[devin-ai-integration]

What and why

Loading an empty model into a REPL allocates 2 155 times on the 0.12.0 candidate against 2 022 on 0.11.0's source built with the same toolchain (six rounds; exact). The +6.6 % is #1462's duplicate-library check: DuplicateLibraries walked every library document's root packages and appended each package symbol to a per-name slice, so the ~130 library packages, none of them duplicated, cost a slice allocation (and its growth) each on every load.

This PR keeps the check and its findings and changes only what it keeps for the common case: the first declaration of a name is held in a plain map until a second declaration of the same name is met, and only then does the name get a slice (seeded with both) and a place in the report order. The grouping of duplicates by the pair of trees, the used-declaration choice and the warning text are unchanged, as is the walk itself — a document count of root packages still has to be read, so the load does not return to 0.11.0's time.

A/B, six rounds (go test ./internal/frontend/repl -bench . -benchmem -count 1 × 6, benchstat; pre-#1462 / #1462 / candidate 1c6bb1972 / this branch; same host, "Intel(R) Xeon(R) Processor", 8 CPUs; bytes and allocations exact)

Row	pre-#1462	#1462	candidate	fix	0.11.0 source, go1.27.3
LoadModel/elements=0 allocs/op	2 021	2 151 (+6.43 %)	2 155 (+6.63 %)	2 038 (+0.84 %)	2 022
LoadModel/elements=0 B/op	227.9 Ki	246.7 Ki (+8.24 %)	246.8 Ki (+8.29 %)	238.1 Ki (+4.46 %)	228 Ki
LoadModel/elements=250, /1000, /4000 allocs/op and B/op				unchanged vs the candidate (their +3.4 % allocs / +5.3 % bytes over pre-#1462 are #1478, fixed in #1744)
LoadModel/* live-B/op				unchanged

Residual, stated in the fragment: 2 038 vs 2 022 allocations and 238 vs 228 Ki — the first map and the walk over the library roots are the check's cost. No sec/op claim: a full go test ./... gate ran on the host alongside rounds 1–3 and the timing columns carry ±100–300 % spread.

Memory acceptance

repl.test -test.bench '^BenchmarkLoadModel/elements=0$' -test.benchtime 200x, candidate and fix built the same way, three cold runs each, interleaved (/usr/bin/time, 8 CPUs):

run	candidate peak RSS	wall	fix peak RSS	wall
1	112.7 MiB	0.15 s	116.0 MiB	0.15 s
2	111.3 MiB	0.15 s	112.4 MiB	0.15 s
3	112.6 MiB	0.15 s	112.0 MiB	0.15 s

Hard cap: under ulimit -v 4000000 both pass (119 / 118 MiB peak RSS), and under ulimit -v 2000000 both pass too (117 / 113 MiB) — one earlier attempt of the fix under the 2 GB cap died in the Go runtime's arena reservation before loading anything, at 38 MB RSS, which the same binary did not repeat; that cap sits at the test binary's address-space reservation on either revision. Whole binary, sysml -validate over the 12 086-element generated model, three cold runs: 198 / 198 / 198 MiB candidate, 199 / 189 / 199 MiB fix, 0.24 s both; tests/stressmodel Load/512 1.27–1.36 GiB on both. Resident memory is not what this change recovers: it is 117 allocations and 9 KB per empty load, and the PR does not claim an RSS change.

How it was verified

• The duplicate-library tests in internal/check/workspace (TestDuplicateLibrariesIsQuietWhenNothingIsDeclaredTwice, …ReportsAUserLibraryDeclaredTwice, …ReportsTwoCopiesOfAStandardLibraryFile, …GroupsATreeDeclaredTwice, …SurvivesAnInterfaceRecordReload) pass unchanged — the finding set is the same, only the bookkeeping for single declarations changed.
• go test ./... below.

Gate

$ go build ./...        # ok
$ go vet ./...          # ok
$ gofmt -l .            # (empty)
$ go test -timeout 45m ./...
ok   124 packages, 0 FAIL, exit 0

(corpora present: training examples, pilot corpora, spec examples, pilot library XMI, PSSM suite, with their OPENSYSML_REQUIRE_* variables set.)

Checklist

• go build ./..., go vet ./..., gofmt -l . (empty), go test ./... pass locally (above); make lint not run (the gate above is the repository's definition of done)
• Tests added or updated for the change — the existing duplicate-library tests cover the finding set; no new behaviour
• Documentation extended where it already covers the surface — none covers the check's bookkeeping
• Changelog entry added as changes/unreleased/duplicate-libraries-walk.fixed.md
• baselines regenerated and make docs-counts run if a gate count moved — no count moved
• No internal work-item labels in the body, docs, or changelog

Link to Devin session: https://nasa-jpl-demo.devinenterprise.com/sessions/15408827f8104c7fbbdbd1b2c852ed2c
Open in Devin Desktop: https://nasa-jpl-demo.devinenterprise.com/desktop/session/15408827f8104c7fbbdbd1b2c852ed2c?variant=devin
Requested by: @HuiJun

[devin-ai-integration]

I'll fix CI failures and address comments from users with write access. I'll skip comments containing "(aside)".

• Disable automatic comment, CI, and merge conflict monitoring

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no bugs or issues to report.