[Deepya1502]

Summary

Adds RFC 8525 YANG Library parsing and opt-in lenient handling for partial schemas. The parser builds a schema context from the library XML and module directory; schema and data processing use the validation policy attached to that context.

Validation behavior

• Strict validation is the default.
• Lenient behavior is explicitly selected with LenientValidationOptions.LENIENT, for example via YangLibraryParser.parse(xmlFile, modulesDir, options) or the corresponding YangYinParser.parseFromYangLibrary overload. The immutable option is stored on the schema context; there is no static or thread-local state.
• YangSchemaContext.getValidationOptions() defaults to STRICT, and its setter is a default no-op, preserving compatibility for external implementations.
• Invalid identityrefs against a complete schema and missing list keys remain errors. Leniency is limited to partial-schema cases such as unavailable referenced modules and inactive feature-gated nodes.
• Case/choice schema building preserves feature checks in strict mode; CaseImpl merges child insertion validation results.
• Revision-aware module lookup rejects a file whose declared revision does not match the YANG Library request.

Review follow-up

• The invalid-name augment-order fixtures and custom second-pass workaround were removed; this PR relies on mainline behavior with valid YANG identifiers.
• YangStructure JSON support was split into PR #54.
• Base/Identity resolution, including the BASE SCHEMA_BUILD registration, was split into PR #55. This PR keeps BASE registered for GRAMMAR only.
• The mutable ValidatorResultBuilder.getRecords() addition and YangYinParser debug logging were dropped. The list-key severity comment now matches the behavior.

Verification

Rebased onto yang-central/main at 573a6f18 (including merged PRs #50, #52, and #53). Full mvn test passes across the reactor, including lenient-validation and revision-match regressions.

[lllyfeng]

Thanks for the contribution. The overall YANG Library integration looks useful, but I found several issues that should be addressed before merging:
IdentityRefStringValueCodecImpl now accepts values even when restriction.evaluate() returns false or throws an exception. This changes validation globally and allows invalid identityref values with a complete schema. Lenient handling should be opt-in and limited to genuinely missing modules.

ListDataJsonCodec downgrades a missing list key to a warning and returns null, silently dropping the entire list entry. This can cause data loss while parsing appears successful. Missing keys should remain errors by default.

YangAbstractDataContainer.validateChildren() now accepts inactive nodes, including nodes disabled by if-feature or deviate not-supported. This weakens validation for every caller, not only partial YANG Library schemas. Please preserve strict validation by default and expose lenient behavior through an explicit option.

The second-pass cross-module augment resolution occurs after the original module.build() errors have already been merged into the validation result. Even if the target is resolved during the second pass, the earlier MISSING_TARGET error remains, so validation may still fail.

When a specific revision is requested, findYangFile() falls back from name@revision.yang to name.yang without verifying the module’s actual revision. This can load the wrong revision while the attached YangSchema reports the requested one.

There are also no new regression tests, and the PR currently has no CI status. The description only mentions a build with tests skipped. Please add tests covering invalid identityrefs, inactive nodes, missing list keys, cross-module augment ordering, and revision mismatches.
For these reasons, I recommend requesting changes before merging.

[Deepya1502]

The requested review changes have been addressed and verified. Regression tests pass, and the branch is ready for re-review.

[lllyfeng]

Thanks for the updates — identityref validation, missing list key handling, revision fallback and the new tests/CI all look good now. A few items from the previous review are still open, so requesting changes again:

1. Lenient mode is controlled by a static ThreadLocal (LenientValidationOptions). Please pass it as an explicit option/parameter instead; global thread-local state is fragile with thread pools and async callers.

2. CaseImpl/ChoiceImpl unconditionally add children whose if-feature is disabled, and CaseImpl swallows errors from addSchemaNodeChild (e.g. duplicate nodes). This weakens schema validation for all users. Please keep the existing behavior and surface the errors.

3. The cross-module augment test uses module names like "01-auger"/"02-target", which are not valid YANG identifiers, so the augment never resolves. With valid names, main already handles this case. Please fix the fixtures and show a case that actually needs the re-resolution logic, or drop it.

4. Please split out unrelated changes (YangStructure extending TreeNode, BaseImpl/IdentityImpl changes, ValidatorResultBuilder.getRecords returning a mutable list, debug logging in YangYinParser) into separate PRs, and update the PR description.

5. main has moved (#50, #52, #53 merged, including schema path changes), so please rebase and re-run the tests.

Thanks!

[Deepya1502]

Addressed the latest review requests in commit fb61e179 and rebased onto current main. The full mvn test reactor passes. The unrelated YangStructure and Base/Identity changes are now split into PR #54 and PR #55; the mutable result accessor, parser debug logs, and invalid augment workaround were removed from this PR. Ready for re-review.

[lllyfeng]

Thanks — this addresses the previous review well: explicit LenientValidationOptions instead of the ThreadLocal, Case/Choice now propagate addSchemaNodeChild results, augment re-resolution and the unrelated changes are gone. A few remaining items before merge:

1. YangSchemaContext.java:24/26 — please make getValidationOptions()/setValidationOptions() default methods (e.g. return LenientValidationOptions.STRICT) so external implementations don't break.
2. YangStatementImplRegister.java:72 — the BASE registration change (GRAMMAR+SCHEMA_BUILD) belongs to #55, which actually depends on it; please move it there and leave the registrations untouched here.
3. JsonCodecUtil.java:561 — the comment about "missing list key downgraded to a WARNING" no longer matches the code.
4. Minor: indentation at ChoiceImpl.java:322; missing trailing newline in LenientValidationOptions.java.

[Deepya1502]

Follow-up to the latest review: YangSchemaContext option accessors now have strict/no-op defaults for external implementations; BASE SCHEMA_BUILD registration and its regression tests are in #55; the stale list-key warning comment and formatting issues are corrected. Full mvn test passes.

[lllyfeng]

Thanks for the quick turnaround — default methods on YangSchemaContext, BASE registration moved out, stale comment fixed. Full reactor build/tests pass locally on top of main. Minor nit (non-blocking): indentation around ChoiceImpl.java:320-322. LGTM.