[nealrichardson]

Fixes #57.

Problem

is_port_open() made one TCP connect right after the container started. On Docker Desktop the published host port can take a fraction of a second to bind after containers.run() returns. During that window the connect is refused at once, so with-connect reported "Posit Connect did not start within 60 seconds" after a few milliseconds, then stopped a container that was starting normally. The timeout=60.0 only limited how long a single connect attempt could take. It didn't make with-connect keep trying.

Fix

• is_port_open() retries every 0.25 seconds (poll_interval) until the timeout expires. Each attempt's connect timeout is capped at the time left. After a failed attempt it sleeps for whichever is shorter, the poll interval or the time left, so it keeps trying right up to the deadline.
• is_port_open() takes an optional container. On each attempt it calls container.reload() and stops waiting if the status is exited or dead, so a Connect that crashes on startup doesn't wait the full 60 seconds. The call site passes the container and raises "Posit Connect container exited before port N opened." in that case. A timeout still raises the "did not start within 60 seconds" error.

Verification

Run locally on macOS with Docker Desktop and ghcr.io/posit-dev/connect:latest:

• uv run main.py --license README.md --image ghcr.io/posit-dev/connect:latest --port 43821 -- true got past "Waiting for port 43821 to open...", logged "Running Posit Connect v2026.10.0", and failed as expected with "Unable to obtain a valid license" (README.md stands in for a license, since no valid one was available).
• The race didn't reproduce on this machine with the unfixed code; containers.run() returned after the port was already bound. To exercise it, a script polled a port and started the container 2 seconds later with docker run -p:
  • A single connect (the old behavior) got ConnectionRefusedError after 0.001s.
  • The new is_port_open() returned True after 2.3s and 10 attempts.
  • With a container whose entrypoint is false, is_port_open(..., container=c) returned False right away (status exited) instead of waiting 60s.

Tests

Added after test_extract_server_version_dev and registered in the __main__ block:

• test_is_port_open_retries_until_port_opens: two refused connects, then a success, gives True after 3 attempts.
• test_is_port_open_times_out_on_closed_port: a closed local port gives False after the full 0.5s timeout, not right away.
• test_is_port_open_stops_when_container_exits: an exited container gives False without trying to connect.

uv run test_main.py and uv run --with pytest pytest -q pass.

Merge conflicts with other open PRs

• fix: Fail fast when the command after -- is missing or can't be run #69 (fix-command-errors), fix: Report Docker and installed-command errors as a short message #70 (clean-errors), fix: Stop Connect containers promptly by signaling Connect directly #71 (fast-stop), feat: Use a free port when 3939 is in use and --port isn't given #73 (auto-port): merge cleanly (checked with git merge-tree).
• feat: Stream Connect's startup logs to stderr while waiting for it to start #72 (stream-startup-logs): one small conflict in main.py at the is_port_open call site, because both PRs change the same lines. To resolve it, keep container=container and the exited-status check from this PR, and replace the two Container logs: print lines with feat: Stream Connect's startup logs to stderr while waiting for it to start #72's startup_logs.dump(). test_main.py merges cleanly.

🤖 Generated with Claude Code