[sujal-maheshwari2004]

What was wrong

priompt/priompt#33 changes the server's URI-to-subject mapping. Each URI segment becomes one subject token, and every byte outside [A-Za-z0-9_-] is percent-encoded with uppercase hex (. becomes %2E). _subject() only replaced / with ., so for a URI with a dot in a segment, subscribe() listened on a subject the server no longer publishes to, and hot reload got no events.

Fix

_subject() now ports pubsub.Subject / subjectToken from #33: split on /, then encode each segment's UTF-8 bytes. Dot-free URIs map exactly as before.

Tests

• New test_subject cases: priompt://acme.corp/x/a.b maps to priompt.acme%2Ecorp.x.a%2Eb, and agent-v2_x stays unchanged. Wildcard bytes and UTF-8 are encoded too.
• New test_subject_is_injective uses the server's own collision pairs from TestSubjectIsInjective: acme/a.b vs acme/a/b, acme/corp/x vs acme.corp/x, and acme/a.b/c vs acme/a/b.c.

before the fix: 6 failed, 18 passed
after the fix:  24 passed in 0.27s

Parity with the server: a throwaway Go test on #33's branch (91eb76b) printed Subject() for the same five URIs, and the output is identical to _subject():

priompt://acme/support/agent-v2_x -> priompt.acme.support.agent-v2_x
priompt://acme.corp/x/a.b -> priompt.acme%2Ecorp.x.a%2Eb
priompt://a/*/> -> priompt.a.%2A.%3E
priompt://a/café -> priompt.a.caf%C3%A9
priompt://a/b/ -> priompt.a.b.

Real run against a server built from #33's branch: subscribe to both URIs, then publish to each.

before:  priompt://acme/support/agent events: 1    priompt://acme.corp/x/a.b events: 0
after:   priompt://acme/support/agent events: 1    priompt://acme.corp/x/a.b events: 1

Compatibility

This matches the server after #33. Against a v0.7.2 server, URIs with a dot in a segment would stop getting events, because that server still publishes on the unencoded subject. Dot-free URIs work with both. Release this in step with priompt/priompt#33.

Fixes #10

🤖 Generated with Claude Code