github.com/moarvm/moarvm | IRC logs at colabti.org/irclogger/irclogger_logs/moarvm
Set by AlexDaniel on 12 June 2018.
MasterDuke would this be useful (e.g., faster reading precomp files) or will this be at a lower level than moarvm uses? www.phoronix.com/scan.php?page=new...all-Review 08:26
nine That's meant for _small_ files, i.e. reading through lots and lots of entries in /proc or /sys. We don't read precomp files in one piece. For some, we never even read the whole thing (most prominently the setting). 08:48
We read the header to get the checksum and dependency information and mmap the rest for MoarVM to lazily deserialize the accessed parts 08:49
MasterDuke hm, didn't realize we don't read always read all of a precomp file 09:15
lizmat bare startup time would be much bigger then :-) 09:16
nwc10 It hates me 14:26
I seem to be deep in a yak forest of uknown unknowns 14:27
(ie al the assumptions we didn't even realise we were assuming)
gen/moar/stage2/MASTNodes.nqp in particular hates me 14:33
it's just thrown me a second yak
timotimo oof 14:39
nwc10 so... 14:41
we like to keep asking iterators "Are you at the end?" and assume that end is sticky
and we need to handle "fetch/store into a hash that is actually a fetch" at the same time as there is an iterator 14:43
and now my paranoia checks have just discovered that we ask an iterator "are you at the end?" and it's not even for the current hash
the answer is probably actually "Yes, I'm at the end"
but I want to figure out why it's passing the wrong hastable pointer to the function
ie I'd like to figure out if there's an actual bug before working around it 14:44
(Work around is to move the assertion that it's the correct hash *after* the test for "am I at the end?")
(and to only fail an assertion if the answer is "not yet" and the hashtable pointer is wrong)
but by now, I have a really good hash fsck function :-/ 14:57
timotimo do you have visualizers? %) 15:23
nwc10 I'm not quite sure what you mean by visualizers, but it will dump all the keys, their hashes, their offset from the ideal bucket, whether the key is in the nuresery (1st or 2nd) or gen2, and a whole bunch of "this would explode because it's ..." (such as "in frompsace") 15:26
timotimo but is it a nice-to-look-at output? ;) 15:33
lizmat feels for nwc10, having had a similar experience trying to optimize Match objects 17:19
nwc10 virtual hugs are certainly appreciated. But have you got any silver bullets? :-) 17:20
(to be clear, for the bugs. It's not *that* bad)
lizmat the only silver plated bullet I have, is stepping away and sleeping a night on it 17:24
nwc10 this was my plan 17:25
it's cool enough to attack the garden for a bit
going to Mistlebach tomorrow - might not get any more time until Tuesday 17:26
lizmat sounds idyllic 17:28
nwc10 lizmat: I wouldn't quite say that - we're visiting (one of) Andrea's aunt and uncle. It should be fine 20:27
and they have to be somewhere else in the evening, so there's no confusion about staging too long 20:28