Re: m4sugar and m4 1.6, bison

[Ralf Wildenhues]

* Ralf Wildenhues wrote on Sat, Jul 12, 2008 at 08:11:55AM CEST:
> So then here's a first prototype for fast m4_append_uniq, completely
> untested, and would probably need some more-unique separator after $1 in
> the definition of m4_key.

I'll blame it on lack of coffee...

> # m4_make_macro_name(STRING)
> # Turn all characters not fitting to be a macro name into '_'.
> m4_define([m4_make_macro_name],
> [m4_bpatsubst([$1], [...])dnl
> ])
> 
> # m4_append_uniq(MACRO-NAME, STRING, [SEPARATOR], [IF-UNIQ], [IF-DUP])
> m4_define([m4_append_uniq],
> [m4_pushdef([m4_key], [_$1_[]make_macro_name([$2])])dnl
> m4_ifdef(m4_key,
>   [m4_if([m4_defn(m4_key)], [$2],

This is completely ignoring hash key collision.  So, assuming that
collisions are rare, i.e., usually strings differ not only in characters
not allowed in a macro name, then the duplicate test here can just reuse
the one from the version of m4_append_uniq currently in Autoconf.  That
will probably still give amortized fast check here.  Actually, I guess
the uniq part is even O(n), only the reallocs necessary now and then
will be loglinear.

>     [$5],
>     [_$1($@)])],
>   [_$1($@)])dnl
> m4_popdef([m4_key])dnl
> ])
> 
> m4_define([_m4_append_uniq],
> [m4_define(m4_key, m4_defn(m4_key))dnl

Then this line can simply be:
[m4_define(m4_key)dnl

> do the appending work, expand $4...
> ])

Still completely untested, of course...

Cheers,
Ralf