[aeberhart]

Fixes #95, #79, #83, #86

Root cause

jsonata-js partially applies a native function by building function($a, $b){ _ } and replacing the body with the native function (bodyAST.body = native). When invoked, applyProcedure calls applyNativeFunction, which does not validate the signature. The Java port instead generated a trampoline that called the builtin by name ($each($a, $b)), so the signature was checked again and the function had to be looked up by name.

Changes (following the JS code)

• partialApplyNativeFunction sets bodyAST.body = _native (Symbol.body is now Object); implemented the commented-out applyNativeFunction / getNativeFunctionArguments and the native branch in applyProcedure
• partialApplyProcedure: only ? leaves a parameter unbound; missing args are bound to null like JS binds undefined (incorrect result when $map($, $append(?)) #86)
• ? comparison: the token value is a Character, so arg.value.equals("?") was always false
• no lookup by name, so custom / unnamed JFunctions work (Custom function pipe error #83)
• $each / $merge iterate arrays like JS for..in (string index keys), since the signature is no longer enforced for partials (Incorrect each function signature validation #79)
• T1007 check was inverted (proc != null vs JS typeof proc === 'undefined')
• implemented the JS typeof proc === 'function' branch: Fn* lambdas bound via Frame.bind(...) can be partially applied

Arity of Java functions

JS takes the parameters of a native function from func.toString(). In Java, JFunction.getNumberOfArgs() now uses, in order:

1. the parameters of the implementing Method
2. the declared arity of a JLambda (Fn0–Fn8 → 0–8), kept by the Jsonata.function(...) factories
3. the number of parameters in the signature
4. unknown (raw JFunctionCallable / FnVarArgs without signature): the number of args supplied at the partial application, e.g. $add(?, 3) → 2. This is the only deviation from JS (no dropping of extra args / binding of missing ones)

API note

JLambda was an empty marker interface and now declares getJFunctionCallable() and a default getNumberOfArgs(). All Fn* types implement them; a class implementing JLambda directly would need to add getJFunctionCallable().

Testing

New PartialTest (11 cases); full suite: 1747 tests pass, also with -Dmaven.compiler.release=11. Compared 31 partial-application expressions against jsonata-js 2.1.0.

🤖 Generated with Claude Code

[github-actions]

Test Results

1 825 tests  +19   1 823 ✅ +19   8s ⏱️ +2s
  124 suites + 2       2 💤 ± 0 
  124 files   + 2       0 ❌ ± 0 

Results for commit 64ec3c7. ± Comparison against base commit 6850f8d.

♻️ This comment has been updated with latest results.

[aeberhart]

might be better to split this large PR into the following smaller parts:

• $each / $merge iterate arrays like JS for..in: only the Functions.java changes from ab7e60b and 64ec3c7. Since no expression can reach it yet, it would be tested by calling Functions.each / merge directly from Java. About 60 lines.
• Native partial application, matching jsonata-js (Incomplete implementation of partial functions #95, incorrect result when $map($, $append(?)) #86): the rest of ab7e60b plus cb16ba0. That covers the bodyAST.body = native approach, applyNativeFunction, the ? comparison fix, the T1007 fix, Parser.body becoming Object, the $each(?) / $merge(?) tests, and the Issues-79: fix each partial invoke #80 and ISSUES-86: fix arguments checked #87 tests from OldPrsTest.
• Partial application of callables and lambdas without a declared arity (Custom function pipe error #83): ef116f6 plus the Issues-83: Custom function pipe error  #85 test and the map_in_pipe resources. This one changes the public API, because JLambda gains getJFunctionCallable(). Reviewing it separately is the main benefit of splitting.

[sergeevik]

maybe extract (Symbol)result to var?
so many cast. so many code. and now it's twice cast

[sergeevik]

bad formatting
closing parenthesis looks like for "if (next instanceof Symbol)" but not

[aeberhart]

closing this PR. will be done in 3 smaller steps