Skip to content

chore(codeboarding): full analysis from core 0.14.0 (e38c07e3) - #2

Open
Svilen-Stefanov wants to merge 1 commit into
devfrom
codeboarding-analysis-20260907
Open

chore(codeboarding): full analysis from core 0.14.0 (e38c07e3)#2
Svilen-Stefanov wants to merge 1 commit into
devfrom
codeboarding-analysis-20260907

Conversation

@Svilen-Stefanov

Copy link
Copy Markdown

Full CodeBoarding analysis of abp at 626c8c7592, generated with core main @ e38c07e3 (0.14.0).

Run

Wall clock ~55 min (31 min static analysis, 24 min LLM)
LLM 128 calls, Kimi-K2.6 via Azure
Peak RSS 2.9 GB
LSP ref-timeouts 0

Result

504 components across 3 levels (9 → 121 → 374), 1,955 component relations, 87/87 subcomponent scopes generated with 0 errors.

Scope: 6,616 of 16,845 files analyzed — 6,284 C#, 210 JS, 113 TS, 8 Go. The rest: 6,328 non-source, 2,700 excluded by .codeboardingignore, 1,151 by abp's own .gitignore (mostly generated Angular API proxies under npm/ng-packs).

Bundle exclusions

.codeboardingignore gained three vendored third-party JS bundles under npm/packs. These are git-tracked, so .gitignore gave no cover and they went straight into the analysis:

  • npm/packs/codemirror/src/codemirror.js is 1 MB / 26k lines with individual lines up to 23k chars
  • npm/packs/bootstrap-daterangepicker/src/daterangepicker.js (66 KB)
  • npm/packs/qrcode/src/qrcode.js (33 KB)

abp's own glue in the same tree (abp.js, abp.jquery.js, bootstrap.enable.*.js) is deliberately still analyzed.

Known issue in this output

The root scope overflowed the model context — 471,947 tokens against a 262,144 limit, 1.8x over. It failed safely (Semantic analysis failed for scope root; retaining deterministic output), so the run completed, but the 9 top-level components carry template descriptions and directory-derived grab-bag names (CmsKit owns 2,774 files including Identity/Docs/Blogging; Cli owns Volo.Abp.AI.Abstractions). All 495 descendants are named normally.

Cause is the uncapped groups[].files block in the scope-agent prompt: 6,428 file entries at ~239 chars each ≈ 384k tokens, ~81% of the prompt. Core abpframework#565 capped known_connections but left this term unbounded — same bug shape, different field. No release ruler covers a repo of abp's file count, so it ships unnoticed.

Treat the top level of this diagram as not-yet-validated. Levels 2 and 3 are sound.

Note

logs/ is included at request and contains local filesystem paths from the run host. analysis.json is 59 MB, above GitHub's 50 MB recommendation — LFS is worth considering if this fork accumulates runs.

🤖 Generated with Claude Code

Full analysis run of abp at 626c8c7 using CodeBoarding core main
e38c07e3 (0.14.0). Run took ~55 min (31 min static analysis, 24 min LLM),
128 LLM calls on Kimi-K2.6, peak RSS 2.9 GB.

Result: 504 components across 3 levels (9 -> 121 -> 374), 1,955 component
relations, 87/87 subcomponent scopes generated with 0 errors.

Scope: 6,616 of 16,845 files analyzed (6,284 C#, 210 JS, 113 TS, 8 Go).
.codeboardingignore gained exclusions for three vendored third-party JS
bundles under npm/packs that are git-tracked and so not covered by
.gitignore. The worst was codemirror.js (1 MB, 26k lines, individual
lines up to 23k chars); daterangepicker.js and qrcode.js went with it.
abp's own glue (abp.js, abp.jquery.js) is deliberately still analyzed.

Known issue in this output: the root scope overflowed the model context
(471,947 tokens vs a 262,144 limit), so it fell back to deterministic
output. The 9 top-level components therefore carry template descriptions
and directory-derived names; all 495 descendants are named normally.
Cause is the uncapped groups[].files block in the scope-agent prompt
(6,428 file entries, ~384k tokens, ~81% of the prompt).

Logs are included and contain local filesystem paths from the run host.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant