SameerMesiah97 commented on PR #74536: URL: https://github.com/apache/airflow/pull/74536#issuecomment-6712358971
> Related to the issue that was opened [NVIDIA/sphinx-llm#203](https://github.com/NVIDIA/sphinx-llm/issues/203) it looks like there are some problems with cache reuse between html and markdown docs builds in this project. Specifically, it seems like there's a custom extension that makes output-specific transformations during the read of the sources. I'm happy to explore disabling cache reuse in `sphinx-llm` as an option, but it would also be interesting to see whether those transformations can be deferred to a later stage. > > I had Codex put together this dish, which should avoid these caching problems. Let me know what you think. > > ```diff > diff --git a/devel-common/src/sphinx_exts/airflow_llms_txt.py b/devel-common/src/sphinx_exts/airflow_llms_txt.py > index 050bc335d6..9b4d1f18db 100644 > --- a/devel-common/src/sphinx_exts/airflow_llms_txt.py > +++ b/devel-common/src/sphinx_exts/airflow_llms_txt.py > @@ -43,6 +43,7 @@ from typing import TYPE_CHECKING, Any > from urllib.parse import urljoin > > from docutils import nodes > +from exampleinclude import ExampleHeader > from sphinx import addnodes > from sphinx.util import logging > from sphinx.util.matching import Matcher > @@ -51,7 +52,6 @@ from sphinx_markdown_builder.contexts import IndentContext, SubContext, SubConte > > if TYPE_CHECKING: > from sphinx.application import Sphinx > - from sphinx.config import Config > from sphinx.environment import BuildEnvironment > > logger = logging.getLogger(__name__) > @@ -219,40 +219,32 @@ def _xref_title(xref: addnodes.pending_xref, env: BuildEnvironment) -> str | Non > return None > > > -def _separate_markdown_doctrees(app: Sphinx, config: Config) -> None: > - """ > - Give the Markdown build a doctree cache of its own. > - > - Without parallel mode sphinx-llm points the Markdown build at the HTML build's doctree cache. > - The Markdown build would then reuse doctrees read for HTML (curly quotes, none of this > - extension's ``doctree-read`` changes) or, once re-read, overwrite them. Its own directory sits > - inside sphinx-llm's temporary output, which is deleted after the Markdown files are copied. > - Sphinx creates the environment after ``config-inited``, so the change takes effect. > - """ > - if app.tags.has("sphinx_llm_markdown"): > - app.doctreedir = Path(app.outdir) / ".doctrees" > - > - > def _is_markdown_build(app: Sphinx) -> bool: > return app.builder is not None and app.builder.name == MARKDOWN_BUILDER > > > -def _link_example_sources(app: Sphinx, doctree: nodes.document) -> None: > +def _remember_example_sources(app: Sphinx, doctree: nodes.document) -> None: > + # exampleinclude replaces ExampleHeader during doctree-read. Keep its filename on > + # the wrapper so either writer can render the same cached tree appropriately. > + for header in doctree.findall(ExampleHeader): > + header.parent["example_source_filename"] = header["filename"] > + > + > +def _link_example_sources(app: Sphinx, doctree: nodes.document, docname: str) -> None: > """ > Replace each ``exampleinclude`` header with a link to the whole example file. > > - Runs before ``exampleinclude``'s own ``doctree-read`` handler, which would otherwise turn the > - header into a bare module path plus a "View Source" button that has no Markdown equivalent. > + Run during output processing so Markdown can reuse the HTML build's cached doctrees > + without changing their headers or running the source parser again. > """ > if not _is_markdown_build(app): > return > - # Imported by bare name, as conf.py loads it: another import path gives a different class. > - from exampleinclude import ExampleHeader > - > repo_root = Path(app.config.llms_txt_repo_root).resolve() > source_url = app.config.llms_txt_source_url.rstrip("/") > - for header in list(doctree.findall(ExampleHeader)): > - path = Path(header["filename"]).resolve() > + for header in list(doctree.findall(nodes.paragraph)): > + if "example-header" not in header["classes"] or "example_source_filename" not in header.parent: > + continue > + path = Path(header.parent["example_source_filename"]).resolve() > if not source_url or not path.is_relative_to(repo_root): > header.parent.remove(header) > continue > @@ -477,8 +469,8 @@ def setup(app: Sphinx) -> dict[str, Any]: > app.add_config_value("llms_txt_root_description", "", "env") > app.add_config_value("llms_txt_optional_docs", [], "env") > app.add_config_value("llms_txt_skip_docs", [], "env") > - app.connect("config-inited", _separate_markdown_doctrees) > - app.connect("doctree-read", _link_example_sources, priority=400) > + app.connect("doctree-read", _remember_example_sources, priority=400) > + app.connect("doctree-resolved", _link_example_sources) > app.connect("doctree-resolved", _clean_markdown_doctree) > app.connect("build-finished", _write_llms_txt, priority=102) > - return {"parallel_read_safe": True, "parallel_write_safe": True} > + return {"env_version": 1, "parallel_read_safe": True, "parallel_write_safe": True} > ``` I have implemented this fix and it appeared to have worked based on my manual tests. I have covered this in my latest commit with the associated test updates. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
