Downgrade: how to handle memory_reset_peak_usage() (no polyfill possible)

[roxblnfk]

Feature Request

Downgrading from PHP 8.2 to 8.1 currently leaves memory_reset_peak_usage() untouched, so the resulting code fatals on < 8.2 (Call to undefined function). I'd like to add a downgrade rule for it, but unlike array_is_list() / json_validate() / stream_isatty() this function cannot be polyfilled: it resets the engine's internal memory peak counter, and there is no userland API to do that before 8.2. Any downgrade therefore silently drops the semantics (a later memory_get_peak_usage() returns the non-reset peak). Because that's a behavior/policy question, I'd like maintainer guidance before implementing.

Option A — function_exists() guard, no-op fallback

-memory_reset_peak_usage();
+if (function_exists('memory_reset_peak_usage')) {
+    memory_reset_peak_usage();
+}

Runs on 8.2+, no-op on older PHP. Consistent with the existing function_exists idiom, but silently loses the peak reset. Simpler than the polyfill rules — the function is void and takes no arguments, so no closure/snippet is needed, just the guard.

Option B — don't add a rule

Leave the call as-is. The function is purely diagnostic/profiling, so arguably it's more honest to let the developer decide than to silently change memory-monitoring behavior.

Question

Would you accept Option A, prefer no rule, or have another shape in mind? Happy to open the PR once the direction is clear.

For reference, the semantically-clean sibling case ReflectionMethod::hasPrototype() (also 8.2, but polyfillable via getPrototype() + try/catch) is already up as rectorphp/rector-downgrade-php#398.

[TomasVotruba]

Thanks for reporting. In what context do you run this function?

[roxblnfk]

Benchmarks https://github.com/search?q=repo%3Aphp-testo%2Ftesto+memory_reset_peak_usage&type=code

In my case it's OK to replace it with something like

(function_exists('memory_reset_peak_usage') and \memory_reset_peak_usage());

[TomasVotruba]

I'd go with option B, as this has no exact replacement and generic approach might break other codebases.

Or make a custom Rector rule that fits your needs.