Introduce pex3 venv create.
#2140
Conversation
|
I know this PR is in a draft but I just wanted to say that instead of a |
Yeah, I'm not super happy about a The thing is, the code used is >50% the same and the use cases I'm aware of are similar; so I've smooshed them together. The motivating use case for Saying that out loud, perhaps |
This new subcommand can create either a venv or just populate a flat `sys.path` directory entry (ala `pip install --target`) given a set of requirements to resolve, potentially from a lock or an existing PEX file, but otherwise from indexes and find links repos. Unlike the sibling `venv` `pex-tool` subcommand, the target can be selected and, in the flat `sys.path` directory entry case, it can be a foreign platform. Fixes pex-tool#1752 Fixes pex-tool#2110
|
Alright, reviewers thanks in advance - this is large. It's mainly just the new sub-command and its integration tests, but there was some code moved around and a small bit of new code in Creating a venv directly from a lock (subset) was what I needed for a-scie/lift, but it was easy enough to roll in features satisfying a few Pex issues as well as the unrealized goal @benjyw had over in Pants of having the lambda function package results be exactly what AWS / GCP / Azure call for in both deployed zip and, in the case of AWS, Lambda layers. |
| from typing import DefaultDict, Iterable, Iterator, List, Optional, Tuple, Union | ||
|
|
||
|
|
||
| def find_dist( |
There was a problem hiding this comment.
This block of code to ensure pip is installed moved from the existing venv PEX_TOOL here: https://github.com/pantsbuild/pex/pull/2140/files#diff-44e8fd52cd3a63a2d3f2fc516418c2aa55e8ee96f6eb82b2186765a8b853922fL83-L180
There was a problem hiding this comment.
I created a pex from this commit and it seems to work fine for me. I tested both venv and flat layouts from a lockfile with a local interpreter.
I don't quite understand the --force flag here with the venv layout. If I provide --force and the directory exists, is the venv destroyed and recreated? I my understanding is that the pex-tools command deletes the venv and recreates it entirely.
Correct.
That is not correct when doing partial exports ( These are semantics I'm not convinced about, although a hard line is supporting the srcs / deps split. If you have alternate semantics that make more sense and support the split let me know. |
| @@ -549,6 +549,7 @@ def _populate_sources( | |||
| dst=dst, | |||
| exclude=( | |||
| "__main__.py", | |||
| "__pex__", | |||
There was a problem hiding this comment.
A venv PEX never needed the magic __pex__ import hook, which won't work anyhow since a venv PEX has no Pex .bootstrap code carrying the import hook implementation code.
This new sub-command can create either a venv or just populate a flat
sys.pathdirectory entry (alapip install --target) given a set ofrequirements to resolve, potentially from a lock or an existing PEX
file, but otherwise from indexes and find links repos. Unlike the
sibling
venvpex-toolsub-command, the target can be selected and,in the flat
sys.pathdirectory entry case, it can be a foreignplatform.
Fixes #1752
Fixes #2110
Fixes #2111