lsp_all.o not rebuilt when cbm.h changes → stale-object ODR mismatch (ASan stack-buffer-underflow in cbm_calls_push) #661

Version

codebade-memory-mcp 0.8.1

Platform

Linux (arm64)

Install channel

Built from source

Binary variant

standard

What happened, and what did you expect?

Makefile.cbm builds lsp_all.o (the unity object for all lsp/*.c resolvers) with a rule that depends only on lsp_all.c, not on the headers it transitively includes. Because lsp_all.c pulls in cbm.h, and CBMCall is copied by value across the lsp→pipeline boundary (cbm_calls_push), any change to CBMCall's layout leaves a stale lsp_all.o on incremental builds, an ODR mismatch. ASan flags it as a stack-buffer-underflow.

Reproduction

On a tree where lsp_all.o predates the most recent CBMCall change (i.e., any incremental build spanning #477):

$ make -f Makefile.cbm build/c/test-runner   # incremental; lsp_all.o NOT rebuilt
$ ./build/c/test-runner                       # suite_rust_lsp
==NNNNN==ERROR: AddressSanitizer: stack-buffer-underflow ... in __interceptor_memcpy
    #1 cbm_calls_push            internal/cbm/cbm.c:128
    #2 rust_inject_syn_call      internal/cbm/lsp/rust_lsp.c:3366
    #3 rust_emit_resolved_call   internal/cbm/lsp/rust_lsp.c:3382
    ...
    #19 test_rustlsp_dbg_macro_inner_call  tests/test_rust_lsp.c:861

  This frame has 1 object(s):
    [48, 352) 'call' (line 3363) <== Memory access at offset 0 partially underflows this variable

Confirm it's staleness, not a source bug:
$ ls -l build/c/lsp_all.o internal/cbm/cbm.h   # lsp_all.o older than cbm.h
$ make -f Makefile.cbm clean-c && make ... && ./build/c/test-runner   # clean build → no abort

Logs

Project scale (if relevant)

No response

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.