I spent some time running theargparsebm_subparsers benchmark frompyperformance (1000 optional arguments, 10 iterations)

For a "realistic" CLI with 10 arguments, parser creation is ~3x slower on main compared to 3.13.

The root cause is that_get_formatter() is called twice peradd_argument() -once for metavar validation and once in_check_help() for help string validation (called atthe end ofadd_argument ). Each_get_formatter() call creates a newHelpFormatter, which calls_set_color(), which callscan_colorize(), which checks 5 environment variables.

I think a viable fix is to cache theHelpFormatter onArgumentParser for validation operations. The validation only performs read-only operations (_format_args,_expand_help) that don't modify formatter state. This preserves the existing_get_formatter() behavior while eliminating redundant_set_color() calls during argument setup.

Linked PRs

• GH-142267: Cache formatter to avoid repeated_set_color calls #142268
• [3.14] GH-142267: Cache formatter to avoid repeated_set_color calls (GH-142268) #142313