Is there a plan to stop supporting py27 in pex?
# development
e
Is there a plan to stop supporting py27 in pex?
h
Nope, quite the opposite. @enough-analyst-54434 has done work to support a vendored pip, to retain py27 support, alongside a resolved, current pip, to support the latest pip features on newer interpreters
e
interesting. Coming from the developer side we definitely lose out on some nice newer stdlib and syntax stuff from python3
e
I don't break users TM
Users trump devs
e
👌 good to know the philosophy there
Though releasing a new major version which dropped support for py27 doesn’t seem like it would break existing users
e
I had looked up the Red Hat end of support and forget, but I think it's 2025 and so wanted to coincide with that
👍 1
Ah, I think that's not true if the tool is half baked. PEX entered a period of half baked after it left Wickman's hands . It's approaching fully baked (~no known bugs).
There was lots of ill thought out stuff added 2013-2018 or so.
Has taken a while to sort back out.
e
makes sense
e
One thing that's interesting to do is look at shard timings for Pex CI.
e
are there distributions of many runs that github ci provides?
or do you spot check individual runs?
e
No, I just spot check and that backs up what I see on my machine (I have the full complement of PyPy and CPython supported versions). Python 2.7 is still fastest.
e
that would not be my initial expectation given all the perf improvements touted in python3
rewriting the container types and such
e
Yeah, perf improvements relative to prior Python 3's definitely.
Not against 2 though for Pex / Pip workloads anyways.
😲 1
e
I guess once you fix too many bugs you start to add overhead
there some local minima there lol
e
Python 3 up through 3.7 is pretty atrocious, getting better steadily since then though. 3.11 is now pretty close for Pex.
e
so basically python2.7 is dead long live python2.7. learned something today
e
I am not too impressed by the language features and shiny toys. The security fixes have my attention of course, but that's the user's choice.
e
heh yeah I don’t use walrus or anything. pathlib is nice
f-strings. I guess there is some stuff I like
e
That all is nice but I think it's like 1% nice. So noise. If Pex could leverage asyncio or something like that that makes a real difference to users via perf I'd sing a different tune.
1
👌 1