Re: Empty matches in global search (string library versus sed)

[Roberto Ierusalimschy <roberto@...>]

> In string.gmatch and string.gsub, when an empty (zero-length) match is 
> encountered, it is always accepted; the search position is then incremented 
> by 1 character.
> Examples:
>   ("ab"):gsub(".*", "#") --> "##", 2
>   ("ab"):gsub("a*", "#") --> "##b#", 3
>
> In sed's "s" command (with "g" flag), when an empty match is encountered, 
> it is accepted only if it's not adjacent to the previous match, otherwise 
> it's ignored.
> Examples:
>   s/.*/#/g
>   ab
>   #
>
>   s/a*/#/g
>   ab
>   #b#
>
> Notes:
> *  I'm not proficient with sed; feel free to correct me.
> *  Python's re.sub behaves (in regard to empty matches) like sed.
>
> Questions:
> a) which behavior seems more intuitive?
> b) if it's that of sed, is it worthwhile for string library (in 5.2) to 
> adopt sed's handling of empty matches?

For me none of them is intuitive. (In particular, the Lua behavior
is only a way to avoid an infinite loop, which would be the correct
[if not the intuitive] behavior.)

-- Roberto