Re: One more string functions change

[Dmitry Antipov]

On 06/29/2014 07:13 PM, Eli Zaretskii wrote:

> It's possible that we should consider this now, but the answer to your
> question is not a trivial one in any case.  Emacs traditionally
> exposed to Lisp all the Unicode character properties, as char-tables.

Are these exposed properties really used from Lisp in a high-level,
user-defined manner? For example, is it desirable/possible to customize
related things via .emacs? Or is there major/minor mode which relies
on the Lisp-visible character properties?

> If we decide to use ICU, we'd need to think what to do with those
> char-tables: remove them, populate them using ICU, something else?
> (Having these databases twice would be an unnecessary bloat, IMO.)

Yes, ICU itself is bloated enough. On my system, shared library
with compiled-in Unicode data is > 20M. Nevertheless it's commonly
considered "not too bloated" even for relatively small systems like
the modern Android-based gadgets.

> Some of these properties need to support very fast access (e.g., for
> bidi display), and the question is how fast is ICU in this regard.
> Also, many Unicode features are already implemented, so they should be
> reworked or refactored, or maybe the corresponding ICU features left
> unused.  And features that depend on Unicode, like font selection,
> will have to be adapted.

IIUC the things are even worse because ICU uses 16- and 32-bit quantities
to represent Unicode characters; this doesn't look too compatible
with our internal variable-size, 1-5 bytes-width encoding.

Dmitry