We use `markdown_exec.markdown_config.save()` to store the user configuration, so that markdown-exec's inner Markdown instances can render contents with the exact same configuration. Before, it would fallback to `md.registeredExtensions`, which isn't reliable.
Signed-off-by: Timothée Mazzucotelli <dev@pawamoy.fr>
## Summary
This version fixes a regression introduced in 0.0.55, making Zensical report false positives for the deprecated `unresolved_references` setting. Additionally, checking of autorefs is moved to the `invalid_links` setting. Please disable `unresolved_references` - it is not supported anymore, and will be replaced by the reference parser that we're currently testing in [Zensical Studio](https://zensical.org/studio/).
Signed-off-by: squidfunk <martin.donath@squidfunk.com>
## Summary
This version fixes two false positives in the reference extractor and reverts Zensical's bootstrapped `zensical.toml` to TOML 1.0, since most editor tooling does not yet support TOML 1.1. Zensical understands both, TOML 1.0 and TOML 1.1, so this change does not affect the functionality of Zensical itself.
Signed-off-by: squidfunk <martin.donath@squidfunk.com>
## Summary
This version significantly reduces peak memory usage by 8–10×, making builds of large documentation projects substantially more efficient. It also improves reference validation by preventing false positives for links successfully resolved through autorefs.
Additionally, the [user interface](https://github.com/zensical/ui) is updated to [v0.0.25](https://github.com/zensical/ui/releases/tag/v0.0.25), fixing instant navigation for inline scripts and version aliases created with `mike`. This release also corrects `mike` configuration defaults and updates `webbrowser` to address a security vulnerability.
Signed-off-by: squidfunk <martin.donath@squidfunk.com>
This commit fixes validation reporting false unresolved-reference errors for Markdown links that are successfully resolved by autorefs. Validation now runs after pages render, so it uses the autorefs results from the current build state and emits one complete issue report for that state.
**Validation implementation summary:**
- Page rendering now returns the identifiers that autorefs resolved, alongside rendered output.
- Validation combines those resolutions with the source-page references and anchors; a reference resolved by autorefs is no longer reported as unresolved.
- The workflow tracks a generation for page content and navigation, so it waits for matching render results from every page before creating issues. This prevents validation from using stale render data or printing duplicate reports during incremental updates.
- Reference collection now follows the configured documentation directory, and tests cover reporting after rendering and a subsequent clean rebuild.
Signed-off-by: Timothée Mazzucotelli <dev@pawamoy.fr>
## Summary
This version adds support for enabling strict mode directly in `mkdocs.yml` or `zensical.toml`, allowing warnings to fail builds consistently without the `--strict` command-line option. It also reduces memory usage by sharing cross-reference data between navigation clones.
Additionally, the user interface is updated to `v0.0.24`, improving search rendering for right-to-left languages and adding four new Lucide icons. The generated `zensical.toml` now uses TOML 1.1 syntax, and `pymdownx` is updated to version 11.0 to address a vulnerability.
Signed-off-by: squidfunk <martin.donath@squidfunk.com>
## Summary
This version adds search support for CJK languages: Chinese, Japanese, and Korean. Search needs to segment text into individual words before it can index and match them. While many languages separate words with spaces, this is not consistently the case for CJK text, particularly Chinese and Japanese. Search now uses locale-aware segmentation to identify meaningful word boundaries, ensuring that content and queries are indexed and matched correctly.
Set the [site language](https://zensical.org/docs/setup/language/#site-language) to `zh`, `zh-Hant`, `zh-TW`, `ja`, or `ko` to enable segmentation.
The search modal is also larger, showing more results and context at once. Additionally, keyboard keys inside admonitions now use the correct background color in the modern theme.
Dependencies and development tooling were updated, including TypeScript 7. The build scripts now use `tsx` instead of `ts-node`, and `npm-run-all2` replaces the unmaintained `npm-run-all`. Updated icon packages add 56 new icons: 23 Font Awesome icons, 21 Lucide icons, 8 Octicons, and 4 Simple Icons.
Signed-off-by: squidfunk <martin.donath@squidfunk.com>
## Summary
This version improves cached rebuilds by persisting `objects.inv` and autorefs data, fixes stale builds after removing files, and upgrades `soupsieve` to address vulnerabilities.
Signed-off-by: squidfunk <martin.donath@squidfunk.com>
On a first build without cache, pages are built and cross-references resolved thanks to the autorefs data retrieved from the Python interpreter.
On a second build, if a page changed, the cache is not reused, so the page is rebuilt and cross-references within it must be resolved again, but if the target symbols were rendered in other, already cached pages, the Python interpreter doesn't hold the necessary data to resolve these cross-references.
We fix this by caching autorefs data. We still need to update the data retrieved from the cache with the data retrieved from the Python interpreter, to give precedence to autorefs data for pages that were rebuilt.
This solution still has the issue that stale data is never removed from the cache (for example, symbols that were completely deleted from the sources and API docs). The API would have to change a lot and often for it to have an impact though. Cleaning (`rm .cache/autorefs.json` or `zensical build --clean`) from time to time easily mitigates this.
Signed-off-by: Timothée Mazzucotelli <dev@pawamoy.fr>
Martin Donath