feat: modifying code generation to reduce bundle size #4978
Conversation
gvwilson
force-pushed
the
codegen2
branch
2 times, most recently
from
0c7ea5e to
2b3d47b
Compare
Removing the check for Python < 3.7 using `sys.version_info` and as a backup checking `typing.TYPE_CHECKING`; this saves us a little space and also cleans up the code. Proposing this as an enhancement beyond what's in the `codegen2` branch / PR #4978.
gvwilson
force-pushed
the
codegen2
branch
2 times, most recently
from
000a5e9 to
dbbbaa8
Compare
1. Add `bin/get_size.py` so that `python bin/get_size.py plotly build`
reports the number of files and total size in bytes of the `plotly`
directory (where generated code is put) and the `build` directory
that is populated by `python setup.py build`.
1. Modify `codegen/__init__.py` and `./setup.py` so that
`python setup.py --reformat=false` disables reformatting.
1. Assign an empty string to the `data_docs` field of generated
validators. (This has a major impact because those docs are
duplicated many times.)
1. Alias name of base validator during import in
`codegen/validators.py`.
1. Remove the long list of CSS colors from help strings for color
properties.
1. Replace `super(Parent, self)` with `super()` in generated code.
1. Drop use of sys.version_info and TYPE_CHECKING. Removed the check
for Python < 3.7 using `sys.version_info` and as a backup checking
`typing.TYPE_CHECKING`; this saves a little space and also cleans
up the code.
1. Remove mention of Chart Studio and explicit enumeration of system
font names from plotly.js / plot-schema.json so that this text
isn't copied dozens of times into the plotly.py bundle.
1. Introduce `_init_provided()` for `BaseFigure` and `BasePlotlyType`
that calls a helper function `_initialize_provided()` to replace
repetitions of:
```
_v = arg.pop("something", None)
_v = something if something is not None else _v
if _v is not None:
self["something"] = _v
```
Original size of plotly/**/*.py: 42283582 bytes
Current size of plotly/**/*.py: 31931739 bytes
Change: -25%
gvwilson
requested review from
alexcjohnson,
emilykl,
marthacryan and
LiamConnors
1. Modify `commands.py` to run code generation. 1. Remove comments from generated code. 1. Replaced named arguments in constructors with positional arguments. 1. Regenerate all code. Notes: The generated code is reformatted once again: this slightly increases size but makes it more readable. There is one flaky test: `tests/test_plotly_utils/validators/test_colorscale_validator.py::test_acceptance_named[Inferno_r]` It fails when the entire test suite is run but does *not* fail when only `test_colorscale_validator.py` is run (using Python 3.11.9). | branch | format | bytes | %age | | -------- | ------- | -------- | ---- | | master | .whl | 14803768 | | | codegen2 | .whl | 12215667 | -18% | | master | .tar.gz | 8100014 | | | codegen2 | .tar.gz | 6114458 | -24% |
|
see also #4951 |
1. Update required version of `black` in `pyproject.toml` and `.circleci/config.yml`. 2. Update Python version identifiers in `tool.black` section of `pyproject.toml` to include `py311` and `py312`. 3. Modify invocation of `black` in `commands.py` to format with `py311`. 4. Regenerate and reformat code.
|
@gvwilson High-level question -- what is the practical impact of some of these changes on the docs / docstrings, for example, removing the list of CSS colors or the list of dict properties? Completely agree that having those large chunks of text repeated many times over is a very poor use of bundle bytes. But are there cases where it makes the docstrings less useful, and if so are there any changes we need to make in the docs to mitigate the impact? |
|
@gvwilson Could we add a single line at the top of all codegen files saying I realize that runs exactly opposite to the goal of reducing bundle size, but it would make development a lot easier to have all code-generated files identified in the file itself. (Could be a separate PR.) |
codegen/resources/plot-schema.json
Outdated
There was a problem hiding this comment.
Why is this file showing up in the diff? Bad merge? Should probably just defer to whatever version of this file is on main.
|
@gvwilson Looks good, see my comments for some questions I have, but none of them are blockers to merging — except possibly the removal of the |
@emilykl agreed - we will need to provide a canonical list of colors and fonts. @LiamConnors would you like to figure out where you'd like to put this and then I can add the URL into plotly.js (to be copied into plotly.py by codegen) that can stay in the docstring? |
Excellent idea - it will only add a few bytes per file but will save a lot of grief. |
- Add caveat to the top of every machine-generated file to warn people not to edit them. - Make name of initial value property setter more readable. - Run black formatting with multiple Python versions.
feat: modify code generation to reduce bundle size
Assign an empty string to the
data_docsfield of generatedvalidators. (This has a major impact because those docs are
duplicated many times.)
Alias name of base validator during import in
codegen/validators.py.Remove the long list of CSS colors from help strings for color
properties.
Replace
super(Parent, self)withsuper()in generated code.Modify
commands.pyto run code generation.Remove comments from generated code.
Replaced named arguments in constructors with positional arguments.
Drop use of sys.version_info and TYPE_CHECKING. Removed the check
for Python < 3.7 using
sys.version_infoand as a backup checkingtyping.TYPE_CHECKING; this saves a little space and also cleansup the code.
Introduce
_init_provided()forBaseFigureandBasePlotlyTypethat calls a helper function
_initialize_provided()to replacerepetitions of:
Used in #5008.