<antocuni>
we're already seeing mojo run untyped code 8x-ish faster than cpython. Just because we have a compiler instead of an interpreter. We're not doing anything fancy.
<antocuni>
"""
<antocuni>
I am VERY skeptical about this statement. If they are getting 8x faster without doing "anything fancy", then they are not implementing Python semantics at all
<nimaje>
hm, I throught they wanted to use cpython to execute python code and integrate their language with cpython? do they compile python to a cpython c-extension and run that?
<antocuni>
from what I understand, they have two ways
<antocuni>
1. import CPython modules by linking to libpython.so: this is fully compatible of course, and it works right now
<antocuni>
2. compile Python code directly. From what I understand, this is just vaporware, it's not even close to exist
<mattip>
vaporware has less bugs
<mattip>
but I doubt they are totally vaporware, they must have some MVP
<antocuni>
indeed, it's not clear what is the status
<antocuni>
in some section they say that "high level classes" and "full python dynamism" is yet to be implemented
<antocuni>
but then they say that they can compile python 8x faster
<antocuni>
so I don't really know what it means
slav0nic has joined #pypy
jcea has joined #pypy
lehmrob has quit [Quit: Konversation terminated!]
slav0nic has quit [Ping timeout: 248 seconds]
_whitelogger has joined #pypy
slav0nic has joined #pypy
<phlebas>
antocuni: doesn't it just mean they picked a benchmark that is 'static enough' to be analysed and compiled to reasonably fast code? I mean, you can write benchmarks that happen to be translatable with rpython and run them, too, right?
<antocuni>
yes, I think you are right
<antocuni>
but then saying "we are 8x faster" is really a lie :)
<antocuni>
basically, they don't explain why they should be faster than Cython