On Mon, 7 Sep 2020 11:08:23 +0200
Corinna Vinschen wrote:
> Hi Takashi,
>
> On Sep 7 13:45, Takashi Yano via Cygwin-patches wrote:
> > #if 0 /* Let's try this if setting codepage at pty open time is not enough
> > */
> > - if (!cygheap->locale.term_code_page)
> > - cygheap->locale.term_code_page = __eval_codepage_from_internal_charset
> > ();
> > + if (!get_ttyp ()->term_code_page)
> > + get_ttyp ()->term_code_page = __eval_codepage_from_internal_charset
> > (NULL);
> > #endif
>
> *If* we revert back to using setup_locale, these #if blocks would
> go away.
>
> > -__eval_codepage_from_internal_charset ()
> > +__eval_codepage_from_internal_charset (const WCHAR *envblock)
> > {
> > - const char *charset = __locale_charset (__get_global_locale ());
> > + const char *charset;
> > + __locale_t *loc = NULL;
> > + if (__get_current_locale ()->lc_cat[LC_CTYPE].buf)
> > + charset = __locale_charset (__get_current_locale ());
> > + else
> > + {
> > + char locale[ENCODING_LEN + 1] = {0, };
> > + if (envblock)
> > + {
> > + const WCHAR *lc_all = NULL, *lc_ctype = NULL, *lang = NULL;
> > + for (const WCHAR *p = envblock; *p != L'\0'; p += wcslen (p) + 1)
> > + if (wcsncmp (p, L"LC_ALL=", 7) == 0)
> > + lc_all = p + 7;
> > + else if (wcsncmp (p, L"LC_CTYPE=", 9) == 0)
> > + lc_ctype = p + 9;
> > + else if (wcsncmp (p, L"LANG=", 5) == 0)
> > + lang = p + 5;
> > + if (lc_all && *lc_all)
> > + snprintf (locale, ENCODING_LEN + 1, "%ls", lc_all);
> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> sys_wcstombs (locale, ENCODING_LEN + 1, lc_all);
>
> OTOH, if you read these environment vars right from our current POSIX
> env, you don't have to convert from mbs to wcs at all. Just call
> getenv("LC_ALL"), etc. After all, envblock is just the wide char
> copy of our current POSIX env.
IIUC, envblock is not a copy of environment if exec*e() is used.
In this case, getenv() cannot retrieve the new environment values
passed to exec*e(). This is needed by the test case bellow.
int pm = getpt();
if (fork()) {
[do the master operations]
} else {
char *env[] = {"LANG=ja_JP.SJIS", ...., NULL};
setsid();
ps = open(ptsname(pm), O_RDWR);
close(pm);
dup2(ps, 0);
dup2(ps, 1);
dup2(ps, 2);
close(ps);
execle("/bin/tcsh", "/bin/tcsh", "-l", NULL, env);
}
> > + else if (lc_ctype && *lc_ctype)
> > + snprintf (locale, ENCODING_LEN + 1, "%ls", lc_ctype);
> > + else if (lang && *lang)
> > + snprintf (locale, ENCODING_LEN + 1, "%ls", lang);
> > + }
> > + if (!*locale)
> > + {
> > + const char *env = __get_locale_env (_REENT, LC_CTYPE);
> > + strncpy (locale, env, ENCODING_LEN);
> > + locale[ENCODING_LEN] = '\0';
> > + }
> > + loc = duplocale (__get_current_locale ());
> > + __loadlocale (loc, LC_CTYPE, locale);
> > + charset = __locale_charset (loc);
> > + }
>
> Oh, boy, this is really a lot. I have some doubts this complexity is
> really necessary. It's a bit weird to go to such great lengths for
> native applications. Still, why not just do this once in the process
> creating the pty rather than trying on every execve?
This is executed just once for a pty. Because
__eval_codepage_from_internal_charset() is called only when
get_ttyp ()->term_code_page is not set yet.
--
Takashi Yano <takashi.y...@nifty.ne.jp>
