Fetching documentation…
Status: active · Priority: high
Core promise: a Morph binary contains exactly what the app uses — nothing else. A hello-world app (single
h1) must never exceed 150KB without UPX (a 120KB no-UPX binary was achieved before, so the budget is proven, not aspirational). UPX size is reported separately as a bonus, never counted toward the budget.
tests/runtime/size/hello-size/ — single-h1 fixture, the measuring stick.tests/runtime/check-size.sh — builds it without --self-test,
asserts ≤153600 bytes, prints file bytes + UPX size. Last gate before
every lean commit.-Os vs -O2)No global default — the feature scan already knows whether the app needs performance:
-Os: no animation, transform, forge, or list anywhere.
Static content pays nothing to be made "faster".-O2: any of those present (unrolling/inlining pay off there).Derived in morph-build from the defines list it already receives
(zero new plumbing). Static mode keeps its -Os path. No manual knob
(a knob users set wrong is worse than a scan that never misclassifies).
| Flag | Detected from | Gates |
|---|---|---|
MORPH_FEATURE_REACTIVITY |
state/shared/event/effect/channel/mid decls, reactive bindings, conditionals, lists, animations | effect.cpp TU skip + whole-file guard; template omits run_pending_effects() / destroy_all_effects() |
MORPH_FEATURE_TASKS |
timers, async/await/Promise, fetch (implied) |
task.cpp TU skip + guard; template omits process_tasks() |
MORPH_FEATURE_NET |
fetch( (implies TASKS) |
net.cpp TU skip + whole-file guard |
MORPH_FEATURE_OWNERSHIP |
parent/modal/role in opts, route windowConfig, [window] defaults |
Follow chain (gated pos-callback registration lets GC drop it); close/create plain paths when off (zero popup bytes) |
MORPH_FEATURE_PAGECACHE |
navigation.cache != 0 (config, no scan) |
navigate plain destroy path when 0 |
MORPH_FEATURE_EVENT (new) |
node events non-empty |
events.cpp TU skip; mouse-button callback registration |
| existing CSS flags | unchanged | unchanged |
Deliberately not gated: window system (every app opens a window),
JsValue/types (props flow through them), header-only code (zero bytes
when unused), dev builds (all-on rule stands).
FeatureSet::scan (IR decls, reactive markers, body
markers) + note_* for routes/config + build.rs additions.required_defines() arms; template {% if %} for loop
calls, channel.h, setColor; codegen plain-path branches
(navigate, close, create); mount MountScope/teardown gating.runtime_sources_with_features skips TUs by defines;
strip --strip-unneeded every release link (~77KB of symtab on
hello-size); morph build --self-test opts the test body back in
(default lean; run-selftests.sh passes the flag).#ifdefs (animation.cpp precedent);
unconditional-trivial API stays (3-line setters cost nothing and
keep generated code compilable); window.cpp callback registrations
follow input/event/hover features so bodies GC-drop.| Step | hello-size file bytes |
|---|---|
| Baseline (self-test in binary, symtab) | 371,100 |
--self-test opt-in (default lean) |
305,100 |
strip on every release link |
227,700 |
| Template branches + TU skips + header gating | 162,100 |
| Callback registration gating | 153,900 |
| Forge backend skip | 145,700 |
-Oz + -flto + -fno-exceptions/-rtti/-unwind + regex-header split |
112,800 |
| Static: HarfBuzz skipped (no shaping content) | 1,049,800 → 560,500 |
Static: GLFW gamepad DB stubbed + unwind-free dep builds + static -Oz |
→ 397,500 |
| Static UPX | ~177,000 |
-Os/-O2 conditional rule + -fno-exceptions try (keep iff the
hello link succeeds and tests stay green).forge.cpp unless chosen; verify flash
isn't load-bearing under forge).logic_prelude.h only with startup_logs/native code;
windowConfig JsObject only with native code.glad.c slimming — almost certainly not worth it (correctness
minefield for ~8KB); listed so nobody re-investigates.