(...args: unknown[])
| 19 | * run uninformative. What a compile-cost test is actually defending against is |
| 20 | * an algorithmic blowup — a traversal that revisits nodes, or per-compile state |
| 21 | * that accumulates across calls — and that is visible as a node-visit count, |
| 22 | * which is identical on every machine. `countCompileVisits` below counts calls |
| 23 | * to `BaseCompiler.compile`, the single recursive entry point every target's |
| 24 | * emitter routes through. |
| 25 | * |
| 26 | * The execution-speed comparisons further down (compiled runner vs |
| 27 | * interpreter) are still timed, because there is no counter that can stand in |
| 28 | * for "the generated code runs faster than the interpreter". They compare two |
no test coverage detected