Canvas: consolidate the drawing state into a State struct

[bkaradzic-microsoft]

Follow-up to #1824, as agreed in review.

State struct

save()/restore() had to rewind seventeen separate wrapper-side members, and did it by hand: Save() built a SavedStyle from a positional initializer list, Restore() copied each field back out one at a time. The field list was written out three times over -- the members, the SavedStyle mirror, and the two copy loops -- so adding an attribute meant editing four places, and missing one silently produced an attribute that never rewound. That is exactly the bug class #1824 spent several commits fixing.

They are now a single State struct, pushed and popped whole:

void Context::Save(const Napi::CallbackInfo&)
{
    nvgSave(*m_nvg);
    m_savedStates.push_back(m_state);
}

void Context::Restore(const Napi::CallbackInfo&)
{
    nvgRestore(*m_nvg);
    m_isClipped = false;
    if (!m_savedStates.empty())
    {
        m_state = std::move(m_savedStates.back());
        m_savedStates.pop_back();
    }
}

Save/Restore no longer name a single attribute, so a new one is rewound correctly by construction.

No behavior change: the saved set is exactly what SavedStyle held. The current path stays outside the stack, since per spec save() does not save it, and m_fonts stays out too -- it is a context-wide face cache, not an attribute.

globalAlpha (found by the refactor)

Collecting the fields made a dead one obvious: globalAlpha was pushed and popped by every save()/restore(), but had zero readers. It was registered write-only --

InstanceAccessor("globalAlpha", nullptr, &Context::SetGlobalAlpha),

-- so ctx.globalAlpha read back undefined, and the setter passed the value straight to nvgGlobalAlpha without ever mirroring it.

Registered the getter and mirrored the value. Rendering was already correct (nvgSave/nvgRestore rewinds nanovg's own copy), so this only changes what the attribute reports.

While validating the value I also applied the spec's range rule: a value that is not finite, or outside [0, 1], is ignored and leaves the previous value in place, rather than being clamped or thrown. The old code forwarded NaN and out-of-range values to nanovg unchecked.

Validation

• Unit tests: 21/21 suites, 45 JS assertions (was 43).
• Visual sweep: ran=305 passed=305 failed=0; "Native Canvas" unchanged at 1.850%.
• A/B: reverting only the C++ change fails exactly the three globalAlpha assertions and nothing else, so the tests are actually gated on the fix.

[Copilot]

Pull request overview

This PR refactors the Canvas2D Context wrapper-side drawing state into a single State struct that is saved/restored as a whole, and fixes globalAlpha to be readable (with spec-compliant range/finite-value handling).

Changes:

• Consolidate wrapper-side drawing attributes into Context::State and push/pop that in save()/restore().
• Register globalAlpha getter, mirror its value in wrapper state, and ignore invalid values per spec.
• Extend Canvas2D unit tests to cover globalAlpha readback, save/restore, and invalid-value behavior.

Reviewed changes

Copilot reviewed 3 out of 4 changed files in this pull request and generated 1 comment.

File	Description
Polyfills/Canvas/Source/Context.h	Introduces State struct and saved-state stack; adds GetGlobalAlpha declaration.
Polyfills/Canvas/Source/Context.cpp	Switches attribute storage to m_state, updates save/restore to push/pop State, and implements globalAlpha getter + validation.
Apps/UnitTests/JavaScript/src/tests.javaScript.all.ts	Adds assertions/tests covering globalAlpha readback and invalid values.
Apps/UnitTests/JavaScript/dist/tests.javaScript.all.js	Updates built test output to match the TypeScript test changes.

Suppressed comments (1)

Polyfills/Canvas/Source/Context.h:150

• State defaults for miterLimit and lineWidth are 0, but Canvas2D defaults are miterLimit=10 and lineWidth=1. With the current defaults, new contexts will report incorrect values until explicitly set, even though nanovg likely renders with its own defaults.

            float miterLimit{0.f};
            float lineWidth{0.f};

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

Pull request overview

Copilot reviewed 3 out of 4 changed files in this pull request and generated no new comments.

Suppressed comments (1)

Previously missed (1) — in code that hasn't changed since the last review.

Polyfills/Canvas/Source/Context.h:155

• Now that globalAlpha is readable, storing it as float makes valid JavaScript values fail to round-trip: assigning 0.1 reports approximately 0.10000000149. The attribute is an unrestricted double; keep the wrapper-side value as double and cast only when passing it to NanoVG. The new tests currently use binary-exact values (0.5, 0.25), so they do not expose this loss of precision.

            float globalAlpha{1.f};

[bghgary]

[Commented by Copilot on behalf of @bghgary]

This is still on master at c8a8071 — SetGlobalAlpha narrows before assigning into the double, so ctx.globalAlpha = 0.1 reads back 0.10000000149011612.

m_state.globalAlpha = alpha;
nvgGlobalAlpha(*m_nvg, static_cast<float>(m_state.globalAlpha));

The three assertions use 1, 0.5 and 0, all exact in binary float, so they pass either way.