[hikmetba-bit]

Summary

Fixes #84.

dirty_put has two clauses: one that matches %ConCache.Item{ttl:, value:} and unwraps it into perform_put(cache, key, value, ttl), and a fallback for a plain value that uses the cache's default ttl. dirty_insert_new had no such split — it stored whatever was passed straight into ETS, so an %ConCache.Item{} ended up cached as the raw struct instead of its .value, and its .ttl was silently ignored in favor of the cache's default ttl.

ConCache.put(cache, 1, %ConCache.Item{value: %{a: 1}, ttl: :timer.seconds(60)})
ConCache.get(cache, 1)  # => %{a: 1}                          (correct)

ConCache.insert_new(cache, 2, %ConCache.Item{value: %{a: 1}, ttl: :timer.seconds(60)})
ConCache.get(cache, 2)  # => %ConCache.Item{value: %{a: 1}, ...}   (wrong)

Fix

Mirrors dirty_put's two-clause pattern for dirty_insert_new via a new perform_insert_new/4, so ConCache.Item behaves the same way for both functions.

Testing

• Added a regression test asserting insert_new with a %ConCache.Item{} unwraps to the plain value.
• Verified the new test fails against the code without this fix (gets the raw struct back), and passes with it.
• mix test test/con_cache_test.exs: 54/60 passing both before and after this change — the 6 remaining failures are pre-existing on unmodified master too (timing-sensitive ttl-expiry tests using ~10-40ms sleeps, flaky in this sandboxed environment; confirmed identical failure set on master).
• mix compile --warnings-as-errors: no new warnings from the changed file (the one pre-existing warning, unused require Logger in lib/con_cache.ex, is unrelated and untouched by this change).

🤖 Generated with Claude Code