feat: expose python package entrypoints via an opaque struct #1189

aignas · 2023-04-30T09:02:08Z

Fixes #958.

For bzlmod users we did not have a working alternative for the entry_point macro for the non-bzlmod users. The reason it is tricky is that in order to know the name of the entry_point available to the user one needs to download the Python package and parse the metadata. This works around the limitation by exposing the list of entrypoints back to the hub repo as an opaque struct, which the users can use in their code.

Summary:

Added support for users setting incompatible_generate_aliases to True. They can use the bin struct exposed via @<hub_rep>//<dep>:bin.bzl.
Added support for bzlmod users not setting incompatible_generate_aliases. They can use it via use_repo("<hub_repo>_<dep") and then importing the same struct via @<hub_repo>_<dep>//:bin.bzl.
The legacy behaviour of the entry_point macro is unchanged for the remaining users for now.

Design notes:

Expose the struct in separate .bzl files in order to have no eager fetches.
Exposing it via a struct will give users an error message if the target with the specified name does not exist and it will tell the available struct attributes.
The inspiration comes from rules_js which @alexeagle has pointed to.

Open questions:

I know that @groodt was toying with an idea of removing the repository macros (requirement, et. al) and I am wondering if this may be the last piece that we need in order to give our users an alternative.
If the design looks good, I am wondering about the documentation we should have. Is the current addition to the bzlmod example sufficient?

aignas · 2023-04-30T09:29:36Z

Depends on #1190 being merged to main. For now I have merged the contents of that PR into this branch to ensure that CI is happy.

rickeylev · 2023-04-30T22:23:46Z

python/pip_install/pip_repository.bzl

@@ -475,6 +495,7 @@ def _pip_repository_impl(rctx):
        ]),
        "%%ANNOTATIONS%%": _format_dict(_repr_dict(annotations)),
        "%%CONFIG%%": _format_dict(_repr_dict(config)),
+        "%%ENTRYPOINT_REPO%%": "//{pkg}" if rctx.attr.incompatible_generate_aliases else "_{pkg}//",


Where is this %ENTRYPOINT_REPO% used? I'm not seeing it.

rickeylev · 2023-04-30T22:26:58Z

python/pip_install/pip_repository.bzl

@@ -475,6 +495,7 @@ def _pip_repository_impl(rctx):
        ]),
        "%%ANNOTATIONS%%": _format_dict(_repr_dict(annotations)),
        "%%CONFIG%%": _format_dict(_repr_dict(config)),
+        "%%ENTRYPOINT_REPO%%": "//{pkg}" if rctx.attr.incompatible_generate_aliases else "_{pkg}//",


Something about this looks weird.

The variable is named "repo", but "//{pkg}" isn't a repo-name.

The "_{pkg}//" value: I dont't repos can start with underscore? I got an error when I tried to do that recently. Shouldn't it also have "@" in it?

rickeylev · 2023-04-30T22:31:01Z

python/pip_install/tools/wheel_installer/wheel_installer.py

+def _generate_bin_bzl_contents(entrypoints: dict[str, str]) -> str:
+    return textwrap.dedent(
+        """\
+        # Autogenerated by wheel_installer.py


Suggested change

# Autogenerated by wheel_installer.py

# Autogenerated by wheel_installer.py#_generate_bin_bzl_contents

Thanks for adding the "generated by" text. Those comments have been very helpful as I've tried to trace code around. I just a small suggestion to point directly to the part doing it. I'm also +1 on including any other extra info in the comment that makes it easier for maintainers.

rickeylev · 2023-04-30T22:32:06Z

python/pip_install/tools/wheel_installer/wheel_installer.py

+        # Autogenerated by wheel_installer.py
+        bin = struct(**{})
+        """.format(
+            repr({name: Label(target=target) for name, target in entrypoints.items()})


Please use a dedicate formatting function instead of relying on repr(). The output of Python's repr isn't guaranteed to be stable between Python releases, nor is it guaranteed to be compatible with Starlark.

rickeylev · 2023-04-30T22:35:06Z

python/pip_install/tools/wheel_installer/wheel_installer.py

+        # Autogenerated by wheel_installer.py
+        bin = struct(**{})
+        """.format(
+            repr({name: Label(target=target) for name, target in entrypoints.items()})


Please sort the values on output. This just helps avoid any determinism issues from the dict ordering. While Python dicts now have guaranteed insertion order, it's easy for some other code to change the insertion order of the dict (e.g. someone calls set on some values earlier). We can avoid that problem by just sorting the output.

rickeylev · 2023-04-30T22:38:10Z

examples/bzlmod/BUILD.bazel

+    # This is using the struct defined by the spoke repo 'yamllint' and it is
+    # re-exported by the hub repo named 'pip'. This allows bzlmod and
+    # non-bzlmod users to access the entry point targets.
+    actual = yamllint_bin.yamllint,


I don't really follow this part -- why does it have to be put into a bzl file and loaded like this? This is pointing to a target, right? So why can't it just be "@pip//yamllint:bin"?

(I'm guessing the answer has something to do with bzlmod, repo mapping, and repo visibility, but I'd like to understand more).

rickeylev · 2023-04-30T22:42:12Z

python/pip_install/pip_repository.bzl

+        # Prefix the re-export with a dep name in order to avoid collisions
+        # during loading of multiple bin structs at the same time.
+        bin_content = """\
+\"\"\"Autogenerated by 'pip_parse'.\"\"\"


nit: be a bit more specific in what generated it. "pip_parse" is a bit vague and there's several pieces to it.

rickeylev · 2023-04-30T22:42:54Z

python/pip_install/tools/wheel_installer/wheel_installer.py

@@ -32,6 +33,14 @@
 from python.pip_install.tools.wheel_installer import namespace_pkgs, wheel


+@dataclass
+class Label:


Can we call this LabelWrapper or something? I just want it to be easier to tell when it's Python code calling the dataclass named Label vs generated bzlcode in the Python code calling the bzl Label. The behavior of Label() is confusing enough; adding "Which label object is this?" is going to make it very confusing.

Actually, can we remove this dataclass entirely? It looks like it's just used once?

rickeylev · 2023-04-30T22:44:10Z

python/pip_install/tools/wheel_installer/wheel_installer.py

@@ -32,6 +33,14 @@
 from python.pip_install.tools.wheel_installer import namespace_pkgs, wheel


+@dataclass
+class Label:
+    target: str


nit: "target_name" instead of just "target". A Target is a 3-tuple of (workspace, package, and name), and this is only the name portion. That, or add a comment to indicate it's only the name portion of a target.

rickeylev · 2023-04-30T22:50:28Z

I am wondering about the documentation we should have.

Probably also update the "entry_point" function docs. If I'm trying to use an entry_point in a certain way, I'll consult its reference documentation to see how to use it. I'll look at examples when I'm first using a library, but after I use it a few times, I'm unlikely to look at the examples (they'll just tell me most of what I know).

aignas · 2023-05-01T12:47:45Z

@rickylev, thanks for the great comments. It seems that the documentation for the bzlmod extensions is non-existent right now so I added minimal text to the right place even though it is not generated right now.

It will take me some time to fully understand how to best add the extensions.bzl documentation generation. I'll see how far I can get and I'll maybe include to this PR.

aignas · 2023-05-01T13:29:36Z

So I have generated some docs for bzlmod extensions specifically, but I am not sure if how this is turning out is the best way it can be. See the link here.

It may be a good start, but without stardoc generating docs automatically for module_extension, there is just no good way to have consistent docs for this. Putting the docs to the non-bzlmod macros may be as, if not more, confusing.

aignas · 2023-05-01T13:35:16Z

Depends on #1190, but marking as ready for review as I have addressed the first round of comments.

…tc (#1190) It seems that the macros for specifying the requirements break when the user starts using `incompatible_generate_aliases=True`. This PR fixes this. Testing done: 1. Modify the example: ``` $ git diff diff --git a/examples/bzlmod/MODULE.bazel b/examples/bzlmod/MODULE.bazel index ce91228..1750210 100644 --- a/examples/bzlmod/MODULE.bazel +++ b/examples/bzlmod/MODULE.bazel @@ -26,6 +26,7 @@ register_toolchains( pip = use_extension("@rules_python//python:extensions.bzl", "pip") pip.parse( name = "pip", + incompatible_generate_aliases=True, requirements_lock = "//:requirements_lock.txt", requirements_windows = "//:requirements_windows.txt", ) ``` 2. Run `bazel build ...` and check that it is still working. I noticed this when working on #1189 and creating a separate PR for easier cherry-picking if we wanted to make a patch release which includes this. I am not sure how I could make an automated test for this other than creating a separate example.

@alexeagle

* Added support for users setting `incompatible_generate_aliases` to `True`. They can use the `bin` struct exposed via `@<hub_rep>//<dep>:bin.bzl`. * Added support for bzlmod users not setting `incompatible_generate_aliases`. They can use it via `use_repo("<hub_repo>_<dep")` and then importing the same struct via `@<hub_repo>_<dep>//:bin.bzl`. * The legacy behaviour of the `entry_point` macro is unchanged for the remaining users. Design notes: * Expose the struct in separate `.bzl` files in order to have no eager fetches. * Exposing it via a struct will give users an error message if the target with the specified name does not exist and it will tell the available struct attributes. * The inspiration comes from `rules_js` which @alexeagle has pointed to. Fixes bazel-contrib#958.

aignas · 2023-05-04T06:43:58Z

Whilst trying to add tests for this I have noticed that it would be better to wait for a few PRs already in the queue, so that I can add better tests. What is more, maybe splitting the documentation for the extensions.bzl into its own PR would be a good idea.

aignas · 2023-05-08T13:52:15Z

I am wondering if this is really the best way to solve this. The reason for it is that if the user would like to use a select statement to select the entrypoint based on a configuration option, it would inevitably eagerly fetch all of the repos used in the select statement. If we had an entry point macro that is in rules_python, it might be a different story.

aignas · 2023-05-12T05:32:45Z

I am going to close this as #1220 is the approach I would recommend for now.

aignas requested review from hrfuller and rickeylev as code owners April 30, 2023 09:02

aignas marked this pull request as draft April 30, 2023 09:22

aignas mentioned this pull request Apr 30, 2023

fix(bzlmod): correctly template repository macros for requirements, etc #1190

Merged

rickeylev requested changes Apr 30, 2023

View reviewed changes

aignas force-pushed the feat/opaque_entry_point branch from 2387137 to e16e3a9 Compare May 1, 2023 13:26

aignas marked this pull request as ready for review May 1, 2023 13:34

aignas mentioned this pull request May 1, 2023

Documentation for bzlmod extensions #1178

Closed

aignas and others added 11 commits May 4, 2023 15:41

comment: format the struct creation in a separate function

23ab27a

fixup: previous

93d79d7

fixup! fixup: previous

c3f377c

comment: get rid of the unused tmpl var

c636f1f

doc: Add documentation to the pip_parse extension

89f962b

doc: add a not on why the design is the way it is.

5d3c38b

comment: improve variable naming

6cb7e85

doc: add a new extensions.md file for bzlmod extensions

d972879

docs: Use subsections for bzlmod tag classes

80b80c6

fixup: use different type annotations

cef4507

aignas marked this pull request as draft May 4, 2023 06:42

aignas force-pushed the feat/opaque_entry_point branch from b4f7025 to cef4507 Compare May 4, 2023 06:42

aignas mentioned this pull request May 11, 2023

feat(bzlmod): support entry_point macro #1220

Merged

aignas closed this May 12, 2023

aignas deleted the feat/opaque_entry_point branch May 13, 2024 06:48

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat: expose python package entrypoints via an opaque struct #1189

feat: expose python package entrypoints via an opaque struct #1189

aignas commented Apr 30, 2023

aignas commented Apr 30, 2023

rickeylev Apr 30, 2023

rickeylev Apr 30, 2023

rickeylev Apr 30, 2023

rickeylev Apr 30, 2023

rickeylev Apr 30, 2023

rickeylev Apr 30, 2023

rickeylev Apr 30, 2023

rickeylev Apr 30, 2023

rickeylev Apr 30, 2023

rickeylev commented Apr 30, 2023

aignas commented May 1, 2023 •

edited

Loading

aignas commented May 1, 2023

aignas commented May 1, 2023

aignas commented May 4, 2023

aignas commented May 8, 2023

aignas commented May 12, 2023

	# Autogenerated by wheel_installer.py
	# Autogenerated by wheel_installer.py#_generate_bin_bzl_contents

feat: expose python package entrypoints via an opaque struct #1189

feat: expose python package entrypoints via an opaque struct #1189

Conversation

aignas commented Apr 30, 2023

aignas commented Apr 30, 2023

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

rickeylev commented Apr 30, 2023

aignas commented May 1, 2023 • edited Loading

aignas commented May 1, 2023

aignas commented May 1, 2023

aignas commented May 4, 2023

aignas commented May 8, 2023

aignas commented May 12, 2023

aignas commented May 1, 2023 •

edited

Loading