[leoweigl]

Follow-up to #125: this adds tests for the problem filter list.

Why the small change in src/main

ProblemTextFilter.isFiltered(String) reads the filter list via SettingsState.getInstance(), which needs a running IntelliJ application. That makes the filter logic impossible to reach from a plain unit test.

So I split it up:

• The public isFiltered(String) still reads the setting, and then passes it to
• a new package-private overload isFiltered(String problemText, String rawFilterList), marked @VisibleForTesting, which contains the existing logic.
• getFilters() now takes the raw filter list as a parameter instead of reading the setting itself.

The behavior of the public method is unchanged, and the static cache is still keyed by the raw filter string, so a changed setting is still re-parsed as before.

What is covered

ProblemTextFilterTest covers the cases described in the Javadoc (15 test cases in total):

• plain entries match the text beginning, ignoring case
• leading spaces and blank entries (e.g. ;;) are ignored, so they don't filter everything
• glob entries match the whole text, and regex symbols inside a glob are treated literally
• re: entries match anywhere in the text, and the re: prefix is case-insensitive
• invalid regex entries are ignored instead of breaking the filter
• null problem text is filtered, null or empty filter lists filter nothing
• a changed filter list is applied (cache invalidation)

Testing

./gradlew test passes locally: ListenerTest 6/6, ProblemTextFilterTest 15/15.

The CHANGELOG has an entry under [Unreleased] → Added.

[0verEngineer]

A few remarks on the tests. The cache race in ProblemTextFilter is tracked separately in #127.

[0verEngineer]

This is the same assertion as the line above, so the sequence checks less than it looks like. Was this meant to use a different list or text? With only two lists that alternate, the test would also pass with a cache that is rebuilt on every call, so a third distinct list (e.g. field → *fiel → re:never → field) would make it actually exercise the invalidation.

[0verEngineer]

All tests share the static cache in ProblemTextFilter (cachedRawFilterList / cachedFilters) and never reset it, so what a test actually exercises depends on what the previous test left behind. For example, if nullFilterListIsIgnored runs first, Objects.equals(null, null) is true on the initial state and parse(null) is never called. A small package-private resetCache() (@VisibleForTesting) called from a @BeforeEach would make the tests independent of the method order.

[0verEngineer]

There is no test for an entry that both starts with re: and contains */?, so the regex-before-glob precedence in toFilter() is not covered. If the two checks were swapped, re:foo.*bar would become a glob with a literal re: prefix and no test would fail. Something like assertTrue(ProblemTextFilter.isFiltered("foo x bar", "re:foo.*bar")) would pin it down.

[0verEngineer]

Nit: ' is JUnit's default quote character in @CsvSource. It works here because the quotes are in the middle of the field, but a row whose text starts with an apostrophe (like 'Thread.sleep()' called in loop used further down) would be parsed as a quoted field. Setting quoteCharacter = '"' avoids that trap for future rows.