Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Support selecting Python version via tool.poetry.dependencies.python #260

Closed
edmorley opened this issue Sep 3, 2024 · 4 comments
Closed
Labels
enhancement New feature or request semver: major

Comments

@edmorley
Copy link
Member

edmorley commented Sep 3, 2024

Initial support for Poetry was added in #7, and included support for bootstrapping Poetry and then using it to install app dependencies.

This issue is for adding support for controlling the Python version via the tool.poetry.dependencies.python TOML field in pyproject.toml when using Poetry. This will be in addition to the existing ability to use runtime.txt, and the planned ability to use .python-version (see #6).

An example Poetry config created using poetry init with when using Poetry 1.8.3 and Python 3.11 contains:

[tool.poetry.dependencies]
python = "^3.11"

However, the next release of Poetry is due to change the default constraint marker from ^ to >= (see python-poetry/poetry#9558), which will give:

[tool.poetry.dependencies]
python = ">=3.11"

It's worth noting that:

  • Poetry's dependency syntax doesn't use PEP-440 style versioning, but instead something closer to the Node.js style semver syntax. This means it differs from the syntax used in requirements files and in other non-Poetry parts of pyproject.toml (including the project.requires-python field, which is what uv uses), which makes things more complicated both for the buildpack and for user UX (for example both syntaxes have a specifier that includes the tilde character, but they do drastically different things for some edge cases). See: https://python-poetry.org/docs/dependency-specification/
  • The Python version used in the output of poetry init uses the version of the local Python version to set the minimum Python version.
  • If the python field isn't specified at all, Poetry defaults to a wide range (currently >=2.7,<2.8 || >=3.4), which then breaks installing any packages that have a smaller compatibility range than that (ie: any package that only supports Python 3) - since Poetry treats its python field as a "this project must be compatible with all of these Python versions" value. As such, omitting python really isn't viable for users - so it's going to be set most of the time.
  • Both the "^3.11" and ">=3.11" forms allow higher major versions of Python such as 3.12 or 3.13, which can include breaking changes. Whilst supporting a range of versions makes sense for a library, for applications these unbounded ranges can cause issues (as we've seen with the Node.js buildpacks over the years), and so ideally we wouldn't want apps to use these forms.

All of the above means that adding support for tool.poetry.dependencies.python is going to mean making compromises in one area or another sadly, since we either have to:

  1. Explicitly support "unsafe" version ranges such as those above. (Pros: Compatibility with the default output of poetry init. Cons: Breaking changes when new major Python versions are released + encourages environment drift between CNB and local development environments - and even from one developer's machine to another.)
  2. Ignore the tool.poetry.dependencies.python field (either completely, or perhaps only if it uses an unsafe range) and instead install the buildpack's curated default Python version. (Pros/cons: Pretty much the same as (1), apart from the curated Python version perhaps being marginally more compatible with packages in the wild, since we wait a couple of months before making new Python versions the default.)
  3. Partially support "unsafe" ranges, by using the lower bound of the range as the Python version to be installed, rather than the upper bound. (Pros: Compatibility with the default output of poetry init and avoids breakage when new Python versions are released. Cons: Doesn't prevent environment drift - developers can still be using a different Python version locally - particularly as new versions are released, unless they remember to bump the project's minimum Python version.)
  4. Error on "unsafe" version ranges telling users to change to a stricter range instead (eg 3.11.*). (Pros: Prevents breaking changes when new versions of Python released + prevents environment drift. Cons: First build on any app using Poetry will likely fail, unless they are using a template that already has a stricter range set.)

At the moment I'm leaning towards (4), since:

  • Our priority should be encouraging users towards safe patterns, rather than those that can cause hard to debug issues later (we already get support tickets where users using pip+runtime.txt were using different version of Python locally vs on Heroku and blame Heroku saying "it works on my machine")
  • Users using a Python package/project manager that uses a lockfile have choosing to use a more advanced tool that offers determinism over short-term convenience. Enforcing that a safe range is used (that forces use of a specific major Python version) would be in line with that.
  • We can always make .python-version take priority over tool.poetry.dependencies.python, and so the error message can say to either adjust the tool.poetry.dependencies.python range or create a .python-version file as an alternative (if they want to keep the wide range in the Poetry config).
  • Poetry's init command supports a --python option, so we could always encourage users to use poetry init --python '3.12.*' in our docs to save them having to fix the default version afterwards.
  • We could always advocate for Poetry supporting an "--type app" (or similar) for poetry init which picks defaults more appropriate for an app vs a library (in addition to it setting a Python version like 3.11.* instead of an unbounded range, it could also set package-mode = false which would avoid all of the other boilerplate)

We'll also want to factor in the pyproject.toml project.requires-python field since that is what uv uses and so we may need to support that too in the future.

See also:

GUS-W-9608268.

@edmorley
Copy link
Member Author

edmorley commented Sep 3, 2024

We could always advocate for Poetry supporting an "--type app" (or similar) for poetry init which picks defaults more appropriate for an app vs a library (in addition to it setting a Python version like 3.11.* instead of an unbounded range, it could also set package-mode = false which would avoid all of the other boilerplate)

I've filed a feature request for this upstream:

edmorley added a commit that referenced this issue Sep 3, 2024
The Python package manager Poetry is now supported for installing app
dependencies:
https://python-poetry.org

To use Poetry apps must have a `poetry.lock` lockfile, which can be
created by running `poetry lock` locally, after adding Poetry config to
`pyproject.toml` (which can be done either manually or by using
`poetry init`). Apps must only have one package manager file (either
`requirements.txt` or `poetry.lock`, but not both) otherwise the
buildpack will abort the build with an error (which will help prevent
some of the types of support tickets we see in the classic buildpack).

Poetry is installed into a build-only layer, so is not available at
run-time to reduce image size. The app dependencies are installed into
a virtual environment (the same as for pip, after #257), which is on
`PATH` so does not need explicit activation when using the app image.
As such, use of `poetry run` or `poetry shell` is not required at
run-time to use dependencies in the environment.

When using Poetry, pip is not explicitly installed, since Poetry
includes its own bundled copy that it will use instead (for the small
number of Poetry operations for which it still calls out to pip, such
as package uninstalls).

Both the Poetry and app dependencies layers are cached, however, the
Poetry download/wheel cache is not cached, since using it is slower than
caching the dependencies layer (for more details see the comments on
`poetry_dependencies::install_dependencies`).

The `poetry install --sync` command is run using `--only main` so as to
only install the main dependencies group and not any other groups (such
as test/dev/... groups).

Relevant Poetry docs:
- https://python-poetry.org/docs/cli/#install
- https://python-poetry.org/docs/configuration/
- https://python-poetry.org/docs/managing-dependencies/#dependency-groups

Work that will be handled later:
- Support for selecting Python version via `tool.poetry.dependencies.python`:
  #260
- Build output and error messages polish/CX review (this will be
  performed when switching the buildpack to the new logging style).
- More detailed user-facing docs:
  #11

Closes #7.
GUS-W-9607867.
GUS-W-9608286.
GUS-W-9608295.
edmorley added a commit that referenced this issue Sep 4, 2024
The Python package manager Poetry is now supported for installing app
dependencies:
https://python-poetry.org

To use Poetry, apps must have a `poetry.lock` lockfile, which can be
created by running `poetry lock` locally, after adding Poetry config to
`pyproject.toml` (which can be done either manually or by using
`poetry init`). Apps must only have one package manager file (either
`requirements.txt` or `poetry.lock`, but not both) otherwise the
buildpack will abort the build with an error (which will help prevent
some of the types of support tickets we see in the classic buildpack
with users unknowingly mixing and matching pip + Pipenv).

Poetry is installed into a build-only layer (to reduce the final app
image size), so is not available at run-time. The app dependencies are
installed into a virtual environment (the same as for pip after #257,
for the reasons described in #253), which is on `PATH` so does not need
explicit activation when using the app image. As such, use of
`poetry run` or `poetry shell` is not required at run-time to use
dependencies in the environment.

When using Poetry, pip is not installed (possible thanks to #258), since
Poetry includes its own internal vendored copy that it will use instead
(for the small number of Poetry operations for which it still calls out
to pip, such as package uninstalls).

Both the Poetry and app dependencies layers are cached, however, the
Poetry download/wheel cache is not cached, since using it is slower than
caching the dependencies layer (for more details see the comments on
`poetry_dependencies::install_dependencies`).

The `poetry install --sync` command is run using `--only main` so as to
only install the main `[tool.poetry.dependencies]` dependencies group
from `pyproject.toml`, and not any of the app's other dependency groups
(such as test/dev groups, eg `[tool.poetry.group.test.dependencies]`).

I've marked this `semver: major` since in the (probably unlikely) event
there are any early-adopter projects using this CNB that have both a
`requirements.txt` and `poetry.lock` then this change will cause them to
error (until one of the files is deleted).

Relevant Poetry docs:
- https://python-poetry.org/docs/cli/#install
- https://python-poetry.org/docs/configuration/
- https://python-poetry.org/docs/managing-dependencies/#dependency-groups

Work that will be handled later:
- Support for selecting Python version via `tool.poetry.dependencies.python`:
  #260
- Build output and error messages polish/CX review (this will be performed
  when switching the buildpack to the new logging style).
- More detailed user-facing docs:
  #11

Closes #7.
GUS-W-9607867.
GUS-W-9608286.
GUS-W-9608295.
edmorley added a commit that referenced this issue Sep 4, 2024
The Python package manager Poetry is now supported for installing app
dependencies:
https://python-poetry.org

To use Poetry, apps must have a `poetry.lock` lockfile, which can be
created by running `poetry lock` locally, after adding Poetry config to
`pyproject.toml` (which can be done either manually or by using
`poetry init`). Apps must only have one package manager file (either
`requirements.txt` or `poetry.lock`, but not both) otherwise the
buildpack will abort the build with an error (which will help prevent
some of the types of support tickets we see in the classic buildpack
with users unknowingly mixing and matching pip + Pipenv).

Poetry is installed into a build-only layer (to reduce the final app
image size), so is not available at run-time. The app dependencies are
installed into a virtual environment (the same as for pip after #257,
for the reasons described in #253), which is on `PATH` so does not need
explicit activation when using the app image. As such, use of
`poetry run` or `poetry shell` is not required at run-time to use
dependencies in the environment.

When using Poetry, pip is not installed (possible thanks to #258), since
Poetry includes its own internal vendored copy that it will use instead
(for the small number of Poetry operations for which it still calls out
to pip, such as package uninstalls).

Both the Poetry and app dependencies layers are cached, however, the
Poetry download/wheel cache is not cached, since using it is slower than
caching the dependencies layer (for more details see the comments on
`poetry_dependencies::install_dependencies`).

The `poetry install --sync` command is run using `--only main` so as to
only install the main `[tool.poetry.dependencies]` dependencies group
from `pyproject.toml`, and not any of the app's other dependency groups
(such as test/dev groups, eg `[tool.poetry.group.test.dependencies]`).

I've marked this `semver: major` since in the (probably unlikely) event
there are any early-adopter projects using this CNB that have both a
`requirements.txt` and `poetry.lock` then this change will cause them to
error (until one of the files is deleted).

Relevant Poetry docs:
- https://python-poetry.org/docs/cli/#install
- https://python-poetry.org/docs/configuration/
- https://python-poetry.org/docs/managing-dependencies/#dependency-groups

Work that will be handled later:
- Support for selecting Python version via `tool.poetry.dependencies.python`:
  #260
- Build output and error messages polish/CX review (this will be performed
  when switching the buildpack to the new logging style).
- More detailed user-facing docs:
  #11

Closes #7.
GUS-W-9607867.
GUS-W-9608286.
GUS-W-9608295.
@edmorley
Copy link
Member Author

edmorley commented Sep 4, 2024

Presuming we pick option (4), then the tool.poetry.dependencies.python specifiers we'll support will be:

  • 3.12.*
  • 3.12.5
  • ==3.12.*
  • ==3.12.5

We would then reject anything else with an error message that says to use one of the above, or to add a .python-version file instead (the .python-version file will accept values like 3.12 or 3.12.5).

Examples of specifiers we would then reject:

  • *
  • 3.*
  • >=3.12
  • ^3.12
  • ~3.12.5
    • whilst this is a "safe" range, it's (a) very similar to the PEP-440 ~= specifier, which has quite different semantics for certain cases (for example whilst ~3.12 and ~3.12.0 are equivalent, ~=3.12 and ~=3.12.0 are not), (b) I suspect there may be Python users who aren't used to seeing the Node.js semver style tilde usage and think this specifier is actually equivalent to 3.12.5, (c) it would needing to do actual version resolution rather than a simple mapping to "latest 3.12.x" (which adds to complexity for minimal benefit).
  • >=3.12,<3.13
    • supporting complex specifiers (that have multiple clauses) would mean needing to do actual version resolution (which adds to complexity for minimal benefit)

@edmorley
Copy link
Member Author

To add yet more things to think about - Poetry has just merged support for PEP-621 and the [project] section of pyproject.toml, in:
python-poetry/poetry#9135

This means Poetry 2.0 (expected later this year, see python-poetry/poetry#3332 (comment) and python-poetry/poetry#9448) will support requires-python too.

Given that:

  • Poetry supporting PEP-621 means it will presumably be slowly moving away from its proprietary tool.poetry.* table
  • We could only ever support a limited subset of the tool.poetry.dependencies.python syntax (syntax which isn't the default, so the first run experience will always be an error)
  • uv is leaning more into using .python-version for picking the single Python version to install (rather than requires-python)

... it's making me wonder whether we should instead double down on .python-version and not support tool.poetry.dependencies.python at all?

@edmorley
Copy link
Member Author

edmorley commented Nov 5, 2024

Given all of the above, I'm wontfixing this for now in favour of having a single way to specify the Python version - using a .python-version file.

@edmorley edmorley closed this as not planned Won't fix, can't repro, duplicate, stale Nov 5, 2024
edmorley added a commit to heroku/heroku-buildpack-python that referenced this issue Nov 6, 2024
After many refactoring/preparation PRs, we're now ready to add support
for the package manager Poetry:
https://python-poetry.org

To use Poetry, apps must have a `poetry.lock` lockfile, which can be
created by running `poetry lock` locally, after adding Poetry config to
`pyproject.toml` (which can be done either manually or by using
`poetry init`).

For now, if a `requirements.txt` or `Pipfile` is found it will take
precedence over `poetry.lock` for backwards compatibility (in the future
this will become a warning then an error). This means users of the
third-party `python-poetry-buildpack` will need to remove that buildpack
in order to use the new native Poetry support, since it exports a
`requirements.txt` file during the build.

Poetry is installed into the build cache rather than the slug, so is not
available at run-time (since it's not typically needed at run-time and
doing so reduces the slug size). The entrypoints of installed
dependencies are available on `PATH`, so use of `poetry run` or
`poetry shell` is not required at run-time to use dependencies in the
environment.

When using Poetry, pip is not installed since Poetry includes its own
internal vendored copy that it will use instead (for the small number
of Poetry operations for which it still calls out to pip, such as
package uninstalls).

During normal (non-CI) builds, the `poetry install --sync` command is
run using `--only main` so as to only install the main
`[tool.poetry.dependencies]` dependencies group from `pyproject.toml`
and not any of the app's other dependency groups (such as test/dev/...
groups, eg `[tool.poetry.group.test.dependencies]`).

On Heroku CI, all default Poetry dependency groups are installed (i.e.
all groups minus those marked as `optional = true`).

Relevant Poetry docs:
- https://python-poetry.org/docs/cli/#install
- https://python-poetry.org/docs/configuration/
- https://python-poetry.org/docs/managing-dependencies/#dependency-groups

See also the Python CNB equivalent of this PR:
- heroku/buildpacks-python#261

Note: We don't support controlling the Python version via Poetry's
`tool.poetry.dependencies.python` field, since that field typically
contains a version range, which is not safe to use. Use the newly
added `.python-version` file support instead. For more on this, see
the longer explanation over in the Python CNB repo:
heroku/buildpacks-python#260

Closes #796.
Closes #835.
GUS-W-16810914.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
enhancement New feature or request semver: major
Projects
None yet
Development

No branches or pull requests

1 participant