Re: proposed Eistring interface

Ben> "Stephen J. Turnbull" wrote: Ben> you are confused. Oh. <+J,$G$7$g$&!# Now please explain to me what good your check is going to do with the absolutely no ASCII nowhere 100% Japanese string (which means, appropriately, "on the contrary, aren't _you_ confused?") on the line above. The point is that there are _no_ characters representable in one byte that can harm _Mule_ if for practical purposes #define eicpy_c(ei, c_string) eicpy_ext(ei, c_string, Qbinary) They can "only" cause mojibake and even external data corruption if written back out under an inappropriate coding system. OTOH, if you want to prevent mojibake and external data corruption, you need to have a full (and 100% accurate) character set autodetector, or at least a 100% reliable undo on all such operations. There may be a way to make handling external-format text Mule-safe, but this isn't it. External-format text is by definition just infested with various ISO-sponsored diseases (and Microsoft toxins) and needs to be hedged round with real error-checking, with sharp teeth and thick hair on its chest. Not half-ass filters that won't detect people doing many common kinds of Stupid Things[tm], while preventing arguably legitimate and convenient usages like ISO-8859-1 literals. (Remember the actual implementation of Qbinary!) If I had a veto, I'd use it on anything that allowed implicit typing of any external format data. Ben> This is defined only to make it easier to use literal ASCII Ben> strings, which is going to be extremely common. If the Ben> string has any non-ASCII characters in it, they need to Ben> explicitly specify the encoding -- hence the restriction. This is just plain evil. People are going to write code that is allegedly Mule-safe because it used your API, and have it randomly abort in non-ASCII locales when somebody feeds it non-ASCII data. And need I mention that such code will be 100% un-gettext-izable? Apparently. That issue needs to be carefully thought through before you allow literals to be handled by these APIs. A better way of dealing with the issues with literals would simply be to specify that all XEmacs source files must be encoded in UCS-4 with the UTF-8 representation. (That means that they must respect the reserved areas in UCS-4 if they want to go beyond the UTF-32 space.) Then string literals in the source code can be treated specially (but generic C strings[1] should not be, since you don't know where they come from), and will not need a coding system associated with them. (Obviously, we have to get a usable Unicode coding-system implemented first so that this is transparent to developers and users.) For the moment, we can just make that a restriction to ASCII, except in mule-packages. This will be easily extensible to UTF-8 with zero fuss as soon as we have the UTF-8 coding system. And this would be transparent to gettext, should we decide that is the way to go. Anyway the pro-ASCII bias is just not PC. Footnotes: [1] Hm. What about .so modules? -- University of Tsukuba Tennodai 1-1-1 Tsukuba 305-8573 JAPAN Institute of Policy and Planning Sciences Tel/fax: +81 (298) 53-5091 _________________ _________________ _________________ _________________ What are those straight lines for? "XEmacs rules."