Context
PR work added transpile <src> --source cpp --target <t> --cbm <repo_root> which recovers the cross-file class/method/macro surface from a codebase-memory-mcp knowledge graph and injects it as a parser preamble (issue #79's multi-file parse wall, automatic OCCT shim for #79's OCCT wall).
Blocker found during testing: the codebase-memory-mcp cli index_repository one-shot CLI crashes on a worker subprocess on this Windows host (ROCm clang 21 present; no Go toolchain to rebuild from source). The already-indexed repos under C:/Github/* work (live query_graph confirmed against the 175K-node transpilers graph), but a new repo like wassimj/Topologic cannot be indexed here via the one-shot CLI.
What's needed to run the Topologic 51-file pass end-to-end:
- Index the target repo via the watcher-server mode (
codebase-memory-mcp index <repo> --project X, which runs the server rather than the one-shot worker) — OR
- Build the fork from source (needs Go + the C toolchain for vendored deps, absent here) — OR
- Use a released
codebase-memory-mcp binary that doesn't hit the worker crash (the v0.9.0 static binary is 114 commits behind the fork HEAD 8c9ac98; the post-v0.9.0 commits are cypher refinements, not API changes, so parity is close but not exact).
Scope of this issue: make the Topologic 51-file pass reproducible — pick one of the above, index wassimj/Topologic, and run transpile TopologicCore/src/*.cpp --source cpp --target mojo --cbm <topologic> to confirm the parse rate moves off 0/51. The engine-side fix (class/method recovery, OCCT shim) is already done and regression-tested; only the indexing step remains.
Acceptance: git status shows a measurable (non-0) parse success rate on the Topologic corpus with --cbm, vs the prior 0/51.
Context
PR work added
transpile <src> --source cpp --target <t> --cbm <repo_root>which recovers the cross-file class/method/macro surface from acodebase-memory-mcpknowledge graph and injects it as a parser preamble (issue #79's multi-file parse wall, automatic OCCT shim for #79's OCCT wall).Blocker found during testing: the
codebase-memory-mcp cli index_repositoryone-shot CLI crashes on a worker subprocess on this Windows host (ROCm clang 21 present; no Go toolchain to rebuild from source). The already-indexed repos underC:/Github/*work (livequery_graphconfirmed against the 175K-node transpilers graph), but a new repo likewassimj/Topologiccannot be indexed here via the one-shot CLI.What's needed to run the Topologic 51-file pass end-to-end:
codebase-memory-mcp index <repo> --project X, which runs the server rather than the one-shot worker) — ORcodebase-memory-mcpbinary that doesn't hit the worker crash (the v0.9.0 static binary is 114 commits behind the fork HEAD8c9ac98; the post-v0.9.0 commits are cypher refinements, not API changes, so parity is close but not exact).Scope of this issue: make the Topologic 51-file pass reproducible — pick one of the above, index
wassimj/Topologic, and runtranspile TopologicCore/src/*.cpp --source cpp --target mojo --cbm <topologic>to confirm the parse rate moves off 0/51. The engine-side fix (class/method recovery, OCCT shim) is already done and regression-tested; only the indexing step remains.Acceptance:
git statusshows a measurable (non-0) parse success rate on the Topologic corpus with--cbm, vs the prior 0/51.