Where your interpreter’s time actually goes#

Before guessing, look at a real measurement#

“Probably parsing.” “Probably the pattern matcher.”

It’s tempting to guess. Resist it once — a real measurement exists, and it does not confirm the obvious guess.

Call and return bookkeeping — ~24.85%#

RCALL, RRTURN, PROC, and the PUSH/POP operations behind them.

RCALL alone: 8.927% RRTURN alone: 6.182%

Descriptor move and fetch — ~24.71%#

GETD and its relatives — the machinery that copies an 8-byte tagged value from one place to another.

Together: roughly half of everything measured.

Pattern matching itself is a smaller slice than either.

What it doesn’t tell you#

A 1969-vintage design, running 1981-vintage assembler, on 1981-vintage hardware, on whatever workload the Arizona project happened to run.

Evidence that call/return and value-shuffling are plausible places for time to go — not a prediction that your OCaml profile will match.

Not bolted on afterward#

A mark-and-compact collector, driven by an explicit stack rather than the call stack.

GCREQ to request a collection. GCGOT to record what it reclaimed. Built into the runtime.

Why the name is worth keeping#

“Regeneration” describes what the collector does to the heap — compacting live descriptors back into a dense region.

“Garbage collection” describes only the other direction: what it removes.

Did checkpoints 1–3 need this? No.#

A let-bound OCaml value doesn’t need a hand-written collector — OCaml already has one.

It becomes live only if your final project’s value type manages its own memory outside what OCaml gives you for free.

Option 1 — Pattern-matching completeness#

BREAKX: behaves like BREAK(s), but can extend itself past the break character on a rematch.

Exactly equivalent to BREAK(s) ARBNO(LEN(1) BREAK(s)), composed from primitives you already have.

Option 2 — A quickscan-style heuristic#

Week 9 covered quickscan and fullscan as optimizations over exhaustive backtracking search.

Implement one for real — with a test suite confirming identical results to unabridged retry, on cases where the two could plausibly disagree.

Option 3 — A small collector#

If your extension gives the value type its own managed structures: a simple mark-and-sweep pass over them.

Not the full “regeneration” design — that would be its own multi-week project.

Option 4 — Performance work, profiled against real numbers#

Last week’s differential-testing harness plus a profiler. Measure where your interpreter spends its time. Optimize the largest measured cost.

Report before-and-after numbers. Not an unmeasured claim that it helped.

A fifty-year-old language, chosen on purpose#

Tagged dynamic values. Success/failure as the sole control primitive. Patterns as data.

Ideas a more comfortable modern language would let you take for granted.

What exists now is not a toy#

An interpreter that reproduces real SNOBOL4 behaviour, against a real reference implementation, on real sample transcripts — extended by a real design decision of your own choosing.