Well I can answer my own question now, as well as a development log in case I forgot this later:
It seems to be a fault of Apple's implementation of libc in iOS. The implementation of iswalpha() is incomplete considering letters in languages other than English. The specific letters(ú,á,ó,...) in different languages could not be recognized by iswalpha(), because they fall out of the 0x7F ASCII boundry, and for somehow it could not be recognized by iOS's locale processing functions, but obviously in different locale those should still be readable alphabet letters.
Some details about it:
iswalph() in iOS is tracked down to:
__DARWIN_CTYPE_static_inline int
__istype(__darwin_ct_rune_t _c, unsigned long _f)
{
#ifdef USE_ASCII
return !!(__maskrune(_c, _f));
#else /* USE_ASCII */
return (isascii(_c) ? !!(_DefaultRuneLocale.__runetype[_c] & _f)
: !!__maskrune(_c, _f));
#endif /* USE_ASCII */
}
and it is __maskrune(_c, _f)) that in the end returns 0.
It is understandable that Apple missed this point since, nobody will use iswalpha() in Objective-C. However it may still be useful to note this point for some porting projects. It was a widely used function so maybe important to many legacy projects that porting to iOS. Hope Apple could fix it in later release.
My workaround now to this problem is to have a wrapper function of iswalpha(), which handle these Latin letters by my own code. Now the app runs flawlessly in my iPhone!