This page serves as a list of current (June 2003) problems with Tcl's regexp library.
So, what proposals have been submitted as TIPs to address the deficits?
Related question: who understands the code enough to be persuaded to write a TIP to address any of these items. (Since, without an implementor, any TIP would only languish)
[Compliments to Vince for this summary.]
DKF: On the last point (greedy/non-greedy) the rules are clear in a certain sense. That sense is that the first quantifier syntactically encountered in a RE determines whether the RE as a whole is greedy or not, and all quantifiers in the RE are either greedy or not greedy. There are no mixed-greediness REs, which is definitely unlike Perl. On the other hand, I've no idea what the automata-theoretic basis for a mixed-greediness RE is; the Perl code is widely accused (probably rightly) of being a horrible hack.
I should note here that there is an exception to the rule on quantifier mixing, and that is for (non-zero-width) lookahead assertions, which are always non-greedy. But they're handled effectively by recursively calling the RE engine internally, and the boundary between which automata are being used is fairly obvious. They're also not very efficient, and you can't use them for subexpression capture.
Vince: Unfortunately, it's very hard to submit a TIP to address any of the above issues (except the addition of a function to the stubs table, which I could happily submit a TIP for), because the code is so incomprehensible that I don't know who (beyond Henry Spencer) would actually be in a position to implement any changes. That's partly why I put together the above list, to see whether discussion here would point to a way forward/out of this mess ;-).