I've noticed that some tests for generated files go in test_generated and some in test_blob (like the Cargo.lock test). Is there a reason to choose one over the other?

Is there a reason to choose one over the other?

@spenserblack I believe test_blob.rb is for testing the functionality and expected behaviour of the Linguist::Blob class proper, while the test_generated.rb file holds individual tests for the contents of Linguist::Generated (the latter is more format-specific). @lildude will correct me if I'm wrong. 😉

@Alhadis is spot on. teat_blob.rb is essentially testing the helper methods used by the Blob class which in turn is doing the same things as the tests in test_generated.rb in this case, so there's no value in adding even more tests for generated to test_blob.rb, or test_file_blob.rb which is essentially the same, as that functionality isn't changing. Doing so only unnecessarily extends the time CI takes to run by effectively adding duplicate tests. This is one of a few places where a refactoring has left things in an incomplete and duplicating state. I've been meaning to find time to clean this all up for years 😁

This is one of a few places where a refactoring has left things in an incomplete and duplicating state. I've been meaning to find time to clean this all up for years 😁

@lildude One straightforward simplification might be to have test_heuristics.rb index the heuristics and test the samples automatically, unless a heuristics.yml field is define that explicitly disables this behaviour:

 - extensions: ['.tsp']
+  auto_test: false
   rules:
   - language: TypeSpec
     pattern: '^(import|using|namespace|interface|op|model|scalar|alias|union|enum)\s'
   - language: TSPLIB data
     pattern: '^(NAME|TYPE|DIMENSION|EDGE_WEIGHT_TYPE|EDGE_WEIGHT_FORMAT)\s*:'

I bring this up because I recently fixed a pedantic test failure in #6775 (392acb1), and all I needed to add to test_heuristics.rb was just copy+pasting the nearest def test_?_by_heuristics method and updating the language/extension names. This could be automated to use all_fixtures in the absence of an explicit test-method.

In fact, a lot of the pedantic-level tests could be fixed automatically without a contributor's input (possibly via a command-line switch). Specifically, the alphabetisation of arrays in languages.yml and friends.