Skip to content

v0.3.52 | Out-of-the-box OCR for the Node.js, Go and C# prebuilts, a Node worker-teardown fix that silenced a spurious exit warning, an OCR detection-unclip fix that restores recognition on wide text lines (native and WASM bindings alike), a Markdown→PDF styling fix that restores headings, bold/italic and monospace, strict CI toolchain-drift gating, and a dependency-maintenance batch.

Choose a tag to compare

View all tags
@github-actions github-actions released this 20 May 18:09
· 171 commits to main since this release
v0.3.52
2da2a35
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Added

  • OCR in the prebuilt native library for Node.js, Go and C#
    (#520)     — the
    build-native-libs release job now compiles the shared library with
    the ocr feature (alongside
    rendering,signatures,barcodes,tsa-client,system-fonts), so auto-mode
    OCR / extractTextAuto works straight from the published Node, Go and
    C# packages with no --build-from-source. docs/OCR_GUIDE.md
    gains an "OCR Support by Binding" matrix and a pure-npm Node recipe
    (npm install pdf-oxide onnxruntime-node, ORT_DYLIB_PATH via
    require.resolve(...), prefetchModels + extractTextAuto). The
    default pdf-oxide-wasm still ships without OCR; see the opt-in
    wasm-ocr build below for the experimental WASM OCR path.

  • WASM OCR backend (#524, experimental, opt-in) — pure-Rust
    tract inference under a new wasm-ocr build feature
    (ocr-tract is
    the bare backend; wasm-ml is retained as a thin back-compat
    alias). The default pdf-oxide-wasm package is unchanged and
    still ships without OCR; only wasm-ocr builds include it. The
    pure-Rust path is output-equivalent to the native ort
    backend     — verified at the inference-engine level (max abs diff
    ≤ 3e-6 on the real PaddleOCR det/rec graphs) and end-to-end
    (byte-identical recognized text on a shared fixture). Cross-target
    (browser / Deno / edge) integration tests and a release .wasm
    size gate are pending and will land in #524's own dedicated release
    cycle. See docs/OCR_GUIDE.md for the JS fetch+Cache recipe and the
    build command.

Fixed

  • OCR: wide text lines were misread (detection unclip bug)
    (#524)     — the
    DBNet box-unclip step scaled each corner by a percent of its own
    dimension     from the centre instead of PaddleOCR's uniform
    area·ratio/perimeter polygon offset. On a long, short text line
    that is badly anisotropic: it over-grew the long axis (pushing the
    box's x origin off-image, negative) and barely grew the short axis
    (the box stayed ~one glyph-band tall), so the recognizer received a
    horizontally-shifted, vertically-clipped sliver. A clean single
    line "OCR fidelity test hello world 2024" came out
    "OcR tdenfy test neno woridZoZ4 s" (confidence 0.66). Replaced with
    the standard uniform offset: the same line now reads exactly, at
    0.98 confidence. Affects the native OCR path (all bindings), not
    just WASM — the two backends are bit-equivalent. Pinned by new
    postprocessor unit tests and an end-to-end regression guard.

  • Node prefetchModels() no longer emits a spurious
    Worker N exited with code 1
    (#521)     — the
    worker_threads pool is now spawned lazily on the first real task
    (importing the library, or calling the synchronous native APIs such as
    extractText* / classifyPage / prefetchModels, spawns zero
    workers); pooled workers are unref()'d so an idle pool never keeps
    the event loop alive; and teardown does an async graceful
    terminate() on beforeExit (deliberately no SIGINT/SIGTERM
    listeners — a library must not change the host's default signal
    semantics) with a synchronous terminated flag flipped on the hard
    exit so a normal process exit killing an unref'd worker is no longer
    reported as an abnormal exit.

  • Markdown → PDF now renders styling instead of flat body text
    (#525)     —
    Pdf::from_markdown (and from_html, which funnels through it)
    computed heading sizes but only used them for line spacing, then drew
    every line in a single regular font, and stripped **bold** /
    *italic* markers to plain text. Headings (#–####) now render in
    the bold face at 2.0/1.5/1.25/1.1× scale; inline **bold**,
    *italic* and `code` produce real per-run font switches
    (Helvetica-Bold/-Oblique, Courier) measured and positioned so a line
    stays visually contiguous; fenced code blocks and GFM tables render
    monospace. Two writer-layer bugs that masked this are also fixed:
    map_font_name discarded explicit Standard-14 weight/oblique names
    (Helvetica-Bold → Helvetica) and had no italic path, and the page
    /Font resource set registered only 6 of the 12 Latin Standard-14
    faces, so any Tf to an oblique/bold-serif face resolved to a missing
    resource and silently fell back to regular. Underscores are no longer
    treated as emphasis, so snake_case identifiers survive intact.
    Reported by @Jethril.

CI / Release

  • Strict toolchain-drift gating
    (#522)     —
    RUSTFLAGS=-D warnings and RUSTDOCFLAGS=-D warnings are enforced on
    the stable matrix leg only, and continue-on-error has been
    removed from the beta/nightly legs so upstream-rustc drift surfaces as
    a real signal instead of being silently swallowed. Residual nightly
    rust-lld SIGBUS risk is mitigated with CARGO_BUILD_JOBS=2 and
    documented inline in the workflow.
  • Dependency maintenance — quick-xml 0.39 → 0.40
    (#494) with all
    seven BytesText::xml_content() call sites migrated to the
    behaviour-identical zero-arg xml11_content() (0.40 added a
    version: XmlVersion parameter; 0.39's xml_content() was literally
    self.xml11_content(), so this is a byte-for-byte no-op);
    tokenizers 0.22 → 0.23
    (#498);
    weezl 0.1 → 0.2
    (#527) — picks
    up the upstream LZW-decoder fix for streams that overwrote initial
    table entries (benefits pdf_oxide's direct PDF LZWDecode filter
    in src/decoders/lzw.rs; the image crate's TIFF reader still pins
    weezl 0.1 transitively and is unaffected by this bump); aws-lc-rs
    1.16.3 → 1.17.0
    (#526) — build-
    hygiene only (jitterentropy CFLAGS filtering for FreeBSD, MinGW
    Win7 fixes, nightly clippy); and
    SHA-pinned action bumps actions/upload-artifact → v7.0.1
    (#495) with
    actions/download-artifact → v8.0.1 for compat and
    astral-sh/setup-uv → v8.1.0
    (#502), unified
    across all nine workflows. Additional pinned-action bumps in this
    release: pypa/gh-action-pypi-publish → v1.14.0
    (#530) which
    carries a security fix
    (GHSA-vxmw-7h4f-hqxh)
    for the action that publishes our Python wheel;
    github/codeql-action → v4.35.5
    (#528);
    taiki-e/install-action → v2.79.2
    (#529) — the
    upstream deprecations (mdbook-alerts, iai-callgrind-runner) do
    not affect us (our tool list is cargo-cyclonedx, wasm-tools,
    taplo-cli, cargo-shear); and codecov/codecov-action → v6.0.1
    (#531).
    The Dependabot ort 2.0.0-rc.11 → rc.12
    bump (#496) was
    declined — rc.12 is an upstream regression (missing
    SessionOptionsAppendExecutionProvider_VitisAI on OrtApi); the pin
    is held at =2.0.0-rc.11.

Thanks

  • @Jethril for reporting
    #525 — the
    PDF-from-markdown-has-no-styling bug — with a minimal repro that led
    directly to the renderer fix.

Installation

Rust (crates.io)

cargo add pdf_oxide

Python (PyPI)

pip install pdf_oxide

JavaScript/WASM (npm)

npm install pdf-oxide-wasm

CLI (Homebrew)

brew install yfedoseev/tap/pdf-oxide

CLI (Scoop — Windows)

scoop bucket add pdf-oxide https://github.com/yfedoseev/scoop-pdf-oxide
scoop install pdf-oxide

CLI (Shell installer)

curl -fsSL https://raw.githubusercontent.com/yfedoseev/pdf_oxide/main/install.sh | sh

CLI (cargo-binstall)

cargo binstall pdf_oxide_cli

MCP Server (for AI assistants)

cargo install pdf_oxide_mcp

Pre-built Binaries
Download archives for Linux, macOS, and Windows from the assets below. Each archive includes both pdf-oxide (CLI) and pdf-oxide-mcp (MCP server).

Platform Support

Platform Architecture Archive
Linux x86_64 (glibc) pdf_oxide-linux-x86_64-*.tar.gz
Linux x86_64 (musl) pdf_oxide-linux-x86_64-musl-*.tar.gz
Linux ARM64 pdf_oxide-linux-aarch64-*.tar.gz
macOS x86_64 (Intel) pdf_oxide-macos-x86_64-*.tar.gz
macOS ARM64 (Apple Silicon) pdf_oxide-macos-aarch64-*.tar.gz
Windows x86_64 pdf_oxide-windows-x86_64-*.zip

Changelog

See CHANGELOG.md for full details.

Contributors

Jethril
Assets 40
Loading