Go: multiple init() per package collapse onto one QN — bodies and registration edges silently dropped

[ilyabrykau-orca]

Status: fix open — PR #1915 (file-stem+line #-suffix QNs for Go init, #495 pattern; stacked on #1913). Post-fix: init nodes 7 → 29, matching source exactly (source count: 29).

Version

codebase-memory-mcp 0.10.8 (also reproduced on a source build @ d706c33)

Platform

macOS (Apple Silicon)

Install channel

Built from source

Binary variant

standard

What happened, and what did you expect?

Go allows any number of func init() per package (several per file included); all run at program start. Every one gets the same QN (proj.pkg.init), so the upsert keeps one node per package and silently drops the rest with their bodies.

Measured: an event-source package registering each decoder in its own file's init() — 21 init functions collapsed into 1 node; repo-wide 29 in source, 7 in the graph. The registration topology (which decoder handles which event) is invisible; the surviving node's edges are a blend of merged bodies.

Reproduction

// a.go
package reg

var handlers = map[string]func(){}

func init() { handlers["a"] = handleA }

func handleA() {}

// b.go
package reg

func init() { handlers["b"] = handleB }

func handleB() {}

Actual: one node reg.init (a.go); b.go's body and its handleB edge gone or misattributed. Expected: two nodes.

Fix (existing #495 pattern)

Rust cfg twins already fold a #-suffix into the QN (rust_cfg_qualified_name); GoogleTest macros derive unique names (#1266). Same move: proj.pkg.init#a.go:L5, mirrored on the call-scope side. Nothing joins on the plain QN — calling init explicitly is illegal in Go — so no resolution path is affected.

Related: #1909, #1911 (same silent-upsert-collapse family); tracked in #1932.

Confirmations

• I searched existing issues and this is not a duplicate.
• My reproduction uses shareable code (a dummy snippet or a public OSS repository), not proprietary code.