[gyanu2507]

Summary

A << merge was re-decoding the anchored mapping at the merge site and writing over keys that were already present. That broke three cases from the merge key type:

1. An explicit key lost to a later merge (or raised a duplicate-key error into a struct).
2. Nested merges of the same alias expanded exponentially instead of reusing the decoded map.
3. An alias inside the merged mapping could resolve against a later redefinition of that anchor.

Merges now only fill missing keys. When the merge target is an alias (or a sequence of them) and the anchor is already decoded, that cached value is reused.

Fixes #964

Test plan

• Explicit key before/after << keeps the local value (map, struct, YAMLToJSON)
• Alias inside a merged mapping keeps the definition-site value
• 12-level nested triple-merge finishes quickly
• Existing merge/roundtrip tests
• go test ./...

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

❌ Patch coverage is 43.62140% with 137 lines in your changes missing coverage. Please review.
✅ Project coverage is 79.68%. Comparing base (edee2f9) to head (afde3bc).
❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #972      +/-   ##
==========================================
- Coverage   80.65%   79.68%   -0.97%     
==========================================
  Files          22       22              
  Lines        6845     7040     +195     
==========================================
+ Hits         5521     5610      +89     
- Misses        905      995      +90     
- Partials      419      435      +16     

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[kimjooyoon]

A native base/head review at e373c6131f19840f764850b843ad755afe5afcc4 confirms the explicit-key and definition-site alias corrections, but finds two data-loss regressions: cached typed anchors disappear in generic merges, and distinct MapSlice keys collapse when AllowDuplicateMapKey is used.

Self-contained 15-case fixture, original outputs and separate cached yaml/v3 reference. The identical fixture distinguishes 6 introduced failures, 2 corrected cases, 3 pre-existing default-parser failures and 4 passing controls. The three default-mode parser failures are not attributed to this PR; the ordered-key findings explicitly use the supported option to reach the changed code.

Execution and review were performed by an AI agent with cached Go1.27.3/macOS arm64, offline modules and bounded small inputs. No SDK/module was downloaded and no production source was modified. Deep alias/performance workloads, full-repository tests, race, vet and hosted CI are outside this review.

[kimjooyoon]

[P2] Handle the cached anchor type before reporting merge success

With Base map[string]int and Copy interface{} fields, decoding base: &base {k: 1}\ncopy: {<<: *base}\n returns nil on both revisions, but Copy changes from {k:1} on the base to an empty map here. Pointer-to-map and struct anchors behave the same way.

The cache contains the typed value, while mergeDecodedMapInto only handles map[string]interface{} and MapSlice. It silently does nothing for these other representations, then this branch returns success instead of using the mapping nodes. Please make success depend on actually handling the cached representation, with an appropriate fallback for unsupported types.

Three introduced cases plus generic/typed-map/ordered controls and independent reference. These are tiny mappings and do not require an alias-expansion benchmark.

[kimjooyoon]

[P2] Preserve typed MapSlice key identity when locating an entry

With yaml.UnmarshalWithOptions([]byte("1: integer\n\"1\": string\n"), &rows, yaml.AllowDuplicateMapKey()), base produces two MapSlice entries, uint64(1)->"integer" and string("1")->"string". This branch now leaves only uint64(1)->"string". Reversed order and bool(true) versus string("true") also lose an entry; no merge key is present.

fmt.Sprint(k) makes different typed keys compare equal and overwrites an unrelated value. Please preserve the actual key identity when updating ordered entries (YAML scalar equality includes tags). The explicit option is part of this reproduction: default mode already rejects these textual key collisions on both revisions, and that pre-existing behavior is recorded separately.

Option-enabled base/head originals and self-contained regression cases.

[gyanu2507]

Addressed both P2 notes from @kimjooyoon:

1. Cached typed anchors (e.g. map[string]int) are copied into a generic merge target via reflect. If the cached value is not a usable mapping, we fall back to walking the AST instead of returning success with an empty map.
2. decodeMapSlice keeps typed key identity under AllowDuplicateMapKey, so 1 and "1" stay as separate entries instead of collapsing through fmt.Sprint.

Added regression tests for both cases.

[kimjooyoon]

Thanks for the follow-up. I re-ran the unchanged15-case fixture at afde3bc8ba5853b79cd405e353d14743bc33dd31 (fixture SHA256 da0c0eb2f956b8edeb1253c610893dedd6973b9a3d218cdaa03814d475253adb). All six cases behind the two P2 findings now pass. The two original intent fixes and four controls still pass. The three pre-existing default-parser failures have exactly the same messages as base and the previous head, so the full fixture command still exits1; those are not new regressions.

Follow-up report and original test events. Both P2 findings are addressed within this fixture.

Execution and review were performed by an AI agent with cached Go1.27.3/macOS arm64 and offline modules. Full-repository tests, race, vet, hosted CI and performance remain outside this review.