jcea has quit [Ping timeout: 260 seconds]
derpydoo has quit [Quit: derpydoo]
lritter has joined #pypy
lritter has quit [Quit: Leaving]
dmalcolm_ has joined #pypy
dmalcolm has quit [Ping timeout: 264 seconds]
antocuni_ is now known as antocuni
otisolsen70 has joined #pypy
epony has quit [Ping timeout: 252 seconds]
epony has joined #pypy
Dejan has joined #pypy
<
cfbolz>
randomly generating traces to feed into the optimizer just found its first bug 😊
<
Dejan>
will 3.11 make PyPy faster too?
<
cfbolz>
Dejan: your question is under-specified, what do you mean exactly?
<
Dejan>
3.11 has several optimisations - i wonder how much will they affect PyPy
<
cfbolz>
Dejan: but pypy is not CPython, so the fact that CPython 3.11 has optimizations does not immediately affect us
<
Dejan>
that is why i asked the question - is there a chance some of them will actually do well on PyPy too?
<
cfbolz>
if we port the same approaches?
<
cfbolz>
that's unclear, basically
<
cfbolz>
those are interpreter optimizations
<
cfbolz>
so for pypy they would affect the pre-jit speed
jcea has joined #pypy
epony has quit [Quit: QUIT]
<
cfbolz>
lots of bugs
<
cfbolz>
moooooore bugs
derpydoo has joined #pypy
<
cfbolz>
nothing totally serious so far, but still
<
cfbolz>
why did we never optimize random traces before?
derpydoo has quit [Quit: derpydoo]
Vishal36 has joined #pypy
Vishal36 has quit [Client Quit]
<
fijal>
cfbolz: I think the main reason was that there is a bunch of assumptions about the traces that are implicit
<
fijal>
or one of the reasons
<
cfbolz>
fijal: no, I am not finding this kind of bug
<
cfbolz>
instead real actual logic bugs in the (int) parts of optimizeopt
<
fijal>
I don't know then
<
fijal>
maybe a much better answer :-)
<
cfbolz>
fijal: I'm only generating int ops for now, maybe that focus helps
<
cfbolz>
I really want a test case minimizer now though
epony has joined #pypy
<
Alex_Gaynor>
cfbolz: I wonder if it'd make sense to hook this up to libFuzzer or something like that -- it has good tools for minimization.
<
cfbolz>
Alex_Gaynor: heh, did I accidentally send out the FuzzMan signal or something? ;-)
<
Alex_Gaynor>
hehe, I don't do much pypy stuff these days, but I'm still here!
<
cfbolz>
Alex_Gaynor: I think I need to finish a bunch of things first, then we could try
<
cfbolz>
some intermediate support missing, basically
<
Alex_Gaynor>
you'd need a way of encoding/decoding ops from bytes.
<
cfbolz>
Alex_Gaynor: we have a pretty-printer and a parser
<
Alex_Gaynor>
that's a string format, yeah?
<
Alex_Gaynor>
as in, human readable
<
cfbolz>
ok, too hard to get the syntax right for a fuzzer maybe
<
Alex_Gaynor>
I'm sure libFuzzer would figure it out eventually, might be less efficient than a binary format, but libfuzzer is like magic.
<
cfbolz>
I mean we do have a random trace generator that takes a source of randomness (=arbitrary bytes) and turns them into operations :-P
derpydoo has joined #pypy
otisolsen70 has quit [Quit: Leaving]