Declare PDC_wcwidth() in curses.h; it defers to the locale even with PDC_FORCE_UTF8

[serhiy-storchaka]

Two related things about PDC_wcwidth().

1. It isn't declared in a public header. The function is exported (PDCEX; nm shows T PDC_wcwidth), but declared only in curspriv.h, so a program linking against PDCursesMod has to declare the prototype itself.

This matters most on Windows, where the C runtime has no wcwidth() at all, so PDC_wcwidth() is the only implementation available in the process. A binding that needs to tell a spacing character from a zero-width combining mark — to validate setcchar() input, say — has to carry its own copy. Python's curses module is gaining PDCursesMod support and currently vendors ~120 lines of Markus Kuhn's wcwidth() for the Windows build, duplicating what the library already has and exports. Declaring the existing symbol in curses.h would let consumers drop that.

2. With PDC_FORCE_UTF8, deferring to the system wcwidth() looks wrong. HAVE_WCWIDTH is chosen from __linux + _XOPEN_SOURCE alone, independently of PDC_FORCE_UTF8. HISTORY.md gives the reasoning:

On Linux, we use the system-provided wcwidth() instead of our own. Aside from avoiding redundant code, this may help in non-Unicode locales (our wcwidth() is for Unicode only).

That reasoning holds for a locale-driven build, but PDC_FORCE_UTF8 means the library encodes cells as UTF-8 whatever the locale says — so classification should be Unicode-based too. Built WIDE=Y UTF8=Y with -D_XOPEN_SOURCE=700, under uk_UA.koi8u:

U+0301 (combining acute):  wcwidth() = -1,  PDC_wcwidth() = -1

The same library in a UTF-8 locale returns 0. Would it make sense for PDC_FORCE_UTF8 to select the built-in Unicode table regardless of platform?

(PDCursesMod 4.5.4, vt port, Linux.)

[Bill-Gray]

At the very least, we could add the following to curses.h :

#if defined( PDC_FORCE_UTF8) || !defined( HAVE_WCWIDTH)
PDCEX int PDC_wcwidth( const int32_t ucs);
#endif

i.e., if we're building PDC_wcwidth() anyway, mention that fact in curses.h. Code using PDCursesMod could then say

#if defined( PDC_FORCE_UTF8) || !defined( HAVE_WCWIDTH)
   width = PDC_wcwidth( c);
#else
   width = wcwidth( c);
#endif

I can't really come up with a down side to this.

Would it make sense for PDC_FORCE_UTF8 to select the built-in Unicode table regardless of platform?

Hmmm... we could do that, yes. We could then have PDC_wcwidth() always be available (and would probably rename it to something like PDC_unicode_width(), so that people didn't call it and expect to get a locale-specific width.)

Programs could then call PDC_unicode_width() to get a character's width in Unicode, and call wcwidth() to get that character's width in the current locale (if the platform actually has that function, i.e., isn't a Microsoft one... and, of course, it wouldn't necessarily be the same character). Whether having both sets of widths available at the same time would be useful is another question.

The only argument against it is that of possibly redundant code. Within PDCursesMod, we'd use #ifdefs such as the above to determine which function to use. Not a big deal, though; a quick 'git grep wcwidth` suggests we'd be talking about changes in less than a dozen lines of code.

We do have the remaining ugliness that on Microsoft® products, it appears that our only knowledge of glyph width is from this Unicode-only function. So we'd be telling people about a width that would not necessarily match the reality of their current (non-Unicode) locale. It may actually be best on Windows to say : if the current locale uses Unicode, call PDC_unicode_width(); otherwise, assume the characters all have size 1. (Unless we wish to create tables for various other character encodings.) We essentially have no way to support non-Unicode encodings of zero- or fullwidth characters; we can only hope they're all half-width.

[serhiy-storchaka]

From the user's point of view, it would be simpler if PDC_wcwidth() was always declared, even if it is just a wcwidth() wrapper.

For now I follow the example of the demos (demos/ncurses_cfg.h):

PDCEX int PDC_wcwidth(const int32_t ucs);
#define wcwidth PDC_wcwidth