2 Commits

Author	SHA1	Message	Date
郭立 Guo Li	0b7213c890	perf(mcp): defer analysis runtime after catalog discovery (#4424) ... Serve build-recorded default compact and explicit full catalogs, then materialize canonical analysis handlers on first use. Keep transport, Evidence delivery, profile identity and caller choices consistent. Share sealed MCP server ownership between eager and deferred lifecycle paths. Validate 125 focused cases, schema/contract repeats, negative controls, builds/static/docs checks and latest-main integration. Record 72 real SDK task trials with independent Rust metadata oracles and whole-family resource bounds; disclose shifted first-RPC cost and unchanged catalog tokens. Qualify the real Rizin small-target debug lifecycle separately. Automated review findings addressed; Codex review unavailable due repository usage limits.	2026-10-22 03:07:31 +00:00
morluto	00dd959b82	fix(mcp): derive shared provider cleanup from loaded ports ... Separate load-failure metadata from the typed optional observation ports and derive the process cleanup inventory from those ports. This includes the JavaScript runtime owner and removes the separate four-provider list. Preserve all load failures in server injection and keep shared providers caller-owned across individual server closure. Retain failed close hooks for retry while removing peers whose cleanup completed. Validated with 19 focused lifecycle/isolation checks and a final rerun of the new boundary regression, check, docs and exact test discovery. The runner test uses real idle providers with controlled SDK/shutdown handles and a close rejection; active Inspector resource consequences, native Linux/Windows and end-to-end stdio shutdown remain unverified. Fixes #5222	2026-10-19 18:03:22 +08:00