The current implementation removes parameters if they are used - however, it only removes the first instance, and removes instances by index. However, when multiple items are being removed, following indices must be decremented by 1 to account for previously removed items.
This introduces the `serde_with` dependency and `rust_type.py`, to allow supporting arbitrary types for serialization.
Since fields may have arbitrary types (eg. `HashMap<_, chrono::Duration>`) which need deserialization, it is necessary to
use type-based serialization to avoid implementing (de)serialization for every permutation of types that require special serialization.
However, `serde` does not let you (de)serialize one type as another (eg. `chrono::Duration` as `Wrapper`) - thus necessitating `serde_with`, which does. `rust_type.py` introduces the `RustType` class, which makes it easy to describe the (de)serialization type used by `serde_with`
The serde traits are now directly implemented for FieldMask - this helps address potential serde issues with wrapper types, and simplifies the serde process somewhat.
If using #[serde(with = ...)] with an Option type, serde will expect all marked fields to be present. Adding #[serde(default)] restores expected behaviour - if no Option value is present, None will be used.
The note says that it's not worth investigating how to make the build
system not look for this file if in API mode, as an empty file that
isn't used hurts no one and probably keeps everything simpler.
Move existing tests into a new __tests__ module under generator/lib.
Also define a new test for `util.new_context`, to prevent regressions as
changes are made.
This is an incremental change towards a strongly-typed util module,
aimed at reducing dependency on the DictObject class. The rough idea is
to annotate everything as Dict, add some tests to codify the existing
behavior, and then start defining dataclasses for the dischovery schema.
We also remove some unused logic & params.
Make a few changes in the toplevel src dir to help separate
templates from code. Specifically, we rename `src/mako` to
`src/generator`, and nest the mako templates inside a
`src/generator/templates` dir.
This isolates most Python code into the `src/generator/lib` dir.
* Update imports to use fully qualified module name
* Add `typecheck` target to Makefile
* Add type annotations to appease pyright
* Switch from importlib_resources backport library to stdlib
import.resources
