v0.1.3: EXSLT extensions, transform soundness fix (#6), CI → GitHub Actions#7
Merged
v0.1.3: EXSLT extensions, transform soundness fix (#6), CI → GitHub Actions#7
Conversation
LaTeXML stylesheets rely on EXSLT extension functions (str:tokenize,
math:*, set:*, date:*). Consumers currently have to reach for their
own extern "C" { fn exsltRegisterAll(); } unsafe declaration; that's
unpleasant and duplicates the crate's FFI policy outside the wrapper.
Changes:
* build.rs — also look up libexslt via pkg-config, falling back to
`cargo:rustc-link-lib=dylib=exslt` so systems with libexslt on the
default search path still link when pkg-config is unhelpful.
* src/bindings.rs — add \`pub fn exsltRegisterAll();\` inside a new
extern "C" block, with a block comment documenting purpose and
idempotence.
* src/lib.rs — new top-level \`register_exslt()\` Once-guarded safe
wrapper. Application code now calls this exactly-once on startup
instead of writing its own unsafe FFI.
No behaviour change; downstream consumers can now drop their
extern "C" blocks for exsltRegisterAll.
Follow-up on a61d0c4. Shifts EXSLT from opt-in to default-on, since the crate already links libexslt unconditionally — the previous opt-in wrapper paid the full cost without the default benefit. `xsltproc` and other mainstream XSLT tooling enable EXSLT by default for the same reason. Changes: * src/parser.rs — `parse_file` and `parse_bytes` call `register_exslt()` on entry; the `Once` guard makes the FFI fire exactly once per process. Stylesheets that reference `str:tokenize` et al. now Just Work without ceremony at the call site. * src/lib.rs — doc comment on `register_exslt` reframed: the function is now an idempotent manual hook for deterministic early init (tests, embedders), not a required call. Thread-safety credited to `std::sync::Once`. * build.rs — symmetric pkg-config handling for libxslt and libexslt. Both probe via pkg-config and fall back to `cargo:rustc-link-lib` on miss, instead of libxslt panicking while libexslt falls back. * src/bindings.rs — `MANUAL EDIT` marker on the `exsltRegisterAll` extern block so a future bindgen regeneration doesn't silently drop it. * tests/base_tests.rs — new `exslt_str_tokenize_auto_registers` regression test, self-contained (inline XSL + `<root/>` source via `parse_bytes` / `parse_string`, no fixture files). Deliberately avoids a manual `register_exslt()` call so it would fail if auto-registration regressed. * Cargo.toml — version bumped 0.1.2 → 0.1.3. * CHANGELOG.md — 0.1.3 section dated 2026-22-04 with release notes; new empty `[0.1.4] (in development)` header opened above. Verified: `cargo clippy --all-targets -- -D warnings` is clean; `cargo test` passes all 16 bindings-layout tests and 3 base tests (including the new EXSLT test). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
After 10a6fe0, the crate unconditionally links libexslt. Reflect that in the two places downstream users actually look: * .travis.yml — add `libexslt-dev` to the apt package list so the existing Linux build matrix keeps passing. * README.md — new "Installation" section listing the development headers (libxml2, libxslt, libexslt) across Debian/Ubuntu, Fedora and macOS. Previously the README said nothing about system deps; this is also the natural spot to land the EXSLT callout. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
travis-ci.org was decommissioned in 2021; the existing .travis.yml has been non-functional for years. Port the Linux build matrix (stable / beta / nightly) to GitHub Actions, keep clippy-on-stable, and retire the dead configuration. * .github/workflows/ci.yml — new. Runs on push to master and on all pull requests. Installs libxml2-dev / libxslt1-dev / libexslt-dev via apt, then `cargo build`, `cargo test`, and (stable only) `cargo clippy --all-targets -- -D warnings`. * .travis.yml — removed. * scripts/doc-upload.sh — removed; it relied on TRAVIS_* env vars and a TRAVIS-era GH_TOKEN secret, so it was dead code. Porting gh-pages deployment is a separate decision (branch, URL, token source) and is deliberately left out of this transition. * Cargo.toml — drop the now-stale `exclude = ["scripts/*"]`. * README.md — swap the Travis build badge for the GitHub Actions CI badge. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Modern Debian/Ubuntu (Bookworm, Noble, and later) bundle the libexslt headers, shared library, and pkg-config metadata inside libxslt1-dev. The separate `libexslt-dev` package was dropped — the GHA run for PR #7 failed with `E: Unable to locate package libexslt-dev` on the ubuntu-latest (noble) image. Verified locally: $ dpkg -L libxslt1-dev | grep -E 'exslt|libexslt\.' /usr/include/libexslt /usr/include/libexslt/exslt.h ... /usr/lib/x86_64-linux-gnu/pkgconfig/libexslt.pc Drop `libexslt-dev` from the CI apt line and from the README's per-distro install snippet; note in the README that libexslt now rides along with the libxslt devel package on all three platforms. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
libxslt mutates the input xmlDoc while applying stylesheet-directed whitespace stripping (see https://gitlab.gnome.org/GNOME/libxslt/-/issues/14). The previous signature pub fn transform(&mut self, doc: &Document, ...) handed that C-side mutation a shared Rust reference, which is undefined behaviour reachable from safe code. Switch to consuming ownership: pub fn transform(&mut self, doc: Document, ...) `doc` is dropped at the end of the transform; libxml's `Document` Drop impl frees the underlying xmlDoc. The returned `real_dom` is a separately allocated xmlDocPtr produced by libxslt, so there is no aliasing between input and output. Matches the resolution I proposed on the issue. This is a breaking API change for every caller — but the crate is already cutting a breaking 0.1.3 release (link-time libexslt dep), so the two fit into a single upgrade cycle. Downstreams should replace `transform(&doc, ...)` with `transform(doc, ...)`; clone the `Document` up front if you need to run it through multiple stylesheets. Verified: `cargo clippy --all-targets -- -D warnings` clean; all 16 bindings-layout and 3 base tests pass. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
dginev
changed the title
dginev
added a commit
that referenced
this pull request
Apr 22, 2026
Modern Debian/Ubuntu (Bookworm, Noble, and later) bundle the libexslt headers, shared library, and pkg-config metadata inside libxslt1-dev. The separate `libexslt-dev` package was dropped — the GHA run for PR #7 failed with `E: Unable to locate package libexslt-dev` on the ubuntu-latest (noble) image. Verified locally: $ dpkg -L libxslt1-dev | grep -E 'exslt|libexslt\.' /usr/include/libexslt /usr/include/libexslt/exslt.h ... /usr/lib/x86_64-linux-gnu/pkgconfig/libexslt.pc Drop `libexslt-dev` from the CI apt line and from the README's per-distro install snippet; note in the README that libexslt now rides along with the libxslt devel package on all three platforms. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Expose libexslt's EXSLT extension functions (
str:*,math:*,set:*,date:*), fix an unsoundness bug inStylesheet::transform(closes #6), and release as v0.1.3.Stylesheets using
str:tokenizeet al. now work without the consumer writing their ownunsafe extern "C" { fn exsltRegisterAll(); }. Also migrate CI from the long-dead.travis.ymlto GitHub Actions.Changes
Auto-registered EXSLT —
parser::parse_file/parse_bytescallregister_exslt()on entry, guarded bystd::sync::Once. Matchesxsltproc's default behaviour.libxslt::register_exslt()— new public, idempotent, thread-safe manual hook for early init.bindings::exsltRegisterAll— new FFI declaration, flagged as a manual addition sobindgenregenerations don't drop it.build.rs— links libexslt via pkg-config with a-lexsltfallback; libxslt gets the same symmetric handling (no more hardpanic!).Soundness fix (unsoundness from libxslt issue #14 #6) —
Stylesheet::transformnow consumes the inputDocumentby value. libxslt mutates the input while applying stylesheet-directed whitespace stripping, which was reachable UB through the previous&Documentsignature. Call sites changetransform(&source, …)totransform(source, …); clone theDocumentat the call site if you need to run it through multiple stylesheets.Compatibility
Stylesheet::transformsignature changes from(&mut self, doc: &Document, …)to(&mut self, doc: Document, …). Every existing call site needs a one-token edit (&source→source).libexsltis now a hard link-time dep — already bundled inside the main libxslt devel package on modern Debian/Ubuntu, Fedora, and macOS Homebrew, so most builds pick it up without any new install.transformno longer introduces UB when libxslt mutates the input during whitespace stripping.