circle-ir-ai 2.8.5 → 2.8.7

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (2) hide show
  1. package/CHANGELOG.md +58 -0
  2. package/package.json +2 -2
package/CHANGELOG.md CHANGED
@@ -5,6 +5,64 @@ All notable changes to this project will be documented in this file.
5
5
  The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
6
6
  and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
7
7
 
8
+ ## [2.8.7] - 2026-06-10
9
+
10
+ ### Changed
11
+
12
+ - **Upgrade `circle-ir` from `^3.27.1` to `^3.33.0`** — picks up
13
+ 6 minor engine versions (3.28.0 through 3.33.0) published on
14
+ 2026-06-10. Typecheck ✓, build ✓, 556/556 tests pass against the
15
+ new engine with no source changes required (no breaking API
16
+ surface in the bump).
17
+
18
+ ## [2.8.6] - 2026-06-10
19
+
20
+ ### Fixed
21
+
22
+ - **#84 postmortem: parse-retry budget was entangled with HTTP retry
23
+ budget in v2.8.5.** Investigation after the v2.8.5 gemma3:12b
24
+ benchmark re-run showed an unchanged score (94/120, same 2 parse
25
+ errors). Inspection of the v2.8.5 verbose trail revealed:
26
+
27
+ ```
28
+ [LLM] JSON parse error (array) — parse-retry 1/1
29
+ [LLM] Error: This operation was aborted — retry 2/3 in 10s
30
+ [LLM] JSON parse error (array) — parse-retries exhausted
31
+ ```
32
+
33
+ The `parseRetries` counter in `benchmarks/runners/run-cwe-bench-java.ts`
34
+ was declared outside the `for (attempt...)` loop and was consumed by
35
+ the first parse error. When an AbortError subsequently triggered an
36
+ HTTP retry, the retry consumed an `attempt` slot but did not refresh
37
+ the parse-retry budget — so the next parse error had no budget left.
38
+
39
+ Fix: decoupled the two counters. Converted the loop to `while
40
+ (attempt <= MAX_RETRIES)` and now only HTTP-retry sites
41
+ (`attempt++`) consume the HTTP budget. Parse-retry `continue`
42
+ statements leave `attempt` untouched. Each error type now has its
43
+ own independent budget (max 4 HTTP attempts + 1 parse retry per
44
+ call). Maximum loop iterations bounded at 5.
45
+
46
+ ### Notes
47
+
48
+ - The v2.8.6 gemma3:12b benchmark (re-run 2026-06-10) scored the
49
+ same 94/120 with 2 parse errors. The retry-logic fix is correct
50
+ but didn't recover the lost detection because gemma3:12b
51
+ consistently fails on the same 2 specific prompts under the
52
+ benchmark's actual conditions (sustained corpus load + cross-file
53
+ context augmentation). Standalone repros with `num_ctx=32768`
54
+ succeed 5/5 and 6/6 respectively, indicating the failure is
55
+ contextual to the runner's full augmented prompt + position in
56
+ the sequence rather than a logic bug.
57
+
58
+ - The fix is the right semantic to ship regardless — removes a
59
+ latent bug from the retry path that would have masked any future
60
+ retry-recoverable failure.
61
+
62
+ - For users seeking zero parse errors, recommend `qwen3-coder:30b`
63
+ (0/113 failures, 91/120 score, ~2x faster per call) over
64
+ `gemma3:12b` (2/109 failures, 94/120 score).
65
+
8
66
  ## [2.8.5] - 2026-06-09
9
67
 
10
68
  ### Fixed
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "circle-ir-ai",
3
- "version": "2.8.5",
3
+ "version": "2.8.7",
4
4
  "description": "LLM-enhanced SAST analysis built on circle-ir",
5
5
  "main": "dist/index.js",
6
6
  "module": "dist/index.js",
@@ -94,7 +94,7 @@
94
94
  "dependencies": {
95
95
  "@ax-llm/ax": "^20.0.0",
96
96
  "@mastra/core": "^1.18.0",
97
- "circle-ir": "^3.27.1",
97
+ "circle-ir": "^3.33.0",
98
98
  "minimatch": "^10.2.5",
99
99
  "p-queue": "^9.1.0"
100
100
  },