[alexanderlinne]

Why

The next three PRs rewrite the component and dependency patterns, so this PR first makes them readable and pins what they match without changing it. PlantUmlPatterns was hard to read and to change:

• The component regex was assembled from five static readonly string fields, some set in a static constructor, through Capture and AnythingBut helpers, and anchored twice (\A(?:^ … )\z). To see what it matched, you had to put it together in your head.
• Stereotypes were read by a second regex over the whole line, instead of from the stereotype section of the component match. MatchStereoTypes() stepped a stored Match forward, so a second call returned an empty set.
• Every component line was matched twice, once by FilterComponents and again by the PlantUmlComponentMatcher constructor, which threw if the line didn't match.
• PlantUmlPatterns was an object without state that the parser created with new, and the "matchers" were really parsed results, read through Match…() methods.

What changes

Tests first. The first commit adds tests through PlantUmlParser only, so that they survive the refactoring:

• the error paths: Parse(null) for a file and a stream, a missing file, and a stream whose Read throws,
• characterization tests for every optional part of the patterns: stereotypes with and without whitespace between them, quoted and unquoted aliases used with and without brackets, surrounding whitespace, and lines that must not count as components,
• dependencies in both directions, with and without brackets and a : description, and lines that must not count as dependencies.

All of them pass on main.

Patterns. Each regex is one readable definition. PlantUmlPatterns becomes a static class that holds the shared stereotype fragment, a named and repeated stereotype group read through Captures.

Matchers. PlantUmlComponentMatcher.TryParse returns null for lines that aren't components, and PlantUmlDependencyMatcher.TryParse replaces TryParseFromLeftToRight and TryParseFromRightToLeft. Each line is matched once, and Name, Alias, Stereotypes, Origin and Target are properties. FilterComponents, MatchComponent, the unreachable throw and the unreachable null checks go away. The matchers stay in PlantUmlPatterns.cs, so that the PRs on top rebase cleanly.

Parser.

• It calls the TryParse methods directly, without the PlantUmlPatterns instance.
• The two ReadLines overloads, which repeated the same catch … when filter, become one.
• A redundant Distinct() before a HashSet is dropped.
• Comment lines are filtered without building a regex for every line.

Main's way of splitting dependency lines stays as it is; #PR2 replaces it.

After this PR, PlantUmlPatterns.cs and PlantUmlParser.cs have 100% line and branch coverage.

Behaviour change

Because stereotypes are now read from the component match, text that only looks like a stereotype is no longer read as one. Before, [A] <<Real.*>> as "<<Fake.*>>" gave A the stereotypes Real.* and Fake.*, and [A] as "<<Foo>>" was accepted although it has no stereotype. A regression test pins this. Apart from that, the import matches exactly what it matched before.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 86.72%. Comparing base (f439e0d) to head (edf1e95).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #521      +/-   ##
==========================================
+ Coverage   86.55%   86.72%   +0.17%     
==========================================
  Files         261      261              
  Lines       12648    12559      -89     
  Branches     1236     1231       -5     
==========================================
- Hits        10947    10892      -55     
+ Misses       1359     1330      -29     
+ Partials      342      337       -5     

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.