Coincident with the release of Pulumi 3.0, we updated the provider SDK codegen for Python to no longer use casing tables for translating Python snake_case names to Pulumi camelCase names (and vice versa). Instead, the mapping is encoded in decorators applied on class properties.
Some of the code that was used to generate and use the casing tables has persisted. This commits removes this code, as it's no longer necessary, and will improve the quality of our generated examples.
This commit modifies the generation of `setup.py` to use Python
variables as the source for the package version and plugin version
instead of placeholder strings. This has the effect of making the
packages installable via the `-e` flag directly from their source
directory rather than requiring a build step, which is useful while
developing a plugin and examples in tandem.
These changes support arbitrary combinations of input + plain types
within a schema. Handling plain types at the property level was not
sufficient to support such combinations. Reifying these types
required updating quite a bit of code. This is likely to have caused
some temporary complications, but should eventually lead to
substantial simplification in the SDK and program code generators.
With the new design, input and optional types are explicit in the schema
type system. Optionals will only appear at the outermost level of a type
(i.e. Input<Optional<>>, Array<Optional<>>, etc. will not occur). In
addition to explicit input types, each object type now has a "plain"
shape and an "input" shape. The former uses only plain types; the latter
uses input shapes wherever a plain type is not specified. Plain types
are indicated in the schema by setting the "plain" property of a type spec
to true.
Every time a resource is created (without an explicit provider version specified), `_utilities.get_version()` is called to determine the package version. This call is expensive and can measurably impact performance, especially for programs with many resources. This change caches the value so it is only determined once.
Python resource constructor overloads were recently added that accept a
`<Resource>Args` class for input properties, as an alternative to the
other constructor overload that accepts keyword arguments. The name of
the new args class is the name of the resource concatenated with an
`Args` suffix.
Some providers (e.g. Kubernetes, Azure Native, and Google Native) have
input types with the same name as resources in the same module, which
results in two different `<Resource>Args` classes in the same module.
When you try to use the new args class with the constructor, e.g.:
```python
pulumi_kubernetes.storage.v1.StorageClass(
resource_name='string',
args=pulumi_kubernetes.storage.v1.StorageClassArgs(...),
opts=pulumi.ResourceOptions(...),
)
```
You run into an error, because
`pulumi_kubernetes.storage.v1.StorageClassArgs` is actually referring to
the existing input type rather than the intended `StorageClassArgs`
class for the constructor arguments.
Having the duplicate classes hasn't broken existing usage of the input
type because we "export" all the input types for a module _after_ all
the resources and resource args classes are exported, so the input type
just ends up "overwriting" the duplicate resource args class.
Other languages don't have this problem because the input type is either
in it's own module/namespace (e.g. Node.js and .NET) or a different name
is used for the input type (Go). But with Python, the input types and
resources are all available in the same module.
To address this for Python, when there is an input type in the same
module with the same name as the resource, the args class for the
resource will be emitted as `<Resource>InitArgs` instead of
`<Resource>Args`.
This commit adds a fallback for the README definition in the generated
setup.py files for Python SDKs, thus allowing editable installs of
packages which not yet been built.
Co-authored-by: Luke Hoban <luke@pulumi.com>
This commit adds a new language option for Python generation to specify
the package name instead of using `pulumi_x` where x is the name defined
in the schema.
A new test is added, and this has also been shown to produce no diff
when run against `pulumi-eks`.
* Import subpackages lazily
* Tighten up lazy_import impl
* Eagerly register resources, but lazily load their impl
* Add CHANGELOG entry
* Satisfy lint
* Restore mypy behavior so the change is not breaking
* Fix golden tests
These changes fix a regression introduced by #6686 that caused the SDK
code generators for .NET, Python, and Typescript to omit definitions for
plain object types. This regression occurred because #6686 drew a
clearer line between types used as resource arguments and types used
as function arguments, but conflated "resource arguments" with "inputty
types". This caused the code generators to generate inputty types for
any types used as resource arguments, even those that are used for
plainly-typed properties.
Fixes#6796.
See #6200 for a complete description of the issue. In short, we generate
inconsistent names for object types depending on whether or not they are
transitively reachable from resources or functions, which risks
unintentional breaking changes due to schema updates.
1. Name "input" types differently: `TArgs` for a type that is used in
resource inputs, having `Input<T>` properties, and `T` for a type
that is used in invoke inputs. The same schema type can produce both.
2. Always keep the name `T` for output types, avoid appending `Result` to
the name.
3. As needed, introduce a flag in the existing providers' schemas to avoid
breaking changes. Consider removing it on a major version bump.
Fixes#6200.
These have been deprecated for a very long time and it's a trivial change to remove them from the generated code. Let's clean this up for the 3.0-based providers.
This change updates the Python SDK codegen to opt-in to the new casing
translation behavior, which will use the passed-in props type's property
name metadata for translations, rather than calling the resource's
`translate_input_property` and `translate_output_property` methods.
- FIX: Keys in user-defined dicts will no longer be unintentionally
translated/modified.
- BREAKING: Dictionary keys in nested output classes are now
consistently snake_case. If accessing camelCase keys from such output
classes, move to accessing the values via the snake_case property
getters (or snake_case keys). A warning will be logged when accessing
camelCase keys.
When serializing inputs:
- If a value is a dict and the associated type is an input type, the
dict's keys will be translated based on the input type's property
name metadata.
- If a value is a dict and the associated type is a dict (or Mapping),
the dict's keys will _not_ be translated.
When resolving outputs:
- If a value is a dict and the associated type is an output type, the
dict's keys will be translated based on the output type's property
name metadata.
- If a value is a dict and the associated type is a dict (or Mapping),
the dict's keys will _not_ be translated.
Add support for creating instances of resources in Python using a
`<Resource>Args` class. This capability aligns with how args are passed
to resources in all the other language SDKs and the separate object bag
allows the properties to be manipulated/validated/passed-around before
creating the resource.
This change adds schema and codegen support for plain properties which
are emitted typed as the plain type rather than wrapped as an `Input`.
Plain properties require a prompt value and do not accept a value that
is `Output`.
This new optional file can be used to include additional metadata about the provider plugin (such as version and custom server URL), which the Python language host will use when determining a program's required plugins.
This change fixes the codegen to emit the file in the correct location (inside the package dir). Note: Providers need to opt-in to emitting this file via a schema option (because it requires some Makefile changes to insert the version in the file) and we haven't done that with any of our providers yet.
Implement GetRequiredPlugins for Python, which determines the plugins
required by the program.
Also, if the `virtualenv` runtime option is set, and the specified
virtual directory is missing or empty, automatically create it and
install dependencies into it.
Generate ResourcePackage and ResourceModule implementations and
registrations. A ResourcePackage is generated for any module that
includes a provider resource (which should be the root module only), and
a ResourceModule is generated for any module that includes a resource.
And add version info to Python registrations.
Generate ResourcePackage and ResourceModule implementations and
registrations. A ResourcePackage is generated for any module that
includes a provider resource (which should be the root module only), and
a ResourceModule is generated for any module that includes a resource.
Note that version information is currently omitted. We should fix this
up before enabling resource reference deserialization end-to-end.
Non-string provider inputs must be projected as JSON formatted strings. The current codegen simply calls `json.dumps` for such properties, but this does not work for the new input types, which aren't JSON serializable.
To address this, make use of the new `pulumi.runtime.to_json` utility function, which is capable of serializing raw dicts and input types as JSON.
We currently emit array types as `List[T]` for Python, but `List[T]` is invariant, which causes type checkers like mypy to produce errors when values like `["foo", "bar"]` are passed as args typed as `List[pulumi.Input[str]]`. Instead, we should move to using `Sequence[T]` which is covariant, and does not have this problem.
We actually already do this for `Dict` vs. `Mapping`, emitting map types as `Mapping[str, T]` rather than `Dict[str, T]` because `Mapping[str, T]` is covariant. This change makes us consistent for array types.