Repository navigation
Optimizer calls getenv on every optimization run #116772
Description
Bug report
Bug description:
The optimizer currently calls getenv during every optimization attempt. This has a small, but statistically significant 0.5% impact on performance. Do we need to do this each time, or could we read it to a flag during initialization?
diff --git a/Python/optimizer.c b/Python/optimizer.c
index 88c45f2e73c..b55556d30ab 100644
--- a/Python/optimizer.c
+++ b/Python/optimizer.c
@@ -1010,8 +1010,7 @@ uop_optimize(
return err;
}
OPT_STAT_INC(traces_created);
- char *env_var = Py_GETENV("PYTHON_UOPS_OPTIMIZE");
- if (env_var == NULL || *env_var == '\0' || *env_var > '0') {
+ if (true) {
err = _Py_uop_analyze_and_optimize(frame, buffer,
UOP_MAX_TRACE_LENGTH,
curr_stackentries, &dependencies);CPython versions tested on:
CPython main branch
Operating systems tested on:
Linux
Activity
ronaldoussoren commented on Mar 16, 2024
IMHO options should be processed using the PyConfig interface, that way all configuration stays in a central place and it is easier for embedders to change this setting. Also: With the current code there's no way to enable this option when using python -E.
savannahostrowski commented on Jan 23, 2026
This issue seems to have been resolved by #144082, which moved PYTHON_UOPS_OPTIMIZE to the _PyOptimizationConfig. The environment variable is now read once during interpreter initialization in pystate.c and cached in interp->opt_config.uops_optimize_enabled so we aren't calling getenv() for this flag during optimization runs.