Re: Bracket expressions with character ranges are slow

[Paolo Bonzini]

On 05/18/2011 10:40 PM, Paolo Bonzini wrote:
> > Suppose grep had a preprocessor that converted any bracket
> > expression containing elements of different byte sizes, whether
> > [美国a] or a range not all of whose characters are a single byte,
> > into a parenthesized alternation like (美|国|a).  Would this use
> > more memory, constituting a space-for-time tradeoff?  If not, is
> > there some other reason not to do this?
>
> There's no justification but laziness. :)  We already optimized a large
> amount of character ranges---basically all that can be optimized except
> this one.

This is now implemented in grep.git.

> I realize this is only potentially possible for egrep, at least at
> the surface level of rewriting the regular expression.

Since the optimization is done inside the matcher, it does not depend on 
extended regex.

Paolo