IxVM: fix false rejections and pathological kernel reductions - #560
Merged
Conversation
Full-environment IxVM sweeps exposed two classes of kernel bugs: valid declarations rejected by canonical/positivity checks, and open def-equality or inference workloads that failed to converge in practical time. Correct the recursor canonical checker to reconstruct each mutual member from its own index telescope and only permit Prop large elimination for solo blocks. Skip strict-positivity checking for unsafe inductives, matching Lean and the Rust kernel. Canonicalize open inference memo keys by rebasing subjects to their minimum loose index, then lift inferred types back into the caller context. This prevents irrelevant inner binder frames from poisoning cache reuse. Keep lazy def-equality reduction bounded and no-delta between steps. Reuse the current lazy budget instead of recursively restarting application congruence, recognize compact Nat offsets, preserve K/projection reductions under no-delta WHNF, and retain a last-resort full-WHNF path for representations that cannot otherwise converge. Add focused fixtures for the mutual-recursors, unsafe-inductive, inference-rebase, large-offset Char, and indexed Time-cast failure shapes. Regenerate the native IxVM and repin all kernel and shard FFT costs after rebasing onto main. Validation: lake exe ix codegen; lake build IxTests; lake test -- --ignored ixvm (187.06s, all execution, FFT, exploit, and bytecode/native parity checks passed).
arthurpaulino
enabled auto-merge (squash)
August 14, 2026 17:39
Member
Author
|
!benchmark aiur fresh |
|
| constant | prove-time (main) | prove-time (PR) | Δ% | throughput (const/s) (main) | throughput (const/s) (PR) | Δ% | peak-ram (main) | peak-ram (PR) | Δ% | execute-time (main) | execute-time (PR) | Δ% | verify-time (main) | verify-time (PR) | Δ% | proof-size (main) | proof-size (PR) | Δ% | fft-cost (main) | fft-cost (PR) | Δ% |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
ByteArray.utf8DecodeChar?_utf8EncodeChar_append |
36.952 s | 34.657 s | -6.2% (1.07× faster) 🟢 | 72.820 | 77.650 | +6.6% (1.07× faster) 🟢 | 72.14 GiB | 69.12 GiB | -4.2% 🟢 | 9.538 s | 8.652 s | -9.3% (1.10× faster) 🟢 | 192.3 ms | 140.6 ms | -26.9% (1.37× faster) 🟢 | 21.63 MiB | 21.77 MiB | +0.6% | 136.87B | 126.12B | -7.9% (1.09× fewer) 🟢 |
Char.ofOrdinal_le_of_le |
28.449 s | 28.171 s | -1.0% | 93.010 | 93.930 | +1.0% | 58.02 GiB | 56.33 GiB | -2.9% | 6.494 s | 6.707 s | +3.3% |
131.3 ms | 174.5 ms | +32.9% (1.33× slower) |
21.58 MiB | 21.78 MiB | +0.9% | 98.00B | 99.10B | +1.1% |
Array.extract_append |
37.674 s | 24.593 s | -34.7% (1.53× faster) 🟢 | 40.930 | 62.700 | +53.2% (1.53× faster) 🟢 | 73.15 GiB | 47.48 GiB | -35.1% (1.54× smaller) 🟢 | 9.460 s | 5.791 s | -38.8% (1.63× faster) 🟢 | 138.3 ms | 138.4 ms | +0.1% | 21.54 MiB | 21.60 MiB | +0.3% | 135.19B | 86.05B | -36.3% (1.57× fewer) 🟢 |
Vector.extract_append._proof_2 |
21.246 s | 16.367 s | -23.0% (1.30× faster) 🟢 | 61.280 | 79.550 | +29.8% (1.30× faster) 🟢 | 39.37 GiB | 31.36 GiB | -20.3% (1.26× smaller) 🟢 | 5.135 s | 3.499 s | -31.9% (1.47× faster) 🟢 | 127.9 ms | 131.7 ms | +3.0% | 21.24 MiB | 21.44 MiB | +0.9% | 76.48B | 53.57B | -30.0% (1.43× fewer) 🟢 |
_private.Init.Data.Range.Polymorphic.SInt.0.Int64.instRxcHasSize_eq |
16.967 s | 16.251 s | -4.2% 🟢 | 106.790 | 111.500 | +4.4% 🟢 | 34.28 GiB | 32.36 GiB | -5.6% (1.06× smaller) 🟢 | 3.544 s | 3.401 s | -4.1% 🟢 | 131.8 ms | 130.4 ms | -1.1% | 21.42 MiB | 21.62 MiB | +0.9% | 54.00B | 51.47B | -4.7% 🟢 |
String.split |
16.020 s | 15.743 s | -1.7% | 110.170 | 112.120 | +1.8% | 32.24 GiB | 31.71 GiB | -1.6% | 3.287 s | 3.179 s | -3.3% 🟢 | 134.8 ms | 134.7 ms | -0.1% | 21.62 MiB | 21.78 MiB | +0.7% | 49.51B | 48.07B | -2.9% |
List.mergeSort |
11.979 s | 12.377 s | +3.3% |
120.880 | 116.990 | -3.2% |
23.48 GiB | 24.60 GiB | +4.8% |
2.360 s | 2.420 s | +2.5% | 133.0 ms | 139.2 ms | +4.7% |
21.48 MiB | 21.68 MiB | +0.9% | 35.52B | 35.66B | +0.4% |
Vector.append |
3.873 s | 4.277 s | +10.4% (1.10× slower) |
125.240 | 113.390 | -9.5% (1.10× slower) |
6.37 GiB | 6.95 GiB | +9.1% (1.09× larger) |
656.0 ms | 704.1 ms | +7.3% (1.07× slower) |
128.4 ms | 125.1 ms | -2.5% | 20.08 MiB | 20.25 MiB | +0.9% | 7.90B | 8.67B | +9.8% (1.10× more) |
Nat.gcd_comm |
3.142 s | 3.267 s | +4.0% |
124.140 | 119.360 | -3.9% |
4.96 GiB | 5.10 GiB | +2.7% | 517.8 ms | 533.5 ms | +3.0% |
122.7 ms | 124.4 ms | +1.5% | 19.84 MiB | 20.08 MiB | +1.2% | 5.15B | 5.48B | +6.4% (1.06× more) |
String.append |
2.367 s | 2.417 s | +2.1% | 128.430 | 125.760 | -2.1% | 5.59 GiB | 4.68 GiB | -16.3% (1.19× smaller) 🟢 | 391.4 ms | 423.3 ms | +8.2% (1.08× slower) |
118.2 ms | 115.7 ms | -2.1% | 19.12 MiB | 19.34 MiB | +1.2% | 2.85B | 3.04B | +6.7% (1.07× more) |
Int.gcd |
1.962 s | 2.056 s | +4.8% |
105.990 | 101.170 | -4.5% |
4.51 GiB | 4.61 GiB | +2.1% | 348.9 ms | 355.0 ms | +1.7% | 113.1 ms | 118.1 ms | +4.4% |
18.67 MiB | 18.91 MiB | +1.3% | 1.83B | 1.96B | +6.6% (1.07× more) |
Nat.sub_le_of_le_add |
1.753 s | 1.882 s | +7.4% (1.07× slower) |
96.970 | 90.320 | -6.9% (1.07× slower) |
5.20 GiB | 4.67 GiB | -10.2% (1.11× smaller) 🟢 | 343.7 ms | 352.2 ms | +2.5% | 114.1 ms | 126.6 ms | +11.0% (1.11× slower) |
19.03 MiB | 19.29 MiB | +1.4% | 1.58B | 1.70B | +7.7% (1.08× more) |
Nat.add_comm |
1.022 s | 1.102 s | +7.8% (1.08× slower) |
41.090 | 38.110 | -7.3% (1.08× slower) |
3.86 GiB | 3.97 GiB | +2.8% | 248.7 ms | 254.9 ms | +2.5% | 105.9 ms | 111.2 ms | +5.0% |
17.20 MiB | 17.67 MiB | +2.8% | 268.28M | 274.30M | +2.2% |
samuelburnham
approved these changes
Aug 14, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Full-environment IxVM sweeps exposed two classes of kernel bugs: valid declarations rejected by canonical/positivity checks, and open def-equality or inference workloads that failed to converge in practical time.
Correct the recursor canonical checker to reconstruct each mutual member from its own index telescope and only permit Prop large elimination for solo blocks. Skip strict-positivity checking for unsafe inductives, matching Lean and the Rust kernel.
Canonicalize open inference memo keys by rebasing subjects to their minimum loose index, then lift inferred types back into the caller context. This prevents irrelevant inner binder frames from poisoning cache reuse.
Keep lazy def-equality reduction bounded and no-delta between steps. Reuse the current lazy budget instead of recursively restarting application congruence, recognize compact Nat offsets, preserve K/projection reductions under no-delta WHNF, and retain a last-resort full-WHNF path for representations that cannot otherwise converge.
Add focused fixtures for the mutual-recursors, unsafe-inductive, inference-rebase, large-offset Char, and indexed Time-cast failure shapes. Regenerate the native IxVM and repin all kernel and shard FFT costs after rebasing onto main.
Validation: lake exe ix codegen; lake build IxTests; lake test -- --ignored ixvm (187.06s, all execution, FFT, exploit, and bytecode/native parity checks passed).