On Thu, Sep 06, 2007 at 01:25:12PM -0500, Patrick R. Michaud wrote:
: > Were we using the procedural conjunction:
: >
: > "foobar" ~~ / <[a..z]>+ && [ ... ] /;
: >
: > I would guess that the LHS matches as much as it can ("foobar"), then
: > the RHS matches "foo" [...and then backtracks the LHS until a
: > conjunctional match is found...]
: >
: > Or it's much simpler than that and both of the regexes above just fail
: > because of the greediness of C<+> and there is no intra-conjunction
: > backtracking.
:
: I think we definitely allow intra-conjunction backtracking.
: PGE implements it that way.
That's what I think.
: On a somewhat similar question, what happens with a pattern
: such as
:
: "foobar" ~~ / foo.+? | fooba /
:
: The LHS initially matches "foob", but with backtracking could
: eventually match "foobar". Do the longest-token semantics
: in this case cause the RHS to be dispatched first, even
: though the token declaration of the LHS _could_ match a
: longer token prefix?
Yow. ICATBW. Non-greedy matching is somewhat antithetical to
longest-token matching. But basically it boils down to this:
Does the longest-token matcher ignore the ? and do
foo.+ | fooba
or is there an implicit ordering above and beyond the DFA engine of
foob | fooba ||
fooba | fooba ||
foobar | fooba ||
I think longest-token semantics have to trump minimal matching here,
and my argument is this. Most uses of *? have additional information
on what terminates it, either implicitly in what it is matching, or
explicitly in the next bit of regex. That is, you'd typically see
either
foo\w+? | fooba
or
foo.+? <wb> | fooba
In either case, the clear intent is to match foobar over fooba.
Therefore I think the DFA matcher just strips ? and does its ordinary
character by character match, relying on that extra info to match
the real extent of the quantifier.
Larry
