Skip to content
formicarium

0002. Interpreter only; defer the JIT

  • Status: accepted (scope 2026-10-03, deferral 2026-10-05)
  • Decided by: the user

Context

Section titled “Context”

A JIT was always planned; its timing was to be decided from measurements. They showed that more than 90% of an aube install is interpretation and that it takes about 75 times as long as native. Against CheerpX 1.3.9, install was about as fast as CheerpX’s slowest run and frozen install was 1.5 to 4 times slower. Correctness issues were still open, and the core was planned to be replaced by paludarium.

Decision

Section titled “Decision”

Build the core as an interpreter only (--disable-jit). Defer the JIT, and do not develop it in the C fork of blink. For now the threshold is “good enough to reproduce a report”: the four aube #1645 steps finish in 12 to 20 seconds.

Since the switch to paludarium, the blink fork and its --disable-jit build are no longer the core. The deferral still holds: formicarium does not enable or develop a JIT of its own.

Consequences

Section titled “Consequences”
  • Correctness can come first.
  • It does not reach everyday speed (1 to 2 seconds per command); that needs both a JIT and a resident core.
  • The JIT investment goes to paludarium.

Alternatives rejected

Section titled “Alternatives rejected”
  • A JIT in the C fork now: likely thrown away when the core is replaced, and it adds correctness risk.
  • Parity with CheerpX as the threshold: frozen install cannot reach it without a JIT or another speed-up first.