Re: libstdc++/4458

[James Kanze <kanze at YAST_ASK>]

Martin Sebor <sebor@roguewave.com> writes:

|>  http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view&pr=4458&database=gcc

|>   |>  It's not required to do so for efficiency (see 22.2.1.3.2, p5).

|>   §22.2.1.1.1/1: concerning std::ctype<>::is:

|>       Returns: do_is(m,c)

|>   Can't be any clearer than that.  The function do_is is required by
|>   the standard to be virtual, which means that the actual function
|>   called (and thus the actual semantics of the function is) depend on
|>   the dynamic type of the object.  This is, after all, the whole
|>   point behind defining the protected virtual functions -- any other
|>   interpretation, and having the protected interface makes sense.

|>  No, you're quoting the requirement on the primary template, but your
|>  issue (titled std::ctype<char>::is ...) is with the specialization
|>  on char. The requirement I referred you to in 22.2.1.3.2, p5 is the
|>  one that's relevant here. The reason for not calling the virtual
|>  function from the ctype<char>::is() is, as I said earlier, for
|>  efficiency.

I've found it now.  It never occurred to me that the standard could
require that a specialization have different semantics than the basic
template.  I guess the bug is in the standard, and not in g++.

I conclude that if I want to implement the iso_8859_1 locale for ctype,
it should be sufficient to provide the correct table.  I don't need it
for my current work, so finding time for it isn't necessarily easy, but
if I can get around to it, do I have to do anything special to get it
integrated into the normal g++ releases?

-- 
James Kanze                                mailto:kanze@gabi-soft.de
Conseils en informatique orientée objet/
                    Beratung in objektorientierter Datenverarbeitung
Ziegelhüttenweg 17a, 60598 Frankfurt, Germany Tel. +49(0)179 2607481