fix(#255): correct the continue-re-stop diagnosis — multi-location macro breakpoints

[debugmcpdev]

Summary

Validating the Rust trio (#272/#273/#274/#276) on Ubuntu surfaced a failing test on main: the strict non-win32 reproducer added in #274 asserts that one continue_execution from a breakpoint at examples/rust/hello_world/src/main.rs:26 runs the program to completion. It does not, on Linux — but not because of the bug #255 describes.

Root cause

Line 26 is let message = format!("Language: {}, Version: {}", name, version);. The format! expansion inlines several call sites onto that one source line, so LLDB plants a breakpoint location at each — the setBreakpoints response says Resolved locations: 3. Each continue advances to the next location, so the session re-pauses on the same file:line, with the same breakpoint id, at a different program counter. The line takes three continues to leave.

Confirmed with a raw DAP client driving the vendored CodeLLDB directly (mcp-debugger entirely out of the loop):

breakpoint line	source	resolved locations	continues to completion
26	let message = format!(...)	3	3
13	let name = "Rust";	1	1
42	let sum = a + b;	1	1

Each intermediate stop reports reason: "breakpoint", hitBreakpointIds: [1], line: 26, col: 19 — at PCs …C6FF, …C73A, then termination. Also ruled out the #236 post-stop setBreakpoints re-sync by disabling it: no change.

So this is ordinary LLDB multi-location behavior on every platform — not Windows-specific, not MSVC/PDB-specific, and not ours. The prior "Linux remains clean" note only held because nothing in the suite had ever continued past a user breakpoint on Linux, which is the exact gap #274 set out to close.

Changes

• tests/e2e/mcp-server-smoke-rust.test.ts — replaces the wrong-premise reproducer with two tests sharing one launch helper, both running on all platforms (the win32 special-casing and its step_over workaround are gone):
  • a single-location line (42) that must complete in exactly one continue — the coverage gap that let this survive;
  • a multi-location macro line (26) that must stay on line 26 between stops, drain its locations within a bound, and then exit 0 in more than one continue.
• skills/debugging/references/rust.md — the quirk bullet and troubleshooting row described a Windows-only re-hit bug with a step_over workaround; they now describe multi-location macro breakpoints and the correct guidance (keep continuing, or step_over once to cross the whole line).
• mcp-server-breakpoint-management.test.ts / mcp-server-smoke-restart.test.ts — comments corrected. Behavior deliberately untouched: both already tolerate multi-location stops (one clears breakpoints first, the other loops), and Windows can't be exercised from this machine.

Testing (Ubuntu 24.04, x86_64, CodeLLDB 1.11.8, rustc 1.91.1)

• mcp-server-smoke-rust 4/4, mcp-server-breakpoint-management 8/8, mcp-server-smoke-restart 8/8
• Full unit + integration with LEAK_GUARD_STRICT=1: 2852 passed
• Lint and typecheck clean

Also verified on Linux while here (no changes needed)

• Rust: panic pause reports lastStop.reason 'breakpoint' instead of 'exception'; CodeLLDB no longer advertises rust_panic #260 — the Linux leg PR fix(#260): normalize Rust panic stops from 'breakpoint' to 'exception' #273 deferred to "ubuntu CI", which never runs it: test:ci-coverage is --project unit --project integration, so no e2e runs in CI at all. The panic trio passes; Linux matches the Windows capture exactly — stopped body {"allThreadsStopped":true,"hitBreakpointIds":[1],"reason":"breakpoint","threadId":…} with no description/text, session reporting {reason: 'exception', rawReason: 'breakpoint'}. An ordinary user breakpoint still reports breakpoint, so the id-disjointness discriminator does not false-positive here.
• adapter-rust: consolidate triplicated CodeLLDB platform-dir/candidate-path logic #265 — real Rust sessions launch, so the consolidated resolver genuinely finds the vendored linux-x64 CodeLLDB (the new unit tests are hermetic and can't show this).
• fix(#275): normalize Rust/Windows break-in pause stops to 'pause' #276 — POSIX pause unaffected: stopReason: 'pause', rawStopReason: 'exception'.
• fix(#255): continue-from-breakpoint reproducer + honest stop-reason fallback #274's cross-language 'unknown' fallback — 42 e2e tests across python/javascript/go/java/ruby/dotnet/mock/restart/logpoints/breakpoint-management, all green.

Closes #255

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[debugmcpdev]

Code review

No issues found. Checked for bugs and CLAUDE.md compliance.

Verified on the platform this PR could not exercise: all 4 tests in tests/e2e/mcp-server-smoke-rust.test.ts pass on Windows 11 (GNU toolchain, CodeLLDB 1.11.8) — the single-location line-42 test completes in exactly one continue, and the line-26 macro test drains its locations and exits 0. This confirms the diagnosis on Windows: the re-stop follows breakpoint-location count, not the OS.

🤖 Generated with Claude Code

_{- If this code review was useful, please react with 👍. Otherwise, react with 👎.}