jinsun has quit [Read error: Connection reset by peer]
derpydoo has joined #pypy
itamarst has quit [Quit: Connection closed for inactivity]
the_rat has quit [Server closed connection]
the_rat has joined #pypy
glyph has quit [Remote host closed the connection]
glyph has joined #pypy
derpydoo has quit [Ping timeout: 252 seconds]
fijal has quit [Server closed connection]
fijal has joined #pypy
eamanu has quit [Server closed connection]
eamanu has joined #pypy
jerith has quit [Server closed connection]
jerith has joined #pypy
Alex_Gaynor has quit [Server closed connection]
Alex_Gaynor has joined #pypy
idnar has quit [Server closed connection]
idnar has joined #pypy
Techcable has quit [Server closed connection]
Techcable has joined #pypy
habnabit_ has quit [Server closed connection]
habnabit_ has joined #pypy
sam_ has quit [Server closed connection]
sam_ has joined #pypy
ronan has quit [Server closed connection]
ronan has joined #pypy
nikolar has quit [Server closed connection]
nikolar has joined #pypy
dustinm has quit [Server closed connection]
dustinm has joined #pypy
itamarst has joined #pypy
mgorny has quit [Server closed connection]
mgorny has joined #pypy
jcea has joined #pypy
ruth2345345 has quit [Server closed connection]
jcea has quit [Ping timeout: 248 seconds]
ruth2345345 has joined #pypy
* arigato
reads https://tonybaloney.github.io/posts/python-gets-a-jit.html and sighs. The basic idea is the same as papers I've read ages ago that were already very old at that point, but apparently "It’s an idea only proposed recently in 2021". Also it gets 2-9% speedups but "it opens the door to some huge optimizations", which, let's say, I have some doubts about
<arigato>
if I remember correctly, it's one of the mainstream approaches when doing partial evaluation
<korvo>
Yeah. I remember reading the copy-and-patch paper (something "stencils"?) and being disappointed that they didn't know the history.
<arigato>
at least the paper seems to have new results because they are using it on new hardware where the trade-offs are different from the old times' hardware. The idea of applying the same technique to Python sounds to me exactly like "we get 2-9% now, and we'll get a lot more later, I'm sure!"
<arigato>
whereas to me, it sounds like it's Pyrex all over again (maybe faster to produce machine code (which means less flexible in the future))
<korvo>
In logic we say sometimes "one person's modus ponens (P -> Q) is another person's modus tollens (~Q -> ~P)"
<korvo>
My main takeaway from the Faster CPython numbers is that CPython's per-bytecode dispatch costs must be enormous, given how much difference it makes to fuse one bytecode into two.