Re: distinguishing multibyte/unibyte ASCII (was: [PATCH] url: Wrap cookie headers in url-http--encode-string.)

[Eli Zaretskii]

> From: Stefan Monnier <address@hidden>
> Cc: Alain Schneble <address@hidden>, address@hidden, address@hidden,
>         address@hidden
> Date: Fri, 09 Sep 2016 16:01:57 -0400
> 
> At some point I tried to change this handling (not exactly fix it) by
> treating multibyte ASCII strings specially (it's easy to recognize by
> checking that the char length is equal to the byte length and both are
> readily available in the "struct Lisp_String" object).  Then when we
> read an ASCII string, instead of making it unibyte, I'd keep it as
> multibyte.  And then change things like "concat" so that those "ASCII
> multibyte" strings don't force the result to be multibyte.
> 
> My local Emacs still runs with those changes, but in the end I don't
> think the result is really better (or sufficiently better to justify
> the subtle incompatibilities it introduces).
> 
> [ Also, I wouldn't be surprised to hear that such a change causes real
>   problems with utf-7 or EBCDIC, or other systems where decoding/encoding
>   a string of bytes/chars all <127 is not a no-op.  ]

We could change concat (and other relevant functions, like format) to
recognize ASCII strings safely and reliably, but I think this would
make those functions slower, which would be a performance hit, since
those functions are used a lot in the inner loops.

This issue is actually rather marginal in Emacs, the url-http use case
is one of a very few that care, because they need to report length in
bytes.