Checking format specifiers (Was: Re: Support for %d$c format specifier in diagnostics.c)

[Wolfgang Bangerth <bangerth at ices dot utexas dot edu>]

Hi Chiaki,
I think it's excellent someone finally looks at positional arguments for 
printf! gcc has repeatedly been bitten by the following problem: there's a 
message like 
  Variable %s doesn't exist in class %T
(or so, don't know whether the format specifiers make sens, just assume 
they're different), and some translation reads
  In der Klasse %T gibt es keine Variable %s

That's easily cured by positional parameters, but unfortunately our 
translations are full of such errors. Another, even more obviously wrong 
problem is that messages do not have the same number of format specifiers as 
the orginal message had. We had several reports for crashes due to this, see 
in particular
  http://gcc.gnu.org/bugzilla/show_bug.cgi?id=9390
and the thread that is referenced in message #5 there, and also
  http://gcc.gnu.org/bugzilla/show_bug.cgi?id=7765
Because the translations happen outside the gcc project, there was not much I 
could do about it. However, I once made an attempt to write a small script 
that goes through the translations and checks for every message whether the 
format specifiers in the orginal and translated messages are the same in 
number and appear in the correct order (in absence of positional 
information). The result was largely devastating: there were translations 
with hundreds of errors. I imagine that if anyone uses them, they would very 
frequently get compiler crashes for this.

I am not familiar with the translation community, but if you are: one of the 
really necessary additions in their processes would be a small program that 
does what my small script tried to do in a thourough way (i.e., for example, 
taking into account positional parameters). If we would then only get "valid" 
translations, we should add this script to our makefiles, so that no broken 
translations can be checked in unless they pass the script (could be done as 
a part of bootstrapping). Since you are already working on a parser for these 
things, writing a checker might (or might not) be a byproduct of your work, 
and one that would certainly be appreciated!

Thanks for your work
  Wolfgang

-------------------------------------------------------------------------
Wolfgang Bangerth              email:            bangerth@ices.utexas.edu
                               www: http://www.ices.utexas.edu/~bangerth/