Camelia, the Perl 6 bug

IRC log for #moarvm, 2013-09-21

| Channels | #moarvm index | Today | | Search | Google Search | Plain-Text | summary

All times shown according to UTC.

Time Nick Message
00:19 benabik joined #moarvm
01:00 benabik joined #moarvm
02:01 FROGGS joined #moarvm
04:09 eternaleye joined #moarvm
04:50 eternaleye joined #moarvm
04:52 cognominal joined #moarvm
05:44 FROGGS joined #moarvm
07:08 cognominal joined #moarvm
10:56 colomon joined #moarvm
12:08 cognominal joined #moarvm
12:16 lizmat_ joined #moarvm
14:02 diakopter hm, the deserializer needs to also de-dupe constant strings so the named-args optimization jnthn & I talked about last week can work
14:08 timotimo oh, you'll have a string table for named arguments and treat them just as integers?
14:13 lizmat wasn't this more general applicable to all string like keys to hashes ?
14:13 lizmat much like p5 handles hash keys >
14:13 lizmat ?
14:13 timotimo probably
14:44 benabik joined #moarvm
14:48 benabik Well, you could check the identity of the string first and then still fall back to value.
14:49 benabik Then you'll get faster access to named values within the same CompUnit at least.  (Well, SC technically but they tend to be related)
14:51 benabik Interning strings would be even faster, perhaps, but then you might end up with a huge uncollectible string pool.
16:48 diakopter lizmat: yep
16:51 diakopter benabik: yep
16:51 diakopter benabik: well it wouldn't be uncollectible if you consider the refs from the compunit's string table as normal gc-tracked references
16:52 benabik diakopter: But you'll want a big global trie of strings to do the de-duplication with.
16:52 diakopter nah I think a hash would be faster if the hash values were precomputed stored in the serialized compunit
16:54 diakopter you agree?
16:54 benabik And just check the dependent SCs for identical hash values when deserializing?
16:55 benabik s/dependent SCs/SC's dependencies/
16:55 diakopter yeah; comparing hash values, length, and spot checking a few characters should make it astronomically unlikely of a false positive
16:56 benabik Well, hash and then check total identity if you want to be paranoid.  If you're going to do string comparison, might as well check the whole thing.
17:11 lizmat joined #moarvm
19:43 timotimo no moarvm commit all day, is moarvm dead? ;_;
19:45 FROGGS no, we're just hard thinking about the outer frame bug :o)
21:25 benabik joined #moarvm
21:29 diakopter timotimo: yes, moarvm died :(
21:30 timotimo oh no, not again
21:30 masak we should have named it after some animal known to survive everything... like Parrot.
22:43 foo_bar_baz joined #moarvm

| Channels | #moarvm index | Today | | Search | Google Search | Plain-Text | summary