i have a question about pex v1. when i run this co...
# pex
a
i have a question about pex v1. when i run this command line to resolve
tensorflow==2.2.0
against a
--find-links
repo with both
tensorflow-2.2.0_....whl
as well as
tensorflow-2.2.0+xxx3yyyy10zzz0_....whl
, it reliably resolves the second one with the
+tag
, not the
tensorflow-2.2.0
wheel, despite requesting
tensorflow==2.2.0
. how is this possible?
if the answer is "this is fixed in pex 2", that would be a great result
let me check
so it is!!
this is not a problem, we can move to pex 2
Copy code
Skipping link: none of the wheel's tags match: cp36-cp36m-linux_x86_64: <https://some-url.com/python/wheels/tensorflow-2.2.0%2Btwtr2cuda10mkl0-cp36-cp36m-linux_x86_64.whl>
beautiful
bumping this because i believe i also may have found another issue: where specifying
--platform=linux-x86_64-cp-36-m
on the pex CLI is silently ignored if the current platform is not linux and
-o <file>
is not specified. separately, is it actually correct to be resolving
tensorflow==2.2.0
to
tensorflow==2.2.0+asdf
? that seems like quite surprising behavior, because if anyone wants to upload a tagged
+asdf
version for testing, the
tensorflow==2.2.0
requirement suddenly redirects to the new tagged version. is there a workaround for this behavior so that tagged versions aren't resolved?
and to be clear, we wouldn't want this to be a global behavior for enabling or disabling tagged releases. it would just be respecting the assumption multiple people have made that the tagged release should not have been preferred over the non-tagged one in this scenario.
it could be a single option that toggles whether
tensorflow==2.2.0
prefers
2.2.0+asdf
over
2.2.0
e
Do you mind posting an issue with exact commands? Good to include would be the usual, -v*9.
a
absolutely! will do!
ok, the "another issue" i found was not an issue, or at least, i can't repro it. i will create an issue about the
+tagging
now.