Re: Bitset statistics

[Akim Demaille]

Included below my first answer, that, according to a guy named Mailer
Daemon, you never received.

----------------------------------------------------------------------

>>>>> "Michael" == Michael Hayes <address@hidden> writes:
Michael> Akim Demaille writes:
>> I'm very interested by your work.  I maintain Bison which spends a
>> significant time in handling sets.

Michael> Thanks.  I've had an underwhelming response from the GCC
Michael> mailing list so far.

Yep, I've seen that.  To be honest, that's a problem for me: I am
looking for a solution for which I'm sure there will be maintenance.
Being hosted by GCC is a strong evidence it would be.

>> Would it be possible (i), to have your implementation, and (ii)
>> your input as to we should steal GCC's implementation, or yours
>> (the last question includes ``is it used somewhere, and do you plan
>> to maintain it'').

Michael> (i) Sure.

Thanks a lot!  It looks cute.  A quick browse doesn't seem to reveal
you left the statistics code in there, did you?  I'd be interested to
know how bison uses its bitsets.

Michael> (ii) I have rewritten the code to work with libiberty but
Michael> have not fully retested it yet.  It is not used anywhere else
Michael> yet but hopefully it will be picked up for use with GCC since
Michael> efficient bitsets are going to become increasingly more
Michael> important.  With any luck the code should be maintenance free
Michael> but I'm willing to support it.

Great!  So I think we will soon integrate your work in Bison.  But
again, seeing it accepted by GCC would certainly speed the process up :)

Michael> I have attached a copy of my latest patch.  I would
Michael> appreciate any feedback to improve the consistency of the
Michael> interface, say for example the naming of functions, and for
Michael> suggested improvements.

OK.  Currently we face other challenges in Bison, so I can't tell you
_when_ we will integrate it.  But we will.

----------------------------------------------------------------------

Today I meant to give a first stab to this.  It pleases me a lot.  I
still have the concerns I described above (and found the stats BTW :),
so its admitance in the GCC club is still something I would like to
hear about.

Nevertheless, a couple of issues: could you sign the papers for Bison?
Given that this code is not part of Liberty or GCC etc., I don't think
I can't accept it now.  And, why the *(& "^&% heck do sbitset have to
be bitset_zero before being usable?  I lost quite some time trying to
understand what was happening: bison was badly broken :(  wit
bitset_zero it works fine but...

Thanks!