Re: WIDECHAR support -- sizeof(stringliteral) patch

[Howard K. <ho...@ka...>]

>o How do we use ctype functions? For example, how do we use=
 isupper()
>for the narrow character version, and iswupper() for the wide
>character version?

>A solution could be to define a macro per ctype function. For
>example, TRIO_ISUPPER() would be a macro for isupper() for=
 narrow
>characters and iswupper() for wide characters.

That's the Microsoft TCHAR solution.
Which actually works pretty well.
I've started down that road actually -- looks promising.
BTW, Microsoft defines _istupper() mapping to isupper/iswupper=
 depending 
on various details. Quite an extensive list actually (TCHAR.H),=
 though we 
won't need that many.

I started down this path, but now that I'm getting a feel for the=
 code I 
want to go back over it from scratch and first come up with the=
 naming 
rules. We have some options, will post a note later in the week=
 or next 
weekend w/details and see what everyone thinks.

>o How do we use triostr functions? As above, we use=
 trio_length() etc.
>in a number of places. But the wide character version has a
>different
>name.
Same issue as above. is*() functions, str/wcs*() functions,=
 triostr 
functions, etc.

>o How do we specify the name of the printf functions? For=
 example, the
>narrow version must be called fprintf() whereas the wide=
 character
>version must be called fwprintf().
This one's actually got some leeway -- do we go for rigorous=
 standard 
compliance (and use similar rules for extensions), or do we go=
 for pure 
trio unique names and provide mapping for 'drop-in replacement'?=
 e.g. do 
we provide
=09int printf(const char *,...)
or
=09int trio_printf(const char *,...)
=09#if defined(TRIO_REPLACE_STD)
=09# define printf trio_printf
=09#endif
I vote for the latter.

If we name everything trio*() then we just have to make sure we=
 have a 
compatible replacement for everything we want to replace, e.g.=
 ANSI C/C++ 
define
=09sprintf
=09snprintf
=09wsprintf
BUT...
1) ...there's no wsnprintf()
2) ...wsprintf() is functionally equivalent to wsnprintf()
Thus even though it's called wsprintf it's really a 
Wide-String-Nsized-PRINTF. Should trio provide all 4 functions or=
 just 
the pure 3, i.e.
=09...narrow flavors...
=09trio_wsnprintf()
=09#define wsprintf=09=09trio_wsprintf
or
=09...narrow flavors...
=09trio_wsprintf()
=09trio_wsnprintf()
=09#if defined(TRIO_REPLACE_STD)
=09#define wsprintf=09=09trio_wsnprintf
=09#endif

I went thru this recently.
This is, IMO, a huge pain converting a 'narrow' program to 'wide'=
 
strings. Sure sprintf may potentially cause buffer overruns, but=
 do you 
know how easy it is to change
=09sprintf(buffer, "fmt", vars)
to
=09wsprintf(buffer, ***size***, "fmt", vars)
It may be the "right" thing to do, but in some code it can be=
 extremely 
time consuming and far lower on the priority list. Also can't=
 really fake 
it out via macros unless you've got varadic(?) macros, which=
 almost no 
one supports today.

Fundamentally we have to answer, should trio provide=
 wsnprintf+wsprintf 
as wide equivalents to snprintf/sprintf, or should trio only=
 provide 
wsprintf as per C/C++ standards (even though, by definition, we=
 need to 
provide snprintf).

Answer that, and the rest is just details.

>o We need to ensure standard compliancy. This involves reading
>carefully through the standards.
What about "implementation dependent" behavior?
Thought I recall some of the printf family varies here (return=
 value? of 
sprintf?). Been a while so maybe not, and even if so maybe C99=
 cleans it 
up.

>o In the wide character version we must determine when to use=
 int and
>when to use wint_t.
This one's going to be rough, though not as bad as the standards=
 
compliance.

>o Grouping and scan ranges will probably fail (or generate huge
>arrays). These needs to be looked at.
Yes. I have no idea how to handle these, though to be honest I=
 haven't 
put any thought into them.



I need to do some 'real' work for a few days so I'll probably=
 hack some 
more trio code next weekend. Top of the list is a single-source=
 triostr.c 
that can be compiled twice (narrow + wide), plus a list of naming=
 
conventions we can discuss.

=09- Howard

--
Email: mailto:ho...@ka...  WWW:=
 http://www.kapustein.com/howard/
--
"IT'S GOD'S RESPONSIBILITY TO FORGIVE BIN LADEN...
=09=09IT'S OUR  RESPONSIBILITY TO ARRANGE THE MEETING."
=09=09=09=09- United States Marines