Add an environment= field to pex_binary, and support for environments to the package goal
#17106
Conversation
…al, and add to `pex_binary`. # Rust tests and lints will be skipped. Delete if not intended. [ci skip-rust] # Building wheels and fs_util will be skipped. Delete if not intended. [ci skip-build-wheels]
| @@ -13,4 +13,5 @@ python_test_utils(name="test_utils") | |||
| _docker_environment( | |||
| name="docker_env", | |||
| image="python:3.9", | |||
| python_bootstrap_search_path=["<PATH>"], | |||
There was a problem hiding this comment.
This is to support the case where [python-bootstrap].search_path is being overridden by a Pants developer to point at <PYENV> instead: this overrides it back to something which is valid for this image.
There was a problem hiding this comment.
@stuhood Note that for python bootstrapping, keeping the default value of ["<PYENV>", "<PATH>"] is acceptable, the <PYENV> will be discarded, leaving only <PATH>.
<PYENV> in a manually-specified value is not acceptable.
There was a problem hiding this comment.
<PYENV>in a manually-specified value is not acceptable.
It is what I use by default via a .pants.rc, in order to avoid ever using a system interpreter.
There was a problem hiding this comment.
ack.
Obviously in favour of making the _docker_environment config explicitly valid here. I'm just suggesting highlighting this as not strictly necessary given the config in Pants' pants.toml, so it's clear to readers why it's necessary.
stuhood
changed the title
environment= field to pex_binary, and support for environments in the package goalenvironment= field to pex_binary, and support for environments to the package goal
| @@ -289,9 +290,19 @@ async def create_docker_build_context(request: DockerBuildContextRequest) -> Doc | |||
| ) | |||
|
|
|||
| # Package binary dependencies for build context. | |||
| embedded_pkgs = await MultiGet( | |||
| Get(BuiltPackage, PackageFieldSet, field_set) | |||
| environment_names_per_field_set = await MultiGet( | |||
There was a problem hiding this comment.
Noting that this is happening over in package.py too, it looks like this pattern of requesting a set of objects corresponding to different environments is likely to be a frequent one -- and one that took me a bit of reading to wrap my head around.
Any chance of factoring that out into something reusable?
There was a problem hiding this comment.
Yea, that's probably worth doing. The three use-cases of environments-on-roots so far all use FieldSetsPerTargetRequest ... will see if it can be applied there generically.
|
To make this less error prone, I'm going to focus on #17129 for now. Moving this back to draft. |
pex_binarytargets gain anenvironment=field, which drives the environment in which they are packaged.This was validated working using the instructions from #17096.
[ci skip-rust]
[ci skip-build-wheels]