SymbolicML / DynamicExpressions.jl / 22823314614

97%

master: 96%

push

github

deps: SymbolicUtils v4 compatibility (release-v1 backport) (#161)

* fix: backport SymbolicUtils v4 compat to release-v1

Restore index_functions round-trip for custom operators under SymbolicUtils v4.

* test: avoid string matching for SymbolicUtils roundtrip

* chore: format test_symbolic_utils.jl

* fix: use term() guards to match master behavior for commutative ops

Adds SymbolicUtils.term() usage to prevent SUv4 from canonicalizing
x*x to x^2 during conversion. This makes the backport behaviorally
equivalent to upstream/master.

- Import term from SymbolicUtils
- Use term(op, ...) for * and + in parse_tree_to_eqs
- Use term(*, ...) in multiply_powers to build explicit products
- Restore test to check x1*x1 stays as explicit multiplication

* ci: add ext/ to workflow paths for fork validation

* ci: trigger workflow

* fix(symbolicutils): handle atoms after multiply_powers

* SU v4: unwrap Const scalar containers + regression tests

* style: run JuliaFormatter

* style: format (JuliaFormatter v1)

* chore: drop CI-trigger / revert unrelated formatting

* chore: revert unrelated changes (keep SUv4 patch minimal)

* test: use BasicSymbolic in simplification tests (SU v4)

* test: fix JET/Aqua and chainrules compatibility under SU v4

* test: ignore OperatorEnumConstruction in JET checks

* fix: silence JET undef-var in precompile workload

JET flagged "local variable  may be undefined" in src/precompile.jl when iterating over a Vector of types.\n\nUse a tuple for the type loop and initialize  so the value is always defined.

* test: call bad_op via @eval in chainrules NaN regression

Switch the chainrules NaN-mode regression to call unary test operators through  instead of explicit Node constructor, while keeping the test resilient to runtime method-extension timing on Julia 1.12. Also add explicit Node-type sanity checks before gradient evaluation.

* test: inline @eval for bad_op calls in chainrules NaN regression

Restore the original  structure ... (continued)

58 of 70 new or added lines in 2 files covered. (82.86%)

2618 of 2692 relevant lines covered (97.25%)

37133.47 hits per line