[Bug regex/52] Repeated and nested subexpressions (reproducible in most other engines)

["bonzini at gnu dot org" <sourceware-bugzilla at sourceware dot org>]

http://sourceware.org/bugzilla/show_bug.cgi?id=52

--- Comment #11 from Paolo Bonzini <bonzini at gnu dot org> ---
The justification for the "suspended" state is that this would be very
complicated to fix and wouldn't really add anything to the quality of the
implementation.

Even if the "(a(b)*)*" case would not be hard to fix, I'm not sure we can say
the same of the backreference testcase in the RH bug ('(a(b)*)*\2' matched
against 'abab') or the more complicated '(a(b)*)*x\1\2' matched against
'abaxa'.

The RH bugzilla was opened by Eric doesn't really add anything to the urgency
of this bug; Fedora bugs that also exist upstream can be closed liberally, and
that's what I did.

The grep bug on Savannah (https://savannah.gnu.org/bugs/?37737) might add
something, but it is not clear if the user actually encountered it in a
real-world usecase.  The same "bug" is present in hardly every regular
expression matcher, and I would suggest that the Austin group gives more leeway
to implementations.  For example, the following rules could work:

- it is undefined _which_ occurrence of the sub-RE is captured by a
parenthesized group (and matched in backreferences) if the subgroup, or any of
its parents, is quantified with + * {};

- the backreference should only match the empty string only if the
corresponding sub-RE can be empty, or if the corresponding parenthesized group,
or any of its parents, is quantified with * or {0,...}.

-- 
You are receiving this mail because:
You are on the CC list for the bug.