Martin Vogel b9b4669a20 fix(lsp): nested-class enclosing QN + constructor resolution to the class node ...

Runtime-diagnosed (LSPDIAG2 at the pass_calls join) LSP-strategy reds.

NESTED-CLASS ENCLOSING QN (cbm_enclosing_func_qn, helpers.c): a method in a
nested class was qualified with only the INNERMOST class (Inner), not the full
chain (Outer.Inner) that the def walk's compute_class_qn produces. So a call
inside a nested-class method sourced to the file node (callable=0) AND its
enclosing_func_qn failed to equal the LSP-resolved caller QN (strategy ABSENT,
e.g. lsp_outer_dispatch). Build the full outer->inner dotted class chain so the
enclosing QN matches the class/method node QN. Single-class and top-level cases
are unchanged (the chain collapses to the prior value).

CONSTRUCTOR RESOLUTION (cs_lsp.c cs_resolve_object_creation, java_lsp.c): `new
Foo()` synthesized a callee `Foo..ctor` / `Foo.Foo` whose ".ctor"/doubled short
name joined nothing (call callee_name is "Foo") and pointed at a non-existent
node -> no edge (cs_ctor, cs_ctor_synthetic, lsp_constructor, lsp_constructor_synth
all dead). A C#/Java constructor registers as <type_qn>.<ClassShortName>, so look
it up by the class short name; when no explicit ctor exists, resolve the call to
the existing Foo CLASS node (short name "Foo" == callee_name -> join matches, node
exists -> edge forms carrying the strategy).

Reproduced by repro_lsp_java_cs (nested-dispatch + constructor strategies RED).
NOTE: the *_unindexed / lsp_unresolved strategies are intentionally below the
confidence floor and target missing symbols (no edge can exist) -- flagged
separately as likely test-scope issues, not fixed here.

Signed-off-by: Martin Vogel <martin.vogel.tech@gmail.com>

2026-06-27 10:56:55 +02:00

..

cbm

fix(lsp): nested-class enclosing QN + constructor resolution to the class node

2026-06-27 10:56:55 +02:00