[PATCH] Fix libstdc++/7811

[Paolo Carlini]

Roland McGrath wrote:

>>I'm unsure about the following points: in case LC_ALL is not set is the 
>>fallback to LANG  really needed (PR submitter implicitly expected this 
>>to be the case)? Also, if neither are set is always correct that "" -> "C"?
>>    
>>
>The standards specify looking at LC_ALL, then at the variable for the 
>particular category if setting just one, then at LANG.
>
I see. However, do you mean the C standards or the C++ standard?
I cannot find in the latter (more relevant here) a detailed discussion 
of this point (well, the existing code just considers LC_ALL in the 
setlocale...)

>Note that a variable
>that is present but empty (e.g. getenv("LC_ALL") => "") is treated as not
>present.
>
This is an interesting technical point, thanks. However, once more, I'm 
not sure if the C++ standard (vs C) actually prescribes something in 
detail about this. Can you possibly help?

>Off hand I am not sure if the behavior when none are set is
>specified or is implementation-defined.  Defaulting to "C" is safe and
>consistent.
>
Good.

>Another thing to note if you want to match the selection behavior of
>setlocale is that if the name contains a slash and the program is setuid
>then the name will be ignored and "C" used instead.  (It doesn't check
>this on each variable, just on the name chosen from any.)
>
Another interesting technical point. As you say, however, it looks like 
the real point is: "Must we really match the detailed selection behavior 
of setlocale?" I don't believe that  for the implementation of the 
locale functionality at issue this is mandated anywhere in the C++ standard.

Benjamin, what do you think?

My personal opinion is that my proposal is a definite improvement on the 
current situation but I'm not sure how many additional details we must 
consider to agree with the C++ standard.

Ciao, Paolo.