Split mixed array ranks into separate variable declarations (#100)

[paulirwin]

Fixes#100.

Problem

Java allows C-style array brackets on individual declarators, so one declaration can mix array ranks:

intmulti[][] = newint[2][2],
single[] = newint[2];

This failed conversion outright with AssertionError: The variables do not have a common type. The error comes from JavaParser's getCommonType(), which asserts every declarator shares a type. Because it was called before the visitor's own array-level check, the friendlier InvalidOperationException on the next lines was unreachable.

Fix

Group declarators by array level and emit one C# declaration per distinct rank.

The groups are emitted as flat sibling statements through the existing PendingStatements mechanism rather than a nested block — a block would put the variables in an inner scope and break every later reference to them. Declarators that share a rank stay together in a single declaration, and declaration order is preserved.

// int single[] = new int[2], scalar = 7, other[] = {8, 9};int[]single=newint[2],other=new[]{8,9};intscalar=7;

Scope

This covers local variable declarations, matching the issue's repro. FieldDeclarationVisitor has the same limitation, but VisitForClass returns a single MemberDeclarationSyntax, so splitting fields would require changing that signature across every body-declaration visitor. That felt out of scope here; fields remain unsupported, as already noted in ArrayField.java.

Note the pre-existing, separate limitation that jagged arrays (int[][]) are emitted as rectangular (int[,]) while indexing stays [0][0] — that's the known issue behind MultidimensionalArrays.java being conversion-only, and is untouched by this PR.

Testing

Unit tests — new ConvertMixedArrayRankDeclarationTests covers the split, same-rank grouping, declaration order and absence of a nested scope, uninitialized declarators, the #100 repro, and a regression guard that single-rank declarations are unchanged.

Integration test — new MixedArrayRankDeclarations.java registered in FullIntegrationTests, which compiles the generated C# with Roslyn, runs it, and asserts on stdout. It converts with no warnings.

Full suite: 389 passed, 0 failed. I also confirmed 6 of the 7 new tests fail without the fix and pass with it (the 7th is the unchanged-behavior guard, which correctly passes both ways).

🤖 Generated with Claude Code

[paulirwin]

Added MixedArrayRankMultidimensional.java, covering the exact int multi[][], single[] declaration from the issue.

One caveat on where it's registered: it goes in GeneralSuccessfulConversionTest (conversion-only) rather than FullIntegrationTests, because FullIntegrationTests compiles and runs the output, and this case still can't compile:

int[,]multi=newint[2,2];// rectangularint[]single=newint[2];multi[0][0]=1;// ...but indexed as jagged

The mixed-rank split is correct — the two ranks land in separate declarations. What's still broken is the pre-existing jagged-vs-rectangular bug already tracked by MultidimensionalArrays.java, which this PR doesn't touch. Same reason that file is conversion-only.

So the coverage is now in two parts:

• MixedArrayRankDeclarations.java — runnable, in FullIntegrationTests, compiles and asserts stdout, no warnings.
• MixedArrayRankMultidimensional.java — the issue's literal example, conversion-only.

Also tightened the multi-dimensional unit test to assert the split explicitly (int[, ] multi = ... / int[] single = ...) rather than just that conversion stops throwing.

390 tests pass. Confirmed the new conversion test fails against the pre-fix visitor and passes with the fix.