chore(diagram-core): seal EngineRenderOptions before it is published

[kookyleo]

Takes nits 1 and 3 from the #245 review. Both are things a publish cannot undo, and neither is published yet — crates.io still has supramark-diagram-core 0.1.0, which predates EngineRenderOptions entirely — so this is the last cheap moment for either.

#[non_exhaustive] on EngineRenderOptions

The struct is a bag of optional readability flags, so a second flag is a question of when, not whether. Without #[non_exhaustive], adding one breaks every struct literal in every downstream crate and forces a major release of diagram-core.

Sealing it costs one thing: a literal is no longer available outside the crate, and ..Default::default() is not a way around that for a #[non_exhaustive] struct. So this adds with_edge_label_decluster, and the single external literal (a mermaid-little test) now reads:

let opts = EngineRenderOptions::default().with_edge_label_decluster(true);

EngineRenderOptions::default() was already the spelling everywhere else.

Manifest metadata

supramark-diagram-core was missing homepage and repository — cargo publish --dry-run warns about it, and a published version cannot gain them. Same values as mermaid-little and supramark-font-metrics.

No version changes

0.1.1 and 11.14.0-6 are still unpublished, so this edits them in place rather than bumping again.

cargo clippy -p supramark-diagram-core -p mermaid-little --all-targets -- -D warnings and cargo fmt --check are clean.

Unrelated note: a workspace-wide clippy on this machine also reports 8 pre-existing errors in plantuml-little (unused doc comment, collapsible if, map_or, four ?-operator blocks, a needless &mut on ensure_create_participant), all in tree changes that predate this branch and none of them in a crate this PR touches. CI is green, so it is presumably a newer local clippy; noting it so it is not a surprise later.

🤖 Generated with Claude Code

[g65537]

Approving. I treated each claim as something to re-derive locally rather than accept; nothing blocking.

Premise holds. crates.io API (/api/v1/crates/<name>/versions) today: supramark-diagram-core has only 0.1.0; mermaid-little tops out at 11.14.0-5; supramark-diagram at 0.1.5. So 0.1.1 / 11.14.0-6 / 0.1.6 are unpublished and editing them in place without a bump is correct. I also pulled the three published .crate files and grepped them: diagram-core 0.1.0 has no EngineRenderOptions at all, and the only struct literal of it anywhere on crates.io is the one in mermaid-little 11.14.0-5's #[cfg(all(test, feature = "metrics-ttf-parser"))] module, which no consumer compiles. Its non-test code only reads opts.edge_label_decluster, which #[non_exhaustive] does not affect. So sealing the struct now breaks nobody, published or not.

Every construction site still compiles. grep -rn EngineRenderOptions over the workspace finds exactly: the trait/registry signatures and setter in diagram-core, EngineRenderOptions::default() and the new default().with_edge_label_decluster(true) in mermaid-little's tests, and a bare re-export in supramark-diagram. No exhaustive match/destructure anywhere. cargo clippy -p supramark-diagram-core -p mermaid-little --all-targets -- -D warnings and cargo fmt --check clean; cargo test -p supramark-diagram-core 3/3; cargo test -p mermaid-little --lib render_options_tests 2/2 both with default features and with --no-default-features (the dev-dep self-cycle still turns on metrics-ttf-parser, so the test is not silently skipped). The rewritten test still asserts what it did (SVG mime + <svg with decluster on).

Escape hatch is sufficient, checked from outside the workspace. A scratch crate under /tmp with a path dependency on the branch's diagram-core: EngineRenderOptions::default().with_edge_label_decluster(true), direct field assignment through a mut binding, and the { edge_label_decluster, .. } rest pattern all compile and run; a struct literal, ..Default::default() functional update, and an exhaustive { edge_label_decluster } pattern are rejected with E0639 / E0639 / E0638 respectively. With one bool field, Default plus the setter (or field assignment) reaches every value of the type. The with_* by-value mut self -> Self shape matches RenderOptions::with_edge_label_decluster in mermaid-little and the with_* setters in font-metrics / plantuml-little.

Manifest. cargo publish -p supramark-diagram-core --dry-run (with verification) packages 5 files and no longer warns; RUSTDOCFLAGS=-D warnings cargo doc -p supramark-diagram-core is clean, so the Self::edge_label_decluster intra-doc link resolves. homepage/repository values match mermaid-little / plantuml-little / d2-little.

Pre-existing clippy noise confirmed. git diff origin/main..HEAD touches only the three files in this PR; a workspace-wide cargo clippy --all-targets -- -D warnings on local clippy 0.1.97 fails in dagre (2 errors: sort_by_key, manual Option::filter) and in plantuml-little (the 8 you listed); cargo clippy -p plantuml-little on main reproduces the same 8. Neither crate is touched here, and CI's pinned clippy is green.

Merging with a merge commit once the two Test & Build jobs finish, as with #237/#238/#239/#244/#245. Non-blocking notes in a separate comment.

[g65537]

Non-blocking notes. The first one is the only "now or never" item I see; the rest can be taken in any later patch release.

1. Copy and Eq on EngineRenderOptions are also one-way doors. #[non_exhaustive] makes adding a field non-breaking, but only if the new field keeps every derived trait: a String/Vec field forces dropping Copy, an f32/f64 (say a label-padding factor) forces dropping Eq, and removing either derive is itself a breaking change. If you are confident every future hint is a bool/small Copy scalar, keep them; if not, dropping Copy (and possibly Eq) before 0.1.1 is free today and a semver-major later. The trait takes &EngineRenderOptions, so nothing in the workspace needs Copy. Adding a derive back later is non-breaking, so erring on the side of fewer derives costs nothing.

2. with_edge_label_decluster could be pub const fn, matching mermaid-little's own RenderOptions::with_edge_label_decluster (which already compiles on the 1.82 MSRV). Adding const later is non-breaking, so purely cosmetic.

3. Manifest: still no documentation = "https://docs.rs/supramark-diagram-core" (supramark-diagram has it), keywords, categories, or readme. None is irreversible: they can land with the next version.

4. Reminder for the actual publish, since this PR now makes the three tree versions final: order must be supramark-diagram-core 0.1.1 -> mermaid-little 11.14.0-6 -> supramark-diagram 0.1.6, then yank mermaid-little 11.14.0-5 and supramark-diagram 0.1.5 (both import EngineRenderOptions from a ^0.1 core that only has 0.1.0 on crates.io, so they cannot build today; after 0.1.1 ships they resolve to it and build again, which is fine but the yank keeps cargo update from ever selecting a pair that was never tested together).