Skip to content

gh-148653: refactored marshal for cycle safety and performance #148700

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
mjbommar wants to merge 6 commits into from
Closed

gh-148653: refactored marshal for cycle safety and performance #148700

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
7c214ea
gh-148653: Fix SIGSEGV in marshal.loads for self-referencing tuples
mjbommar Apr 16, 2026
3835149
marshal: preserve safe recursive hashable refs
mjbommar Apr 17, 2026
eb1c4b7
marshal: add benchmark summary note
mjbommar Apr 17, 2026
4aaf064
marshal: recover loads regression with raw refs table + tagged state
mjbommar Apr 17, 2026
e931b7e
marshal: add final validation section + raw perf data folder
mjbommar Apr 17, 2026
4c18f3a
marshal: add third-party + fuzz validation section to perf diary
mjbommar Apr 17, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Prev Previous commit
Next Next commit
marshal: add final validation section + raw perf data folder 
Appends to Misc/marshal-perf-diary.md the results of the full test
suite rerun (48,932 tests pass, including the new RecursiveGraphTest
combinatoric cases) and a `pyperformance` comparison against main on
the same 10-benchmark marshal-adjacent slice the design doc used.

Significant results on the pyperformance slice:

    python_startup          1.18x faster (t=59.80)
    python_startup_no_site  1.03x faster (t=12.90)

All other slice benchmarks within noise; no regressions.

Adds Misc/marshal-perf-data/ with the raw JSON backing every table in
the diary: all per-experiment microbench runs (exp0..exp9, expC, final)
plus the two pyperf-slice JSONs and a README describing the layout and
reproduction commands.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
  • Loading branch information
@mjbommar @claude
mjbommar and claude committed Apr 17, 2026
commit e931b7ec45c21048389b4235d44a6220ceb9bdfc
53 changes: 53 additions & 0 deletions Misc/marshal-perf-data/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,53 @@
# Marshal Perf Raw Data

Raw JSON artifacts backing `Misc/marshal-perf-diary.md`. Everything here
is regeneratable — kept checked in so reviewers can inspect numbers
without rerunning the full methodology.

## Marshal microbench (`marshal_bench_cpu_stable.py`)

Three pinned runs per commit, 11 repeats per operation, 200k `loads` ×
100k `dumps`. Pinning: `taskset -c 0`.

| File | Commit / idea |
| --- | --- |
| `exp0-run{1..3}.json` | `perf-baseline` self-check at `eb1c4b7` |
| `exp1-run{1..3}.json` | Experiment 1 (raw `PyObject **` refs array) |
| `exp2-run{1..3}.json` | Experiment 2 (tagged pointer state, stacked on 1) |
| `exp3-run{1..3}.json` | Experiment 3 (collapse reserve+publish) |
| `exp4-run{1..3}.json` | Experiment 4 (`has_incomplete` counter) |
| `exp5-run{1..3}.json` | Experiment 5 (zero-init + drop mark_ready) |
| `exp6-run{1..3}.json` | Experiment 6 (drop `allow_incomplete_hashable` param) |
| `exp7-run{1..3}.json` | Experiment 7 (force-inline helpers) |
| `exp8-run{1..3}.json` | Experiment 8 (`frozendict`-then-wrap when `flag==0`) |
| `exp9-run{1..3}.json` | Experiment 9 (preallocate refs capacity) |
| `expC-run{1..3}.json` | Combined (Exp 1+2+6+7) on `exp/combined-winners` |
| `final-head-run{1..3}.json` | Post-squash HEAD `4aaf064344d` (same as Combined) |

Schema per bench (small_tuple / nested_dict / code_obj):

loads_number, dumps_number # inner loop counts
loads_runs: [11 times] # raw timings
dumps_runs: [11 times]
loads_min, loads_median # statistics
dumps_min, dumps_median

## pyperformance slice

Ten marshal-adjacent benchmarks (pickle × 4, unpickle × 3,
python_startup × 2, unpack_sequence). Run with `uvx pyperformance run
-b <list> --affinity 0` via `taskset -c 0`.

| File | Python |
| --- | --- |
| `pyperf-slice-baseline.json` | `main` tip `2faceeec` built at `/tmp/cpython-baseline/python` |
| `pyperf-slice-current.json` | HEAD `4aaf064344d` built in-tree |

These are standard `pyperf` format; compare with:

uvx pyperformance compare pyperf-slice-baseline.json pyperf-slice-current.json

## Regeneration

See `Misc/marshal-perf-diary.md` → "Ground rules" for the exact harness
(`/tmp/marshal_bench_cpu_stable.py` inline script) and methodology.
104 changes: 104 additions & 0 deletions Misc/marshal-perf-data/exp0-run1.json
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,104 @@
{
"code_obj": {
"dumps_median": 0.040405122010270134,
"dumps_min": 0.039836373005528,
"dumps_number": 100000,
"dumps_runs": [
0.040096846991218626,
0.04028077999828383,
0.04065272197476588,
0.04065836101653986,
0.04028131702216342,
0.040413388021988794,
0.039836373005528,
0.04000361601356417,
0.04100019999896176,
0.040806725999573246,
0.040405122010270134
],
"loads_median": 0.09791472702636383,
"loads_min": 0.09674789398559369,
"loads_number": 200000,
"loads_runs": [
0.09674789398559369,
0.09708369200234301,
0.09783264898578636,
0.09744057801435702,
0.09815629699733108,
0.09906345797935501,
0.09835395799018443,
0.09974412800511345,
0.09990381199168041,
0.09737725998274982,
0.09791472702636383
]
},
"nested_dict": {
"dumps_median": 0.07455040101194754,
"dumps_min": 0.07420829197508283,
"dumps_number": 100000,
"dumps_runs": [
0.07494553699507378,
0.07451574600418098,
0.07446633197832853,
0.07566936200601049,
0.07430389401270077,
0.07455040101194754,
0.07424808200448751,
0.07420829197508283,
0.07491164302336983,
0.07538196400855668,
0.0746764489740599
],
"loads_median": 0.08682646200759336,
"loads_min": 0.08603669900912791,
"loads_number": 200000,
"loads_runs": [
0.08728363699628972,
0.08605418298975565,
0.08732464100467041,
0.08639801701065153,
0.08744612199370749,
0.08878127401112579,
0.08682646200759336,
0.08644566498696804,
0.08648836001520976,
0.08734603400807828,
0.08603669900912791
]
},
"small_tuple": {
"dumps_median": 0.0156986290239729,
"dumps_min": 0.01551589198061265,
"dumps_number": 100000,
"dumps_runs": [
0.015826012007892132,
0.015661484998418018,
0.01551589198061265,
0.01563638899824582,
0.0156986290239729,
0.015819393011042848,
0.01584347800235264,
0.01551837200531736,
0.015644578001229092,
0.01609343197196722,
0.01589953200891614
],
"loads_median": 0.032818893989315256,
"loads_min": 0.032478501991136,
"loads_number": 200000,
"loads_runs": [
0.039915669010952115,
0.034840578999137506,
0.032818893989315256,
0.032755594002082944,
0.032478501991136,
0.03285682899877429,
0.032941445999313146,
0.03272099400055595,
0.032805391994770616,
0.0325985630042851,
0.032910926995100453
]
}
}
104 changes: 104 additions & 0 deletions Misc/marshal-perf-data/exp0-run2.json
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,104 @@
{
"code_obj": {
"dumps_median": 0.039624935016036034,
"dumps_min": 0.03912865900201723,
"dumps_number": 100000,
"dumps_runs": [
0.04005826299544424,
0.039624935016036034,
0.03912865900201723,
0.03937379698618315,
0.03985618698061444,
0.0396080739737954,
0.03980735401273705,
0.039562535006552935,
0.03952119100722484,
0.04000154402456246,
0.03970523300813511
],
"loads_median": 0.09828513601678424,
"loads_min": 0.0965760959952604,
"loads_number": 200000,
"loads_runs": [
0.09915574299520813,
0.0985370319976937,
0.0965760959952604,
0.09828084200853482,
0.097619662003126,
0.10028967598918825,
0.0983922420127783,
0.0987230259925127,
0.09828513601678424,
0.09821752097923309,
0.09726348100230098
]
},
"nested_dict": {
"dumps_median": 0.07583842999883927,
"dumps_min": 0.07371149398386478,
"dumps_number": 100000,
"dumps_runs": [
0.07573145598871633,
0.07466877298429608,
0.07562672600033693,
0.07371149398386478,
0.07529966699075885,
0.07686721297795884,
0.07583842999883927,
0.0770502710074652,
0.076849859993672,
0.07778827799484134,
0.07668794898199849
],
"loads_median": 0.08660990899079479,
"loads_min": 0.0859223100123927,
"loads_number": 200000,
"loads_runs": [
0.08658557198941708,
0.08649306499864906,
0.08618080499581993,
0.0874386359937489,
0.08762964798370376,
0.0859223100123927,
0.0869109459745232,
0.0878543850267306,
0.08672871999442577,
0.08612436000839807,
0.08660990899079479
]
},
"small_tuple": {
"dumps_median": 0.015691205015173182,
"dumps_min": 0.015531735989497975,
"dumps_number": 100000,
"dumps_runs": [
0.015568955015623942,
0.015662263002013788,
0.01571622901246883,
0.015571760974125937,
0.015531735989497975,
0.01582504398538731,
0.01576065199333243,
0.015691205015173182,
0.015571972995530814,
0.015708502003690228,
0.015745237993542105
],
"loads_median": 0.03315894500701688,
"loads_min": 0.03273815900320187,
"loads_number": 200000,
"loads_runs": [
0.036090038978727534,
0.03278564300853759,
0.03315894500701688,
0.03321557000163011,
0.032992702996125445,
0.0333719459886197,
0.03368866798700765,
0.03273815900320187,
0.03284121898468584,
0.03342736198101193,
0.032767266005976126
]
}
}
104 changes: 104 additions & 0 deletions Misc/marshal-perf-data/exp0-run3.json
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,104 @@
{
"code_obj": {
"dumps_median": 0.04041685801348649,
"dumps_min": 0.03987288600183092,
"dumps_number": 100000,
"dumps_runs": [
0.040255115018226206,
0.040099431993439794,
0.04095420497469604,
0.03987288600183092,
0.04041685801348649,
0.040494924993254244,
0.04049262698390521,
0.04079927600105293,
0.04118377499980852,
0.04038904799381271,
0.04038159898482263
],
"loads_median": 0.09712665400002152,
"loads_min": 0.09544071301934309,
"loads_number": 200000,
"loads_runs": [
0.09710476599866524,
0.09544071301934309,
0.09796852199360728,
0.09641282400116324,
0.09771751399966888,
0.09875676699448377,
0.09825973899569362,
0.09712665400002152,
0.09814226999878883,
0.09603261601296254,
0.0969423119968269
]
},
"nested_dict": {
"dumps_median": 0.07670464701368473,
"dumps_min": 0.07443057300406508,
"dumps_number": 100000,
"dumps_runs": [
0.07670464701368473,
0.0755682380113285,
0.07518417699611746,
0.07505606301128864,
0.07443057300406508,
0.07728818198665977,
0.07829541899263859,
0.07810401200549677,
0.07622043599258177,
0.07752709698979743,
0.07774014701135457
],
"loads_median": 0.08734592801192775,
"loads_min": 0.08589171001221985,
"loads_number": 200000,
"loads_runs": [
0.08709321398055181,
0.08734592801192775,
0.08714917601901107,
0.08695745901786722,
0.08589171001221985,
0.08879310198244639,
0.08659902500221506,
0.08748295900295489,
0.08759746799478307,
0.08768528798827901,
0.08783682601642795
]
},
"small_tuple": {
"dumps_median": 0.01571595700806938,
"dumps_min": 0.015293899981770664,
"dumps_number": 100000,
"dumps_runs": [
0.01571595700806938,
0.015348431013990194,
0.015293899981770664,
0.015364045015303418,
0.01530077401548624,
0.015608929010340944,
0.015807940013473853,
0.01611778000369668,
0.01599793799687177,
0.01602914900286123,
0.016072057012934238
],
"loads_median": 0.03303753398358822,
"loads_min": 0.032468440011143684,
"loads_number": 200000,
"loads_runs": [
0.036123027006397024,
0.03290336497593671,
0.032468440011143684,
0.032748425990575925,
0.03303753398358822,
0.033069997996790335,
0.03337217401713133,
0.03351389098679647,
0.033144343993626535,
0.03277306197560392,
0.03264542398392223
]
}
}
Loading
Loading