methodcaller vs getattr Call: Localhost Lab

Intro — what this post promises

Call a method on many objects: operator.methodcaller, getattr(obj, name)(), direct obj.method(), and pre-bound methods. This lab reports calls/s on Linux localhost.

It is not attrgetter (lab 65) or itemgetter (lab 59) — those fetch attributes/items; this invokes methods.

Related links:

• 01shoppercove.comattrgetter vs getattr localhost lab
• 02shoppercove.comitemgetter vs lambda sort localhost lab
• 03shoppercove.comfunctools partial vs lambda localhost lab
• 04shoppercove.comgroupby vs manual localhost lab
• 05shoppercove.comhmac compare digest localhost lab
• 06shoppercove.comargparse vs sys argv localhost lab
• 07shoppercove.comfutures as completed vs wait localhost lab
• 08shoppercove.comfnmatch vs re localhost lab

Lab honesty (1 Oct 2026 IST): Python 3.13.5. Affiliates: 0. No Docker. Widget.value() / Widget.add(1).

Verdict up front (10 000 objects): direct value() ~30.38 Mcalls/s; getattr ~17.04; methodcaller ~13.53; prebound calls ~29.68. Use methodcaller for clean map/sorted key hooks; direct calls when the name is fixed in code.

Arms

Lab topology

n in {1000, 10000, 50000} Widget objects · 7 rounds · p50
metric: calls/s = n / p50_s
methods: value() no-arg · add(1) with fixed arg

Script: lab-evidence/99-methodcaller-vs-getattr/results/run_lab.py.

Lead table — n=10 000 (p50 Mcalls/s)

Direct and prebound sit near the top. methodcaller pays for a portable callable. Binding a fresh method object every outer loop (bound_methods) is the costly anti-pattern.

With argument (add(1), n=10 000)

With a bound argument, methodcaller closes some of the gap vs getattr — still behind a direct call on this box.

Scale check (value())

Absolute rates move with cache effects; the ordering (direct / prebound ahead of dynamic lookup) is the useful takeaway.

map/sorted fit

methodcaller shines when an API wants a one-argument callable: sorted(objs, key=methodcaller("score")) or map(methodcaller("strip"), lines). That clarity often matters more than a few Mcalls/s on a microbench. Pair with attrgetter only when you need the attribute itself, not a call.

Reading it

• Direct calls win when the method name is static.
• methodcaller is for higher-order tools (map, sorted(..., key=methodcaller("lower"))) — readable, slight tax.
• getattr+call suits truly dynamic names; expect overhead vs direct.
• Pre-binding helps if you call the same bound methods repeatedly; binding every iteration loses.

Pitfalls

• Using methodcaller where a direct call is fine (noise in profiles).
• Binding methods in a tight loop without reuse.
• Confusing attrgetter (get attribute) with methodcaller (call method).
• Passing wrong arity into methodcaller-bound args.
• Measuring only n=1k — check a larger size before drawing product conclusions.

Reproduce

python3 lab-evidence/99-methodcaller-vs-getattr/results/run_lab.py

Evidence: summary.json, summary.txt.

Limits

One Linux box. Microbench only — real workloads with I/O will dwarf call overhead. Tiny methods returning ints. Not free-threaded surprises. Not slots vs dict object layouts beyond Widget slots.

Takeaway

At 10 k objects, direct ~30.38 Mcalls/s beat getattr ~17.04 and methodcaller ~13.53. Prefer direct in hot loops; methodcaller when you need a portable callable for APIs that expect a function.