Make the functions of a TrackedValue on their first use

[NullVoxPopuli-ai-agent]

tracked(value) allocates less. Each TrackedValue made four arrow functions and one options object. Now it makes a function only when code reads it.

This is a part of #21650, and it merges alone. Only tracked-value.ts and its test change.

case	main	this PR
create: 1000 signals	17.84 µs	9.39 µs (0.5x)
create: 1000 computeds, read each	41.11 µs	40.06 µs (1.0x)
rows: 1000 rows, write all	118.18 µs	111.87 µs (0.9x)
weighted geometric mean	1.0x	0.9x

The gain is in the create of a tracked value. Reads and writes do not change.

What changes

• get, set, update and freeze are accessors that make the bound function on its first read, and keep it.
• Each accessor has a setter, so code can still assign to them. A write through value uses an assigned set, as on main.
• The default options are one frozen object for all instances.

What changes for callers

• The four functions are still bound, and each read gives the same function. The test "methods can be detached from the instance" from RFC 1071 still passes.
• They are now accessors on the prototype, so Object.keys() and spread do not list them.

All 20 cases

case	main	this PR
propagate: 1 chains x 1 deep	124 ns	122 ns (1.0x)
propagate: 10 chains x 10 deep	5.51 µs	5.49 µs (1.0x)
propagate: 100 chains x 100 deep	724.68 µs	713.81 µs (1.0x)
propagate: 1 chains x 1000 deep	57.18 µs	57.57 µs (1.0x)
propagate: 1000 chains x 1 deep	103.01 µs	100.65 µs (1.0x)
kairo: avoidable propagation	382 ns	374 ns (1.0x)
kairo: broad propagation	7.35 µs	7.25 µs (1.0x)
kairo: deep propagation	3.09 µs	3.04 µs (1.0x)
kairo: diamond	428 ns	422 ns (1.0x)
kairo: mux	20.23 µs	20.17 µs (1.0x)
kairo: repeated observers	365 ns	332 ns (0.9x)
kairo: triangle	755 ns	749 ns (1.0x)
kairo: unstable	574 ns	577 ns (1.0x)
rows: 1000 rows, write 1	7.19 µs	7.09 µs (1.0x)
rows: 1000 rows, write all	118.18 µs	111.87 µs (0.9x)
batch: 10 writes, 1 output	667 ns	634 ns (1.0x)
avoidable: write the same value	6 ns	6 ns (1.1x)
create: 1000 signals	17.84 µs	9.39 µs (0.5x)
create: 1000 computeds, read each	41.11 µs	40.06 µs (1.0x)
create: 1000 outputs	50.30 µs	52.33 µs (1.0x)
weighted geometric mean	1.0x	0.9x

The benchmark is https://github.com/NullVoxPopuli-ai-agent/ember-reactivity-bench. One measurement is the writes of one frame, then one flush that brings every output up to date. Each case of each build runs in its own process, pinned to one core. The numbers are the median of 6 mirrored rounds on Node 24.20.

The run has main, each of the five branches, and #21650 as columns: table and method.

Tests

• The full suite passes locally: 9498 pass, 18 skipped, 0 failed.
• type-check:internals, ESLint and Prettier pass.
• New tests: each function is the same on each read, each function can be replaced, and a detached freeze() prevents updates.

The parts of #21650

#21650 has all five changes in one branch. It stays open as the reference for the combined numbers. The five branches merge with no conflict, and their merge is the same code as the head of #21650.

PR	change	reactivity benchmark, mean against main
#21663	Pool the trackers of tracking frames	0.6x
#21664	Reuse the combined tag of a tracking frame. Needs #21663	0.6x, with the pool
#21665	Index loop over the subtags of a combined tag	1.0x
#21666 (this PR)	Make the functions of a TrackedValue on their first use	0.9x, create is 0.5x
#21667	Store 0 first in the value field of a TrackedValue	1.0x, removes one deopt
#21650	all five	0.6x

Rendering was measured for the combined change only: pnpm bench shows 2.1% less script time, in this comment on #21650.

🤖 Generated with Claude Code