aboutsummaryrefslogtreecommitdiffstats
path: root/experiments
diff options
context:
space:
mode:
authorblasty <blasty@local>2026-08-09 22:15:23 +0200
committerblasty <blasty@local>2026-08-09 22:15:23 +0200
commita5137fe0c6d6bb2c44a88f4ebc08e06c08161b2f (patch)
tree27ffa0945a79431022b7591491935b34d67e73a4 /experiments
parentGraph: never hand triskel a block its root cannot reach (diff)
downloadida-tui-a5137fe0c6d6bb2c44a88f4ebc08e06c08161b2f.tar.gz
ida-tui-a5137fe0c6d6bb2c44a88f4ebc08e06c08161b2f.tar.xz
ida-tui-a5137fe0c6d6bb2c44a88f4ebc08e06c08161b2f.zip
Drop the settrace workaround: ida-codemode 0.3.2 deleted the hook
_script() ran every snippet with sys.settrace(None) to detach Code Mode's per-line trace hook, which returned itself and so enabled line tracing in every frame it saw -- 52x on ida_bytes.get_flags, and the whole residual gap against the old private worker. 0.3.2 removes that hook entirely (zero settrace refs); the deadline is now a C-level thread interrupt, runtime._interrupt_thread. Re-measured on a 200-row listing page of main in targets/bash: the workaround buys 0.99x. So it goes, and the caveat goes with it -- a pure-Python loop inside a snippet is once again bounded by its deadline. _PACK_EPILOGUE measures 0.97x for the same reason (dumps_json got a C fast path) but is kept for encoder determinism, with its stale 114x claim corrected. experiments/bench_pack_trace.py is the harness for both numbers. Full gate green: 1031 passed, 0 failed.
Diffstat (limited to 'experiments')
-rw-r--r--experiments/bench_pack_trace.py97
1 files changed, 97 insertions, 0 deletions
diff --git a/experiments/bench_pack_trace.py b/experiments/bench_pack_trace.py
new file mode 100644
index 0000000..c6ca980
--- /dev/null
+++ b/experiments/bench_pack_trace.py
@@ -0,0 +1,97 @@
+"""Measure ``_PACK_EPILOGUE`` against the live ida-codemode runtime.
+
+Our snippets return one pre-serialised JSON STRING instead of a structure, to
+dodge to_jsonable()'s Python-level walk of the result (a 200-row listing page
+is ~10k small objects). ida-codemode 0.3.2 gave that path a C fast path --
+``serialization.dumps_json`` hands the structure straight to ``json.dumps`` and
+only falls back to the walker for values the encoder rejects -- so the packing
+now costs a double encode (escaping the whole payload as a string literal) to
+avoid a walk that may no longer happen.
+
+This script answers whether packing still pays. Its sibling question, the
+``sys.settrace`` strip, is settled: 0.3.2 deleted the trace hook, the workaround
+measured 0.99x, and it has been removed.
+
+Usage::
+
+ PYTHONPATH=. ~/ida-venv/bin/python experiments/bench_pack_trace.py [FILE]
+"""
+from __future__ import annotations
+
+import argparse
+import json
+import os
+import statistics
+import time
+
+from idatui import codemode_client as cc
+from idatui.codemode_client import CodeModeClient
+
+
+def _time(fn, reps: int) -> tuple[float, float]:
+ """Best-of and median wall time in ms; best-of resists co-tenant noise."""
+ samples = []
+ for _ in range(reps):
+ started = time.perf_counter()
+ fn()
+ samples.append((time.perf_counter() - started) * 1000.0)
+ return min(samples), statistics.median(samples)
+
+
+def main() -> int:
+ ap = argparse.ArgumentParser()
+ ap.add_argument("target", nargs="?", default="targets/bash")
+ ap.add_argument("--reps", type=int, default=25)
+ ap.add_argument("--rows", type=int, default=200)
+ args = ap.parse_args()
+
+ target = os.path.abspath(args.target)
+ client = CodeModeClient(target)
+ client.connect()
+
+ # A real listing page: the flow the workaround was tuned for.
+ # list_funcs answers {"result": [{"data": [...], "total": N}]}.
+ index = client.invoke("list_funcs", queries=[{"offset": 0, "count": 40}])
+ funcs = (index.get("result") or [{}])[0].get("data") or []
+ biggest = max(funcs, key=lambda f: f.get("size") or 0, default=None)
+ if not biggest:
+ print("VERDICT: FAIL - no functions")
+ return 1
+ addr = biggest["addr"]
+ if isinstance(addr, int):
+ addr = hex(addr)
+
+ page = lambda: client.invoke( # noqa: E731
+ "heads", addr=addr, count=args.rows, annotate=True)
+
+ # _script() reads _PACK_EPILOGUE at CALL time, so both variants share one
+ # process -- one lease, one warm database, one fair baseline. _unpack()
+ # passes an unpacked answer through untouched, so plain `result` works.
+ packed_epilogue = cc._PACK_EPILOGUE
+ results = {}
+ for packing in (True, False):
+ cc._PACK_EPILOGUE = packed_epilogue if packing else "\nresult\n"
+ payload = page()
+ rows = len(payload.get("heads", []))
+ for _ in range(3): # warm caches; the first sample is always an outlier
+ page()
+ results[packing] = (rows, *_time(page, args.reps))
+ cc._PACK_EPILOGUE = packed_epilogue
+
+ size = len(json.dumps(payload, separators=(",", ":"), default=str)) / 1024
+ print(f"target {os.path.basename(target)} backend={client.backend}")
+ print(f"listing page func {addr}, {results[True][0]} rows, "
+ f"{size:.1f} KiB of JSON, {args.reps} reps")
+ for packing, label in ((True, "packed string (current)"),
+ (False, "plain structure")):
+ rows, best, med = results[packing]
+ print(f" {label:<26} best {best:7.2f}ms median {med:7.2f}ms"
+ f" rows={rows}")
+ print(f" packing buys "
+ f"{results[False][1] / results[True][1]:.2f}x")
+ client.close()
+ return 0
+
+
+if __name__ == "__main__":
+ raise SystemExit(main())