Re: LC_ALL and UTF-8 directory names

[Akim Demaille]

Hi Paul!

Le 4 oct. 2011 à 08:42, Paul Eggert a écrit :

> On 10/03/11 07:39, Akim Demaille wrote:
>> The JVM is quite happy with that.  Unless you make the name "invalid"
>> by setting LC_ALL to C for instance, which Autoconf does.
> 
> Eeuuuw!  (I think that's the proper technical term. :-)

In French that would be « Beurk ! » :)

> Hmm, Autoconf itself doesn't have this issue, right?
> as it does not use Java.  So the issue is more about
> configure.ac files that invoke the JVM.

I agree, yet I expected that if there is at least one program that has a 
problem with that, then maybe there are more.  So it would be safer if Autoconf 
took care of this in all the cases.  But I definitely agree that there is some 
balance to find, and I'm fine with deciding that this should be left to the 
troublesome cases (especially that now this message is on the Internet, so 
maybe people facing the same problem will find the correct diagnostic faster :).

> The Autoconf-generated 'configure' code checks that the
> output of "pwd" works when given as an argument to "cd".
> Perhaps configure.ac programs that invoke the JVM should
> also check the JVM equivalent of "cd", to make sure it
> also works.  I don't offhand see how to automate this under
> Autoconf, though, not unless Autoconf picks up a bunch
> of Java support.

I had something simpler in mind: if the locale is not C/POSIX and there are non 
ASCII characters in pwd, then warn.  And add a warning about this in INSTALL.  
Maybe there would be way too many false positives, I don't know — I have no 
idea how many configurers use extended charsets in their directory names.