Validate change/variable targets against the model they reference

[jcschaff]

Fixes #167.

validate_doc reports valid repeatedTask setValue targets as invalid, because the XPath is evaluated against the last <model> in the document rather than the model the change references.

Cause

model_etrees is built in for model in doc.models: (L131, rebound at L229). Three later validate_target(...) calls then read model_etrees.get(model, None) — the leaked loop variable, which by then is always the last model — where they mean the model belonging to the change or variable being validated:

line	context	now passes
359	repeated-task change variable	model_etrees.get(variable.model, None)
509	repeated-task setValue change target	model_etrees.get(change.model, None)
551	setValue variable	model_etrees.get(variable.model, None)

Every other argument at those call sites already comes from change.model / variable.model — only the etree was taken from the stale name. That is why the message names a model that was never searched, and the element it claims is missing is plainly present in the named model's source.

L190 and L1408 also read model_etrees.get(model, ...) but use genuine loop variables; they are untouched.

Test

test_validate_repeated_task_xpaths_use_the_referenced_model validates one document twice, swapping only the order of two models in doc.models. The change references the same model and carries the same target both times, so both orderings must be valid.

On dev the new test fails in one ordering with exactly the reported message:

AssertionError: Lists differ: [['Task `task2` is invalid.', ...
-  'does not match any elements of model `model1`.']]]]]]]] : unexpected errors with model order ['model1', 'model2']

and passes in the other — the ordering alone decides it. With the fix both orderings pass.

The existing repeated-task fixtures (BIOMD0000000297_repeat_*.sedml) all have a single model, where the leaked variable happens to equal the correct one, which is why this went unnoticed.

Verification

tests/sedml/ run against this branch versus the same clone with only the validation.py change reverted:

	failures
without fix	4 — 3 pre-existing + the new test
with fix	3 — the 3 pre-existing only

The 3 pre-existing failures (test_validate_model_with_language, test_get_parameters_variables_for_simulation, test_build_combine_archive_for_model_xpp_with_plot) are missing optional dependencies in my environment — ModuleNotFoundError: No module named 'libcellml' and XPP — and fail identically before and after. Nothing else changes.

Deliberately not included

Two adjacent points raised in #167 are left out, since they are behaviour calls rather than clear bugs and I did not want to bundle them:

• L551 passes model_change=change.model and change.model.has_structural_changes() where its sibling at L359 uses variable.model. Since the target validated there is variable.target, variable.model looks right, but it can turn warnings into errors for some documents.
• Model.has_structural_changes() returns True for ModelAttributeChange, contradicting its docstring (add/replace/remove only). This is what demotes genuine XPath misses to warnings for any model carrying changes, so it masked half of this bug — but tightening it would surface errors in documents that validate today.

Happy to fold either in, or open them separately, whichever you prefer.

🤖 Generated with Claude Code

https://claude.ai/code/session_01SvqmME7MkRUNEYje5HpiLt