Re: [Qucs-devel] strtod-woes and setlocale-salvation ...

[Stefan J. <st...@gr...>]

Am Mo, 3.07.2006, 10:05, schrieb Raimund 'Raimi' Jacob:

> morning!

hi there again,

>> i just made up my mind about this topic.  i would like to summarize:
>>
>> you have somehow set your locale to "de_DE.utf8", correct?  thus the
>> qucs program failed to read e.g "1.2e8" data, because it uses strtod()
>> which expects "1,2e8" with this locale set?  a solution was to
>>
>>   setlocale (LC_NUMERIC, "C");
>>
>> in the graph constructor.
>>
>> so far for the problem (and its kinda solution).
>>
>> first thing i don't understand is this:  how does the program get this
>> locale state?  there is now statement in the program which
>>
>>     setlocale (LC_ALL, "de_DE.utf8");
>>
>> and with
>>
>>   $ export LC_ALL=de_DE.utf8
>>
>> I can't reproduce the problem.  so how does this work?  anyone an idea?
>
> the locale setting is a process-wide setting just like for example the
> current working directory. i think you just hit one of the weak points
> of low-level programming that evolved with the invention of
> international languages after everyone got used to using ASCII.
>
> while it is very reasonable for a user to set LANG or LC_* to something
> sane for his/her environment, programmers must take extra precautions so
> that their (plain text) file formats dont get b0rked.
>
> so the "bug" is actually that you are using a parsing funtion that is
> influenced by a locale setting while your file format is not localized
> and defined to contain (in this case) numerical values as externalized
> by the C/POSIX locale.

to be precise:  the file formats should be portable, thus we must stay
with the "C" locale representation of numbers.

>> anyway.  I think the actual solution should not be placed in the graph
>> constructor but in the main program.  thus I sent Raimund a patch for
>> CVS which
>>
>>     setlocale (LC_NUMERIC, "C");
>
> see above, it is a process-wide setting. so watch out for side-effects.
> are there any places where users can user numbers according to their
> locale?

as said:  any inputs into qucs should be locale "C".  parsing numbers
should fail if the user tries to enter "1,23" instead of "1.23".

bottom line: you missed the problem i described.  at my computer i can
set LANG=de_DE.utf in the environment.  this does *not* change the
behaviour of the program at all.  but setlocale(LC_NUMERIC,"de_DE.utf")
*does* change it!  also it seems that at christopher's laptop the
environment is setup using LANG=de_AT.UTF-8 and this *does* change
the program behaviour.  this is what i don't understand.

cheers, stefan.