[eitsupi]

CRAN's MKL check exposed a race in the mock test server: the Unix socket path can appear before the server is listening. A refused connection leaves the subprocess waiting, and the old collector then compares a NULL exit status with zero, producing a secondary error.

Skip subprocess/socket integration tests on CRAN at the shared local and TCP mock-server entry points, including direct callers. Publish local readiness only after socket creation and a zero-timeout poll starts connection waiting. On Windows, this registers ConnectNamedPipe before a fast client can connect and close, avoiding ERROR_NO_DATA during the first blocking poll. Share collection logic that terminates timed-out subprocesses and distinguishes unknown, nonzero, and successful exit statuses. Manage temporary paths in the caller's scope with withr.

The existing R check workflow sets NOT_CRAN=true, so integration tests continue to run in CI. No production transport, package version, or E2E workflow changes are needed.

Validation:

• All six PR CI jobs passed: R CMD check on Ubuntu (R devel and 4.1), Windows (R 4.1), and macOS (R release), plus E2E on Ubuntu and Windows.
• Windows i386 and x64 each passed 226 expectations with no failures or warnings; R CMD check reported Status: OK.
• Local regression and frame-output tests passed 56 expectations with no failures, warnings, or skips, including delayed socket construction, delayed blocking polling, immediate client close, and automatic cleanup.
• CRAN-mode full suite passed with 58 expectations and 74 skips; replacing callr::r_bg with an error confirmed that no mock subprocess was started. The follow-up CRAN regression/frame-output run also passed without subprocess calls.
• Local R CMD check --no-manual --as-cran: no errors or warnings; one NOTE about this environment's compilation flags. R syntax and git diff --check passed.

Fixes #75

[grantmcdermott]

Thanks for this @eitsupi. I've been slammed at work and just coming up for air.

We need to submit this patch for CRAN. Is there anything else that you'd like to roll into a 0.2.1 release? (I'll open a separate submission issue later today.)

[eitsupi]

Thank you. I don't have anything in particular to add.