Backward running version look-behinds (Re: Look-behind proposal in trouble)

[Nozomu Katō]

On Thu, 19 Nov 2015 09:04:41 +0100, Yang Guo wrote:
> This implementation supports variable length lookbehind similar to .NET's
> semantics. It does so by emitting code to read backwards inside the
> lookbehind. The size of the change without platform ports and tests is
> about 600 lines.

When regular expressions inside a look-behind are evaluated backwards,
the automaton will not have a chance to know what has been captured in
$1 in /(?<=(.)$1)/, since the automaton reaches $1 prior to (.).

This can be fixed by defining that a backreference referring to a
capture inside the same look-behind has to be put leftwards contrary to
the usual rule, but saying so would impose reading regular expressions
backwards or prefetching forward expressions inside a look-behind upon
the parser, to know if the corresponding capturing parentheses exist.

I am not sure if it is easy or there is a simpler solution, anyhow, if
ECMAScript supports .NET-like look-behinds this matter would need to be
fixed.

Nozomu