Poetry: Support of manylinux2010

[PgLoLo]

• OS version and name: Ubuntu 18.04
• Poetry version: 0.12.17

Issue

As far, as I understand currently poetry incorrectly processes manylinux2010 dependency tag (which can be seen in this issue from TensorFlow project).

I am not sure, that problem is on the poetry side, but from the fact that PipEnv previously had this problem as well and does not have in recent versions, I conclude that that has to be solved in the poetry as well.

[PgLoLo]

Currently, it is a very big restriction. Some libraries pretty much can not be installed with poetry.

[sdispater]

Like I said in this comment https://github.com/sdispater/poetry/issues/1330#issuecomment-538763873 this is not on Poetry's end and it seems it has been fixed in tensorflow (see https://github.com/tensorflow/tensorflow/pull/32758).

So the next release of tensorflow should work and have its dependencies properly specified.

[rbong]

Isn't the Tensorflow fix just a workaround though?

According to this post pipenv was trying to install the MacOS version of Tensorflow since it didn't support manylinux2010 tags (PEP 571). Pipenv managed to fix it on their end by adding manylinux2010 support, see this comment to see that it's fixed on master.

Even though they are able to add a workaround on their end, the last official position still that this issue will be resolved by fixing it in poetry. See this comment.

So couldn't this happen with another package without adding support for manylinux2010 tags, or is poetry's issue different? Are there any plans to add manylinux2010 support, independent of whether the Tensorflow issue is fixed?

Sorry if I got something wrong, I just want to make sure I understand the position on manylinux2010 tags.

[SimonSelg]

Hey there,
this has indeed been fixed on tensorflows end, but only starting from 2.1rc0.
The same issue is present for tensorflow 1.15.0. Is there any workaround or plans to make tensorflow 1.15 work with poetry?

[kafendt]

Same problem here we depend on tensorflow 1.15.0 without this I guess I'll have to try pipenv instead :(

[leesei]

Same problem here we depend on tensorflow 1.15.0 without this I guess I'll have to try pipenv instead :(

pipenv's latest build on PyPI faces the same issue.
But the master branch have it fixed (I haven't personally tested).