fix: make ?cpuMultiplier= run the CPU faster than the peripherals again

[mattgodbolt]

?cpuMultiplier= is meant to run the CPU faster relative to the peripherals, leaving video, sound and VIA timing at real-world rates. That takes two halves working together:

1. main.js feeds cpuMultiplier * cpuSpeed cycles per wall-clock second into execute().
2. Cpu6502.buildPolltime() divides videoCycles and peripheralCycles back down by this.cpuMultiplier, so peripherals still see real time.

Half 2 was dead. emulationConfig captured the cpuMultiplier local by value before the query string was parsed into it, so config.cpuMultiplier was always 1. Only half 1 applied, and the whole machine sped up, sound pitch and screen refresh included.

Archaeology

• 2538773 (2015-10-07) introduced the feature. The parse sat above the emulationConfig literal, so it worked.
• 4b0b301 ("Modernise the code - let and const", Modernise the code - let and const #373, 2022-09-01) hoisted the literal from ~line 560 to ~line 152, above the parse, as part of the var-to-const conversion. Nothing else in that commit touched the feature. That is the regression.
• 076750c (refactor: move per-session settings off the shared Model singletons #714) later added a comment describing the broken state as deliberate. Removed here.

README.md's long-standing caveat that "disc loads become unreliable with a too-slow CPU" only makes sense under the 2015 semantics, where the FDC runs at real time while the CPU is slowed.

Changes

• Resolve cpuMultiplier from parsedQuery where it is declared, so the config cannot capture a stale value, and drop the now-false comment.
• keyboard.js: paste timing no longer multiplies by cpuMultiplier. The paste task lives on the processor's scheduler, which is polled with peripheral cycles, so its delays were already in real time; the multiplier would have stretched paste by the multiplier now that it is live. This path had only ever run at 1.
• Plumb cpuMultiplier through fake6502 / MachineSession, alongside tube, and scale TestMachine.type()'s key hold, which is counted in CPU cycles rather than scheduler ticks.
• New tests/unit/test-6502.js pins the ratio behaviourally: at multiplier 2, 1000 CPU cycles advance the scheduler epoch and video by 500; at 1 the unscaled path passes every cycle straight through.

Sanity-checked headless at multiplier 2: the machine boots, PRINT 6*7 gives 42, and BASIC's TIME advances by the same amount for twice as many CPU cycles.

Behaviour change

This changes emulation behaviour for anyone currently passing ?cpuMultiplier=, restoring the 2015 semantics: the CPU speeds up relative to the peripherals rather than the whole machine speeding up.

Fixes #716

🤖 Generated with Claude Code

[Copilot]

Pull request overview

Restores the intended ?cpuMultiplier= behavior where the CPU runs faster/slower relative to peripherals (video/sound/VIA timing stays real-time), fixing a regression where cpuMultiplier never reached Cpu6502 via emulationConfig.

Changes:

• Resolve cpuMultiplier directly from parsedQuery before constructing emulationConfig, and remove the now-incorrect comment in src/main.js.
• Adjust paste/autotype timing to remain real-time under a live CPU multiplier, and plumb cpuMultiplier through headless/test wiring (MachineSession/fake6502).
• Add/adjust unit tests to pin the CPU-vs-peripheral scaling behavior and update existing keyboard tests accordingly.

Reviewed changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
tests/unit/test-keyboard.js	Updates paste timing expectations to be based on peripheral cycles (real-time), not CPU multiplier.
tests/unit/test-6502.js	Adds new unit coverage for peripheral/video scaling under cpuMultiplier.
tests/test-machine.js	Scales key-hold timing in TestMachine.type() by cpuMultiplier to keep holds constant in real time.
src/main.js	Fixes config wiring so config.cpuMultiplier reflects the parsed query value.
src/machine-session.js	Exposes cpuMultiplier as a per-session option and forwards it into TestMachine.
src/keyboard.js	Removes multiplier scaling from paste delays (scheduler already runs at peripheral rate).
src/fake6502.js	Passes cpuMultiplier into the CPU config for headless/test instances.
README.md	Clarifies that cpuMultiplier is relative to peripherals, not whole-machine speed.

Comments suppressed due to low confidence (1)

tests/test-machine.js:419

• As above for the BBC path, Atom holdCycles can become non-integer for fractional cpuMultiplier values. Ensure the cycle count is integral before using it for runFor()/cycle comparisons.

        const holdCycles = 80000 * this.processor.cpuMultiplier; // Atom at 1 MHz needs longer hold than BBC at 2 MHz