⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 68.80%. Comparing base (9bf4b40) to head (9d39fd7).
❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

@@           Coverage Diff           @@
##             main     #725   +/-   ##
=======================================
  Coverage   68.79%   68.80%           
=======================================
  Files          46       46           
  Lines       30999    31000    +1     
=======================================
+ Hits        21325    21327    +2     
+ Misses       9674     9673    -1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

REV: fix(connection): check socket file existence in default_host(), not directory

Verdict: REQUEST CHANGES — the fix is directionally correct and the bug is real, but there are two issues worth addressing before merge.

1. Correctness: directory scan matches .s.PGSQL.<port>.lock too

The .starts_with(".s.PGSQL.") predicate matches both the Unix-domain socket (.s.PGSQL.5432) and the lock file (.s.PGSQL.5432.lock). The lock file is a regular file created alongside the socket by postmaster. In practice this means has_socket can be true if only the lock file is present — which could happen in a crash scenario where the socket was cleaned up but the lock file wasn't. Not a common case, but it's a latent false positive that partially reintroduces the bug this PR is trying to fix.

The fix is straightforward: check file type explicitly.

e.file_type().map(|ft| ft.is_socket()).unwrap_or(false)
    && e.file_name().to_string_lossy().starts_with(".s.PGSQL.")

Or alternatively, since the port is already known (defaults to 5432), skip the directory scan entirely and check directly:

let socket = path.join(format!(".s.PGSQL.{}", DEFAULT_PORT));
if socket.exists() {
    return (*dir).to_owned();
}

The direct check is O(1) vs O(n) directory scan and cannot accidentally match a socket for a different port (see point 2).

2. Correctness: scanning all sockets ignores the target port

The current implementation returns a directory if any .s.PGSQL.* socket exists in it, regardless of port. If /tmp/.s.PGSQL.5433 exists but there is no .s.PGSQL.5432, the code returns /tmp and then the connection attempt to port 5432 fails. The user gets a confusing error: the directory was "detected" as having Postgres, but the connection still fails.

libpq sidesteps this by not probing at all — it just tries to connect and lets the OS return ENOENT or ECONNREFUSED. If we want to probe, we should probe the specific socket that will actually be used.

3. Performance: read_dir() on every default connection attempt

default_host() is called from Default for ConnParams and from resolve_params(), which is on every connection path when no host is specified. read_dir() is a syscall + directory traversal. For the common case (a single Postgres instance), this is one stat() call if we use the direct path check instead of read_dir(). Not a significant concern in practice for a terminal tool, but the direct approach is both simpler and faster.

4. Error handling: silent .ok() on read_dir()

Silently ignoring read_dir() errors is acceptable here — if we can't read the directory, we can't tell if a socket exists, and falling back to localhost is the right default behavior. No concern.

5. Doc comment: backtick-quoted PostgreSQL is unconventional

contains an active \PostgreSQL` socket file— the backticks aroundPostgreSQL` look odd; that's identifier formatting, not prose formatting. Just write it as plain text: "contains an active PostgreSQL socket file".

6. Unrelated: version bump and CHANGELOG belong in PR #720

This PR contains two commits: the fix (correct) and a chore: bump version to 0.8.1 commit. The Cargo.toml bump to 0.8.1 and the new ## [0.8.1] CHANGELOG entry were already added in the separate version-bump PR #720. Having both in this PR creates a merge conflict risk and makes the history harder to read. The version bump commit should be dropped from this PR.

Note also: the CHANGELOG entry for 0.8.1 describes fixes #708 and #711, not this fix (#721). Either this fix should be added to that entry, or the entry is misplaced here.

Summary

The simplest fix for 1, 2, and 3 together: replace the read_dir() scan with a single path.join(".s.PGSQL.5432").exists() check (using the resolved port, or the default if not yet resolved).