file generated still has the ',' instead of a '.' causing problem when open the resulting file with a postscript/pdf viewer. Of course, an obvious workaround would be to force the locale for numbers to C ( LC_NUMERIC = "C" ), but that could be overriding internationalization features otherwise in R. Any advice ?
Don't set LC_NUMERIC with R: ?Sys.setlocale did warn you about that. It used to cause far more problems than this.