Rewrite foundations of native elements by laurmaedje · Pull Request #6547 · typst/typst

laurmaedje · 2025-07-02T18:42:05Z

This PR rewrites the foundations of native elements and the #[elem] macro. The goal is to move more of the logic into normal as opposed to macro-generated code, improving maintainability. The whole system would've needed an overhaul with the unification of types and elements at the latest, so this is a good incremental preparation step for this.

A theme of the PR is that logic that was previously concretely generated for each element through the macro is now built on a few foundational macro-generated buildings blocks through a mix of generics and const fns.

While generics and const fn of course also have compile-time costs (perhaps even higher ones), from my experiments that part of the Rust compiler is much better incrementalized than the front-end (which isn't really incremental at all), so the hope would be better incremental build times. But that would mostly be a nice-to-have; moving away from macro-generated code is IMO worth it in itself.

If you want to take a look at the changes, I recommend taking a look at the two commits separately. One contains the actual meat of the PR and one all the mass changes across the codebase.

Changes across the codebase

The most visible change is how fields of an element are accessed. It results in the following mass changes across the codebase:

elem.field(styles)     => elem.field.get(styles)
elem.push_field(val)   => elem.field.set(val)
Elem::field_in(styles) => styles.get(Elem::field)
Elem::set_field(val)   => styles.set(Elem::field, val)

As you can see, previously we generated a bunch of functions for each field, which are now superseded through direct methods on the field for instance properties and a single associated constant for everything else.

The PR also removes the two field attributes #[borrowed] and #[resolve]. These are not really properties of the field, but rather of an access to a field. This is reflected in the following new APIs:

elem.field.get_ref(styles) // replacement for #[borrowed]
elem.field.resolve(styles) // replacement for #[resolve] and shorthand for
                           // `elem.field.get(styles).resolve(styles)`

This wasn't possible previously, as the accessor function had to have one concrete signature (and we didn't wanna generate even more of them). This demonstrates the increased flexibility of the new, more generic approach.

Stats

The stats below show the reduction in output of the macro (this is adapted output from the new -Z macro-stats compiler feature):

Branch Macro Uses Lines Avg Lines Bytes Avg Bytes

main #[elem] 129 55_043 426.7 3264247 25304.2

remacro #[elem] 129 25_254 195.8 1457616 11299.3
The effect on build times wasn't quite what I would have hoped. I benchmarked a bit and cargo check was a little faster incrementally than before while cargo build wasn't really, but it was too noisy to really say much. I might try to get more stable numbers again later. But as mentioned above, the PR isn't pointless even if compile times are unaffected.
Runtime performance seems to be unaffected from a few quick experiments.
I also checked binary size and was actually quite surprised to see that the PR seems to increase binary size quite a bit (the CLI jumped from 34.7MB to 35.3MB in release mode on my machine). It might be monomorphization bloat. I'll need to check whether I can figure out the cause.

TODO

Figure out why the binary size grew
Possibly do some more compile time benchmarking
Potentially split get into get where E::Type: Copy and get_cloned to encourage use of get_ref in situations where cloning could be expensive

laurmaedje · 2025-07-07T09:30:40Z

Why did binary size grow?

I managed to figure out why the binary size grew. The problem was that a bunch of of field-specific functions were embedded both standalone (because there are function pointers to them in NativeElementData), but also inlined into the trait Bounds impl. This made up for exactly the 600KB extra binary size I was observing.

The Bounds trait allows us to dispatch to element-specific routines in a type-safe way by letting Rust generate a vtable and taking care of the unsafety involved internally. That said, conceptually, NativeElementData by itself already is a vtable and the functions contained within can perfectly be called directly on Content as long as the content holds an element matching the vtable.

So, the Bounds trait was really just a dance to let Rust generate a vtable that then dispatches to our vtable. Unfortunately, this led to code duplication in the binary. This could probably be avoided somehow by making sure the function pointers of our vtable don't end up in data that is used at runtime.

That said, Rust's vtable is also less general than the one implemented by NativeElementData: It can only hold single methods, while NativeElementData also holds plain data and arrays of methods (for each field). Essentially, NativeElementData holds the platonic ideal of a vtable for our use case, but that vtable cannot directly be expressed in safe Rust.

A manual vtable

Based on this insight, I experimented with using our own vtable directly while still making the whole thing as safe as I can manage. The result has a bunch of unsafe code (in the files raw.rs and vtable.rs), but I believe this is worth it, not primarily for the binary size or the slight performance win, but for the flexibility in data layout it gives us now and down the road (when we implement custom types).

Currently, on main, Content is implemented as an Arc<Inner<dyn Bounds>> where Inner holds the element plus some metadata. This is a fat pointer: A pointer to some Inner holding an element and a pointer to a vtable. In this PR, a similar scheme is implemented, but manually: Content is a pointer to inner data plus an Element and an Element is just a pointer to a static holding our custom vtable. This means that the reference counting had to be implemented manually, but it means we can lay out this data however we like.

This is really useful because, when we unify content and values in the future, I want to adopt this new representation for arbitrary Values. And then we also want to have an inline representation without any allocation. E.g. an integer could be an inline i64 plus a pointer to native type data / a vtable. Essentially, we have complete freedom in mixing inline and heap representation, with reference counting, very efficient dynamic dispatch and super-cheap downcasting.¹ Implementing this with Rust's built-in trait objects is basically impossible as far as I can tell.

Implementation

I put effort into making sure the implementation is as safe as possible even if it uses unsafe. All new unsafe code is encapsulated into two files:

The file raw.rs implements the RawContent type: A custom "trait object" with a custom reference counting implementation (which also skips the unnecessary weak ref count contained in an Arc). This was fairly straightfoward as I had already reimplemented Arc in ecow.
The file vtable.rs implements the custom vtable and exposes a safe interface for it: Once you've unsafely acquired a Handle once (where you need to guarantee that element and vtable are matching), all further operations are safe. The unsafe keyword occurs a bunch of times in this file, but most of it is very straightforward wrapper code. The one exception is ContentVtable::erase, where we transmute from a typed to an untyped vtable (the vtable is repr(C) and the only mention of the types are as arguments of unsafe fns). This is the dynamic dispatch aspect of the vtable.

I also implemented a miri-based test (and will add it to CI). All this to say: I did not decide lightly to add this much unsafe code, but I believe it is worth it, not primarily because of the binary size or perf win now, but because of the flexibility it gives us in data layout of the Value type down the road. This is already quite a step in the direction I imagined the Value type to go in the future.

Results

The new implementation fixes the binary size issue. Binary size on main was 34.7 MB. Binary size temporarily rose to 35.3 MB with this PR, but is now down to 34.5 MB.
I benchmarked a few documents and saw a few percent performance gains on average, but tbh it's still too noisy to really say much. A stable suite of benchmark documents that we can trigger in CI on a machine with no other CPU pressure would be really useful to make more informed statements in such cases.

Checking whether the value/element is of a specific type is just a matter of comparing the value/element's vtable pointer to the well-known static pointer to the vtable of the type/element in question. This is a benefit this PR brings down over main, where every downcast involves dynamic dispatch. ↩

This factors out the memoized hash part of `LazyHash` for manual use

saecki

Still looking at the code, but in the mean time I've found a few doc comments that weren't changed.

crates/typst-library/src/foundations/content/vtable.rs

saecki · 2025-07-07T17:29:51Z

So I've looked at the unsafe code in raw.rs and vtable.rs and some of the surroundings, and it seems reasonable 👍
I think this is a very exciting change!

laurmaedje · 2025-07-08T08:51:10Z

On Discord, there was the question whether it's really worth it to add unsafe code to reduce the amount of macro-generated code. For posterity, I want to also clarify here on the PR that this is not what's happening. Rather, the PR needs to be considered in two parts: The rewrite of the #[elem] macro and the rewrite of the Content representation. Those two only happen to be batched together the rewrite of the macro triggered a binary size increase, which made me think about the data layout more, and which could be fixed by switching to a more manual representation. However, that binary size increase is feasible to fix without the unsafe code. That's not its main purpose. So what is the purpose of each change?

Rewriting the macro: The bad compile time is just the tip of the iceberg. Editing macros is a PITA. You need to constantly restart rust-analyzer and reason about compile errors without actually seeing where they happen (because it's always annotated at the macro). So the first part of the PR gets rid of that by letting the macro just generate the minimal amout of data and primitive accessors as a set of building blocks and then normal code composes further behaviour (like a Repr impl, or reading field while considering styles, or materializing a field, ...) This means that working on content stuff means way less touching of the macro. The new, generic approach is also generally nice because it allows for new generic APIs like Content::set and removes stuff like #[borrowed] and #[resolve], making things more uniform. It does increases the lines of source code because there is one extra representation (the one of the primitives), but that's not a big deal in my opinion. So that's the maintainability part and it has not so much to do with the unsafe code.
Rewriting the content representation: This gives us a lot of flexibility to experiment with and optimize the data layout in the future, with value/content unification coming up. Right now, it's not a big perf win because it mostly manualizes how the old implementation worked. But it makes it totally feasible to have stuff like inline stack-allocated representations for values like integers, bools, etc. Today, this is done with an enum of built-ins and one variant is Content, which internally is dynamic. (And then there's also Dynamic, which is the non-content dynamic variant.) I want to unify all of this, while retaining inline storage and extending it to elements (like TextElem, which is just a string internally, but is separately allocated today). There is no good way to do that in safe Rust that I know of. You basically need an enum with one variant being a trait object, but that wastes space (you need one extra int for the enum tag), introduces more dynamic code paths when doing generic operations on values or downcasting and is not so extensible as you need to mention all inline variants in the enum.

I should also mention that saying Typst has very little unsafe code today is somewhat true, but also gives a warmer fuzzy feeling that it maybe should. In typst/typst there is just a little, but in our dependencies there is more, both in-house and external. For example, ecow has more unsafe code than the PR adds (fairly similar one actually). That unsafe code could be omitted, but it would make Typst 5-10% slower (I just made a small experiment on a few documents). Of course, unsafe code should be avoided if possible, but I think the most central type of the compiler deserves to get the best data layout we can think of, and not the best safe Rust can express. That I happened to work on this now, together with the #[elem] macro was mostly coincidental because of the binary size growth.

* build: use nightly typst, typst-ansi-hl, typstyle, reflexo; update ttf-parser, fontdb * refactor: new eval_string api * refactor: typst::html -> typst_html fixup * refactor: new Source lines api in typst/typst/6308 * refactor: new ast::Expr enum names in typst/typst#5784 * refactor: new color api in typst/typst#6479 * refactor: new element api in typst/typst#6547 * refactor: new html frame api in typst/typst#6505 * refactor: new label constructing api in typst/typst#6332 * refactor: new symbol api in typst/typst#6441 * fix: catch_unwind when printing Ty's debug format * styl: fmt * refactor: use new LibraryExt mod fixup 2 * fix: increase recursion limit for tinymist-query * tests: update test snaps * ci: continue on error * build: bump world crates to 0.13.15-rc1 * build: pin reflexo, typstyle fixup 3 * build: bump to 0.13.15-rc1 * docs: changelog * ci: update windows version

* build: use nightly typst, typst-ansi-hl, typstyle, reflexo; update ttf-parser,fontdb * refactor: new eval_string api * refactor: typst::html -> typst_html * refactor: new Source lines api in typst/typst/6308 * refactor: new ast::Expr enum names in typst/typst#5784 * refactor: new color api in typst/typst#6479 * refactor: new element api in typst/typst#6547 * refactor: new html frame api in typst/typst#6505 * refactor: new label constructing api in typst/typst#6332 * refactor: new symbol api in typst/typst#6441 * fix: catch_unwind when printing Ty's debug format * styl: fmt * refactor: use new LibraryExt mod * fix: increase recursion limit for tinymist-query * tests: update test snaps * ci: continue on error * deps: update rustc * tests: update test snaps * deps: patch reflexo * refactor(typlite): replace manual highlight and strike with official impl * tests: update test snaps for shortened empty attributes syntax * tests(typlite): update test snaps to ignore outline temporarily * fix(typlite): convert link * feat: auto release nightly * feat: auto tag stable release

… upstream (#1) * Numbering implementation refactor (typst#6122) * Pin colspan and rowspan for blank cells (typst#6401) * Clean up some parser comments (typst#6398) * Autocomplete fixes for math mode (typst#6415) * Update docs for gradient.repeat (typst#6385) * Document how to escape lr delimiter auto-scaling (typst#6410) Co-authored-by: Laurenz <[email protected]> * Improve number lexing (typst#5969) * Report errors in external files (typst#6308) Co-authored-by: Laurenz <[email protected]> * Table multiple headers and subheaders (typst#6168) * Use the shaper in math (typst#6336) * Standardize trailing slashes in docs route paths (typst#6420) * Make a more descriptive definition of `module` (typst#6380) * Specify which CSL style is not suitable for bibliographies (typst#6306) Co-authored-by: Laurenz <[email protected]> * Adjust source file API surface (typst#6423) * Do not force `math.mid` elements to have the Large math class (typst#5980) * List both YAML file extensions in bibliography docs (typst#6426) * Fix panic when test source is not found in world (typst#6428) * Use `codex::ModifierSet` (typst#6159) Co-authored-by: Laurenz <[email protected]> * Render `#super` as `<sup>`, `#sub` as `<sub>` in HTML (typst#6422) * Warning when watching stdin (typst#6381) Co-authored-by: Laurenz <[email protected]> * Fix untidy Cargo.lock * Warn when using variable fonts (typst#6425) * Check that all translation files are added to TRANSLATIONS and ends with newline (typst#6424) Co-authored-by: Laurenz <[email protected]> * Unify `EvalMode` and `LexMode` into `SyntaxMode` (typst#6432) * Consume `data` argument in `pdf.embed()` (typst#6435) * Better error message for compile time string interning failure (typst#6439) * Ensure that label repr is syntactically valid (typst#6456) * Hint for label in both document and bibliography (typst#6457) * Prefer `.yaml` over `.yml` in the docs (typst#6436) * Fix align link in layout documentation (typst#6451) * Fix param autocompletion false positive (typst#6475) * Encode empty attributes with shorthand syntax * Add `Duration::decompose` * Generic casting for `Axes<T>` * More type-safe color conversions * Add `typst_utils::display` * Support for generating native functions at runtime * Typed HTML API (typst#6476) * Consistent codepoint formatting in HTML and PDF error messages * Test runner support for HTML export errors * Turn non-empty void element into export error * Handle pre elements that start with a newline * Extract `write_children` function * Properly handle raw text elements * Fix stroke cap of shapes with partial stroke (typst#5688) * Adding Croatian translations entries (typst#6413) * Rewrite `outline.indent` example (typst#6383) Co-authored-by: Laurenz <[email protected]> * Use ICU data to check if accent is bottom (typst#6393) Co-authored-by: Laurenz <[email protected]> * Add docs for `std` module (typst#6407) Co-authored-by: Laurenz <[email protected]> * Improve equation reference example (typst#6481) * Add page reference customization example (typst#6480) Co-authored-by: Laurenz <[email protected]> * Bump `krilla` to current Git version (typst#6488) Co-authored-by: Laurenz <[email protected]> * Check that git tree is clean after build (typst#6495) * Also fix encoding of `<textarea>` (typst#6497) * Minor fixes to doc comments (typst#6500) * Fix typos in page-setup.md (typst#6499) * Support `in` operator on strings and modules (typst#6498) * Consistent sizing for `html.frame` (typst#6505) * Allow deprecating symbol variants (typst#6441) * Disallow empty labels and references (typst#5776) (typst#6332) Co-authored-by: Laurenz <[email protected]> * Fix nested HTML frames (typst#6509) * Basic support for text decoration functions in HTML (typst#6510) * Improve sentence in guide for LaTeX users (typst#6511) * Sort line items by logical order when constructing frame (typst#5887) Co-authored-by: Laurenz <[email protected]> * Fix panic when sampling across two coincident gradient stops (typst#6166) * Bump `typst-dev-assets` (typst#6514) * Acknowledgements (typst#6528) * Support HTML tests in test-helper extension (typst#6504) * Fix typo in Advanced Styling docs tutorial (typst#6517) * Fix typo in PDF standard CLI help (typst#6518) * Fix typo in PDF standard CLI help part 2 (typst#6531) * Fix minor typo in `array.product` docs (typst#6532) * Fix typos in calc module docs (typst#6535) * Use "subs" and "sups" font features for typographic scripts (typst#5777) * Use punctuation math class for Arabic comma (typst#6537) * Remove duplicate language computation (typst#6557) * Fix typo in PackageStorage (typst#6556) * Fix nightly warnings (typst#6558) * Fix minor typo in function docs (typst#6542) * Refer to json function instead of deprecated json.decode in groups docs (typst#6552) * Rewrite foundations of native elements (typst#6547) * Target-specific native show rules (typst#6569) * Construct library via extension trait instead of default & inherent impl (typst#6576) * Move `html` module to `typst-html` crate (typst#6577) * Fix typo of Typst domain in quote docs (typst#6573) * Anti-alias clip paths (typst#6570) * Use "displayed" instead of "repeated" to avoid ambiguity in numbering docs (typst#6565) * Ignore spans when checking for RawElem equality (typst#6560) * Add `default` argument for `str.first` and `str.last` (typst#6554) Co-authored-by: Laurenz <[email protected]> * Add completions subcommand (typst#6568) * Update Swedish translations based on defaults used for LaTeX and cleveref (typst#6519) * Move math styling to codex and add `math.scr` (typst#6309) * More consistent `Packed<T>` to `Content` conversion methods (typst#6579) * Support images in HTML export (typst#6578) * Fix tooltip for figure reference (typst#6580) * Complete movement of HTML export code to `typst-html` (typst#6584) * Handle `lower` and `upper` in HTML export (typst#6585) * Deduplicate labels for code completion (typst#6516) * Fix regression in reference autocomplete (typst#6586) * Use "whitespace" instead of "space" to denote block-level equation in docs (typst#6591) * Fix minor typo in text docs (typst#6589) * Rephrase docs for truncation of float/decimal to integer (typst#6543) * HTML frame improvements (typst#6605) * Change `enum.item.number` to `Smart` instead of `Option` (typst#6609) * Support setting fonts repeatedly with different `covers` (typst#6604) * Support intra-doc links in HTML (typst#6602) * Partially automate span assignment in native show rule (typst#6613) * Bump `zip` dependency (typst#6615) * Restore timing scopes for native show rules (typst#6616) * Slightly improve selector docs (typst#6544) * Add show rule for smallcaps in HTML (typst#6600) * Mention Tinymist in README.md (typst#6601) * Fix documentation oneliners (typst#6608) * Add rust-analyzer to flake devShell (typst#6618) * Add Lithuanian translations (typst#6587) * Bump CI Rust to 1.88 * Bump MSRV to 1.88 * Migrate to 2024 edition * Fix 2024 clippy warnings * Yeet `if_chain` macro * Reformat with 2024 edition * Add support for PDF embedding (typst#6623) Co-authored-by: Laurenz <[email protected]> * Add `pdf` extension to image autocompletions (typst#6643) * Fix bounding box computation for lines in curves (typst#6647) Co-authored-by: Laurenz <[email protected]> * Lint for iterations over hash types (typst#6652) * Create constructor methods for manifest types (typst#6625) * Remove unnecessary `comemo` dependency in `typst-syntax` (typst#6668) * Remove duplicate center alignment style for equations (typst#6667) * Fix incorrect `vline.x` docs (typst#6657) * Miscellaneous minor docs improvements (typst#6651) Co-authored-by: Laurenz <[email protected]> * Allow explicit autocomplete immediately after comma and colon (typst#6550) * Improve Guide for LaTeX users, Query Function, and replace invisible emojis (typst#6620) Co-authored-by: PgBiel <[email protected]> Co-authored-by: Laurenz <[email protected]> * Hint that deprecated items will be removed in `0.15.0` (typst#6617) * Specify the standard smart quotes for `Arabic` (typst#6626) * Use `rustc-hash` for hash maps and sets (typst#6678) * Faster constraint checking in comemo (bumps comemo & krilla) (typst#6683) * Add `--target` argument for `typst query` (typst#6405) Co-authored-by: Laurenz <[email protected]> * Require parentheses in all function-like sub/superscripts (typst#6442) Co-authored-by: Laurenz <[email protected]> * Fix several wrong anchors in `$` docs link resolution (typst#6684) * Add `cargo testit` alias for running integration tests (typst#6682) Co-authored-by: Laurenz <[email protected]> * Update guides welcome text to refer to list instead of specific guides (typst#6685) Co-authored-by: Laurenz <[email protected]> * Prevent broken glyph assemblies when font data is incorrect (typst#6688) * Allow custom element names in HTML tag syntax (typst#6676) Co-authored-by: Laurenz <[email protected]> * Add interface to disable timer (typst#6695) Co-authored-by: Derived Cat <[email protected]> * Add support for drawing COLR glyphs in SVG export (typst#6693) * Apply aspect ratio correction for linear gradients in PDF export (typst#6689) * Wrap raw blocks in `<code>` tag additionally to `<pre>` tag (typst#6701) * Support for raw syntax highlighting in HTML export (typst#6691) * Ensure that whitespace is not collapsed in inline raw blocks (typst#6703) * Show aliases of citation styles in docs and completions (typst#6696) * Correct CeTZ spelling (typst#6706) * Add `title` element (typst#5618) Co-authored-by: Laurenz <[email protected]> * Make HTML data structures cheaper to clone (typst#6708) * Move `par`, `box`, and `block` to show rules in HTML export (typst#6709) * Support smartquotes in HTML export (typst#6710) Co-authored-by: Malo <[email protected]> * Avoid dangling reference output for HTML tests (typst#6711) * Support multiple fonts in math (typst#6365) * Add link to position field of grid.hline and grid.vline docs (typst#6670) * Rename `pdf.embed` to `pdf.attach` (typst#6705) * Include numbering in PDF bookmark (typst#6622) Co-authored-by: Laurenz <[email protected]> * Document escaping semicolon, valid identifiers, and `state` tips (typst#6674) Co-authored-by: Andrew Voynov <[email protected]> Co-authored-by: Yaksher <[email protected]> Co-authored-by: Laurenz <[email protected]> * Ensure table headers trigger rowbreak (typst#6687) * Fix phrasing of iff (if and only if) in docs (typst#6713) * Remove redundant "This can be" from stroke docs of curve and polygon (typst#6715) * Fix curve docs for fill (refer to curve instead of rectangle) (typst#6714) * Add span to `html.frame` node (typst#6716) * Fix return type of `missing_method` (typst#6718) * Encoding fixes for HTML raw text elements (typst#6720) * Update RelativeTo to include tiling in docs (typst#6730) * Add missing "the" for cmyk function of color docs (typst#6731) * Remove use of "last" and "end" for conic gradient circle docs (typst#6733) * Fix broken links in docs (typst#6734) * Simplify links in docs (typst#6732) Co-authored-by: Laurenz <[email protected]> * Bump hayro and krilla (typst#6741) * Deduplicate fallback smart quotes (typst#6747) * Move `is_inline` to `HtmlElem` (typst#6748) * Move grid cell locator creation to GridLayouter (typst#6746) * HTML whitespace protection (typst#6750) * Improve code snippets in Table Guide (typst#6658) * More accessible color scheme for raw blocks (typst#6754) Co-authored-by: Laurenz <[email protected]> * Use `stylistic-set: 1` in favor of `ss01` in docs and tests (typst#6766) * Implement fraction styles: vertical, skewed, and horizontal. (typst#6672) * Bump Rust to 1.89 in CI (typst#6775) Co-authored-by: Drodt <[email protected]> Co-authored-by: Daniel Drodt <[email protected]> * Fix typo in doc on quotes parameter of smartquote (typst#6779) * Fix logical order in bidirectional lines (typst#6796) * Extract `trim_weak_spacing` function (typst#6797) * Separate items for hyphens, fixing style of repeated hyphen (typst#6798) * Fix Unicode mapping of hyphenation artifacts (typst#6799) * Do not consider default ignorables when picking last resort font (typst#6805) * Compute width of shaped text on-demand (typst#6806) * Fix `sub` and `super` oneliners (typst#6791) * Ensure that hyphenation is possible after a tag (typst#6807) * Make links to `$grid` in `table` docs more specific (typst#6776) Co-authored-by: PgBiel <[email protected]> * Fix `repr` of `foo.with(..)` (typst#6773) * Fix case in docs serialization (typst#6808) * Add links and examples in the docs of `str` (typst#6751) Co-authored-by: Laurenz <[email protected]> * Hash page instead of frame for watch command (typst#6810) * Update nix flake and use nixfmt (typst#6827) * Update & fix tutorial part 3 & 4 example code (typst#6778) * Improve docs on customizing `raw` (typst#6000) Co-authored-by: Malo <[email protected]> Co-authored-by: Laurenz <[email protected]> * Better handle large numbers (u128/i128) in deserialization (typst#6836) * Follow the comment on setting the `State`'s mask (typst#6642) * Allow augment line at the beginning and end of a matrix (typst#5806) Co-authored-by: Laurenz <[email protected]> * Fix slicing last n elements using count (typst#6838) * Fix auto hanging-indent for centered headings (typst#6839) * Use rust-analyzer from fenix toolchain in flake (typst#6826) * Type safety for logical indices in line layout (typst#6848) * Load linked bitmap images in SVG images (typst#6794) Co-authored-by: Laurenz <[email protected]> * Fix CJ-Latin space at manual line breaks (typst#6700) Co-authored-by: Laurenz Mädje <[email protected]> * Allow multi-character symbols (typst#6489) Co-authored-by: Max <[email protected]> Co-authored-by: Laurenz <[email protected]> * Keep end of line whitespace glyphs (typst#6866) * Fix punctuation in HTML placeholder (typst#6854) * Unify and document the behaviours of `{json,yaml,toml,cbor}.encode` (typst#6743) Co-authored-by: Laurenz <[email protected]> * Improve docs of various numbering parameters (typst#6757) Co-authored-by: Laurenz <[email protected]> * Document and test `Sides<T>` parameters (typst#6862) Co-authored-by: Laurenz <[email protected]> * Remove unused `Styles::set_family` (typst#6879) * Make clear that `Content::query_first` is naive (typst#6880) * Ensure correct tag nesting with grouping rules (typst#6881) * Avoid breaking after an empty frame (typst#6335) * Add test for default ignorables before a breakpoint (typst#6782) * Fix typos (typst#6878) * Improve the docs on function params of `array.{sorted,dedup}` (typst#6756) * Add a link from `math.vec` to `math.{arrow,bold}` in docs (typst#6867) * Add a disclaimer about embedding PDFs (typst#6888) Co-authored-by: Laurenz <[email protected]> * Improve examples in the docs for under/over functions (typst#6889) * Improve docs of `Celled<T>` params of `table` and `grid` (typst#6764) Co-authored-by: Laurenz <[email protected]> * Fix off by one in tag expansion (typst#6900) * More useful `Debug` impls for `Tag` and `Location` (typst#6901) * Fix potential crash in `Debug` impl of `Property` (typst#6902) * Fix `Debug` impl for end tags (typst#6906) * Make `select_where!` usable outside of typst-library (typst#6910) * Fix footnote links in presence of styling (typst#6912) * Consistently ignore styles for tags (typst#6911) * Expand tags even a bit more around groupings (typst#6909) * Don't warn for zero-sized horizontal spacing in HTML export * Use `singleton!` for `HElem::hole` * Add way to attach role to closest element in HTML export * Add logical parent mechanism to HTML export * Generalize mechanism for root styles to HTML export * Support footnotes in HTML export * Add `default` argument to `array.join` (typst#6932) * Replace the non-existent `$math.arrow` with `$math.accent` in the docs for vector (typst#6918) * Initial Hayagriva bump (typst#6920) Co-authored-by: Laurenz <[email protected]> * Use different colors for object keys and string-based values in JSON listings (typst#6873) * Support for outline in HTML (typst#6606) * Use Title Case for Doc Pages (typst#6936) * Improve the docs for dictionary (typst#6899) * Add an illustration for `par.{leading,spacing}` in docs (typst#6915) Co-authored-by: Andrew Voynov <[email protected]> Co-authored-by: Laurenz <[email protected]> * Initial plan --------- Co-authored-by: Sam Ireson <[email protected]> Co-authored-by: PgBiel <[email protected]> Co-authored-by: Ian Wrzesinski <[email protected]> Co-authored-by: cAttte <[email protected]> Co-authored-by: Andrew Voynov <[email protected]> Co-authored-by: Laurenz <[email protected]> Co-authored-by: Tobias Schmitz <[email protected]> Co-authored-by: Max <[email protected]> Co-authored-by: 3w36zj6 <[email protected]> Co-authored-by: Malo <[email protected]> Co-authored-by: T0mstone <[email protected]> Co-authored-by: Lachlan Kermode <[email protected]> Co-authored-by: Y.D.X. <[email protected]> Co-authored-by: Ilia <[email protected]> Co-authored-by: Noam Zaks <[email protected]> Co-authored-by: Wannes Malfait <[email protected]> Co-authored-by: Ivica Nakić <[email protected]> Co-authored-by: +merlan #flirora <[email protected]> Co-authored-by: Connor K <[email protected]> Co-authored-by: Said A. <[email protected]> Co-authored-by: Florian Bohlken <[email protected]> Co-authored-by: Robin <[email protected]> Co-authored-by: Adrián Delgado <[email protected]> Co-authored-by: frozolotl <[email protected]> Co-authored-by: Jassiel Ovando <[email protected]> Co-authored-by: Patrick Massot <[email protected]> Co-authored-by: Erik <[email protected]> Co-authored-by: pog102 <[email protected]> Co-authored-by: Laurenz Stampfl <[email protected]> Co-authored-by: Niklas Eicker <[email protected]> Co-authored-by: Marcono1234 <[email protected]> Co-authored-by: Igor Khanin <[email protected]> Co-authored-by: hpcfzl <[email protected]> Co-authored-by: zefr0x <[email protected]> Co-authored-by: Abdul-Rahman Sibahi <[email protected]> Co-authored-by: 枚鴉 <[email protected]> Co-authored-by: Myriad-Dreamin <[email protected]> Co-authored-by: Derived Cat <[email protected]> Co-authored-by: Sebastian Eberle <[email protected]> Co-authored-by: Johann Birnick <[email protected]> Co-authored-by: Yaksher <[email protected]> Co-authored-by: Tau <[email protected]> Co-authored-by: Martin Haug <[email protected]> Co-authored-by: Théophile Cailliau <[email protected]> Co-authored-by: Drodt <[email protected]> Co-authored-by: Daniel Drodt <[email protected]> Co-authored-by: ultimatile <[email protected]> Co-authored-by: Poh <[email protected]> Co-authored-by: Clemens Koza <[email protected]> Co-authored-by: Siddhant Agarwal <[email protected]> Co-authored-by: Philipp Niedermayer <[email protected]> Co-authored-by: Toon Verstraelen <[email protected]> Co-authored-by: Eric Biedert <[email protected]> Co-authored-by: Jan Romann <[email protected]> Co-authored-by: copilot-swe-agent[bot] <[email protected]>

laurmaedje force-pushed the remacro branch 2 times, most recently from 6a000bd to 56d4714 Compare July 2, 2025 18:49

MDLC01 mentioned this pull request Jul 2, 2025

Replace non-#[internal] #[synthesized] fields with getter methods #6351

Closed

laurmaedje force-pushed the remacro branch 3 times, most recently from c5ddfc2 to feb25ec Compare July 7, 2025 09:30

laurmaedje force-pushed the remacro branch 4 times, most recently from 93c21c2 to d0150de Compare July 7, 2025 10:04

laurmaedje requested a review from saecki July 7, 2025 10:04

laurmaedje force-pushed the remacro branch from d0150de to 0e9800d Compare July 7, 2025 10:13

laurmaedje added 3 commits July 7, 2025 13:55

Mass changes

d273691

Introduce HashLock type

2f401f2

This factors out the memoized hash part of `LazyHash` for manual use

New #[elem] implementation

533f17c

laurmaedje force-pushed the remacro branch from 0e9800d to 1fd9a10 Compare July 7, 2025 11:56

saecki reviewed Jul 7, 2025

View reviewed changes

crates/typst-library/src/foundations/content/vtable.rs Outdated Show resolved Hide resolved

crates/typst-library/src/foundations/content/vtable.rs Outdated Show resolved Hide resolved

laurmaedje added 2 commits July 8, 2025 10:49

New Content implementation

d12cad6

Miri test

1955946

laurmaedje force-pushed the remacro branch from 1fd9a10 to 1955946 Compare July 8, 2025 08:49

laurmaedje enabled auto-merge July 8, 2025 08:51

laurmaedje added this pull request to the merge queue Jul 8, 2025

Merged via the queue into main with commit 0a3c693 Jul 8, 2025
16 checks passed

laurmaedje deleted the remacro branch July 8, 2025 09:11

laurmaedje mentioned this pull request Jul 10, 2025

Complete movement of HTML export code to typst-html #6584

Merged

ParaN3xus added a commit to ParaN3xus/tinymist that referenced this pull request Jul 15, 2025

refactor: new element api in typst/typst#6547

de43154

ParaN3xus added a commit to ParaN3xus/tinymist that referenced this pull request Jul 15, 2025

refactor: new element api in typst/typst#6547

c0a0b27

ParaN3xus added a commit to ParaN3xus/tinymist that referenced this pull request Jul 16, 2025

refactor: new element api in typst/typst#6547

d1aafd3

ParaN3xus mentioned this pull request Jul 16, 2025

New ContentVtable crashed Ty's debug format Myriad-Dreamin/tinymist#1910

Closed

laurmaedje mentioned this pull request Jul 17, 2025

#[internal] fields are accessible by the user in function.where #6350

Closed

1 task

ParaN3xus added a commit to ParaN3xus/tinymist that referenced this pull request Jul 28, 2025

refactor: new element api in typst/typst#6547

f538618

ParaN3xus added a commit to ParaN3xus/tinymist that referenced this pull request Jul 30, 2025

refactor: new element api in typst/typst#6547

31857bd

ParaN3xus added a commit to ParaN3xus/tinymist that referenced this pull request Jul 31, 2025

refactor: new element api in typst/typst#6547

faf138d

ParaN3xus added a commit to ParaN3xus/tinymist that referenced this pull request Jul 31, 2025

refactor: new element api in typst/typst#6547

6103085

ParaN3xus added a commit to ParaN3xus/tinymist that referenced this pull request Jul 31, 2025

refactor: new element api in typst/typst#6547

d831437

Vanille-N pushed a commit to Vanille-N/typst that referenced this pull request Aug 14, 2025

Rewrite foundations of native elements (typst#6547)

0d72786

Andrew15-5 mentioned this pull request Oct 1, 2025

Ability to pass any value type to type-limited field in where clause #6991

Open

1 task

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Rewrite foundations of native elements#6547

Rewrite foundations of native elements#6547
laurmaedje merged 5 commits intomainfrom
remacro

laurmaedje commented Jul 2, 2025 •

edited

Loading

Uh oh!

laurmaedje commented Jul 7, 2025

Uh oh!

saecki left a comment

Uh oh!

Uh oh!

Uh oh!

saecki commented Jul 7, 2025

Uh oh!

laurmaedje commented Jul 8, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Branch	Macro	Uses	Lines	Avg Lines	Bytes	Avg Bytes
main	`#[elem]`	129	55_043	426.7	3264247	25304.2
remacro	`#[elem]`	129	25_254	195.8	1457616	11299.3

Uh oh!

Conversation

laurmaedje commented Jul 2, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Changes across the codebase

Stats

TODO

Uh oh!

laurmaedje commented Jul 7, 2025

Why did binary size grow?

A manual vtable

Implementation

Results

Footnotes

Uh oh!

saecki left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

saecki commented Jul 7, 2025

Uh oh!

laurmaedje commented Jul 8, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

laurmaedje commented Jul 2, 2025 •

edited

Loading