Two applies can change different tables in the same config. This regression checks that each
successful apply's update survives, or that the competing apply explicitly refuses to write.
The fixture updates an MCP server and a profile; each apply uses --keep to preserve the other's table.
You need Node (declared for 18 or newer; last run on 22 and 24), a POSIX shell, and npm and tar for the second step. No dependency install is needed. From the root of a clean checkout, copy the code and tests into a temp folder:
overlap_demo=$(mktemp -d)
cp -R bin tests "$overlap_demo/"
cd "$overlap_demo"Run the regression:
node tests/review-overlap.test.jsThe tests create temporary config files, pass their paths explicitly to the CLI, and remove them after each case. They do not use your real Codex config.
It exits 0 and prints:
ok CMP sequential mcp then profile keeps both compatible updates
ok CMP sequential profile then mcp keeps both compatible updates
ok CMP a peer commit before the final check is merged or explicitly refused
ok BUG a peer commit during backup is kept or explicitly refused
4 passed, 0 failed
Now put the last release without the lock, 0.1.1, under the same test. This step needs network access:
npm pack [email protected] --silent
tar xzf codex-config-sync-0.1.1.tgz
cp package/bin/* bin/
node tests/review-overlap.test.jsThe last command exits 1 and prints:
ok CMP sequential mcp then profile keeps both compatible updates
ok CMP sequential profile then mcp keeps both compatible updates
ok CMP a peer commit before the final check is merged or explicitly refused
FAIL BUG a peer commit during backup is kept or explicitly refused
both applies exited 0, but the peer's committed profile was lost; backup retains it: false
3 passed, 1 failed
Both applies reported success, but the profile written by the second apply disappeared from the live config and its backup.
With the lock, the first apply holds the lock through its write; the second apply waits, then refuses with exit 2 while that lock is held. The two sequential cases confirm that both updates survive when the applies run one after the other, and repeating the last apply leaves the config and backup alone.
The test pauses the first apply to force the overlap; this does not mean every sync loses settings.