I noticed a failing CI job in https://github.com/Codeception/Codeception/actions/runs/19833673725/job/56825691142 when behat/gherkin:4.15 is used.

The same tests are green when using behat/gherkin:4.14.

Codeception uses Behat\Gherkin\Keywords\CachedArrayKeywords::withDefaultKeywords() when initializing the lexer.

The test tries to load all gherkin files.

File 1 and 3 are using the default language "en" whereas the language "ru" is defined for file 2.

File 1

Feature:
  In order to know that the right step will be executed
  As a user
  I should be notified about steps that are ambiguous

  Scenario: We have multiple step definitions that match a given step
    Given a step which matches multiple step definitions

File 2

#language: ru

Функционал: Генерация сниппетов из сценариев на других языках
  Как пользователь
  Чтобы выполнить проверку сценариев на родном языке заказчика
  Я хочу иметь возможность генерировать сниппеты из них 

  Сценарий: Генерация сниппета из сценария на русском языке
    Допустим я написал сценарий на языке "RU_ru"

File 3

Feature: Run gherkin
  In order to test a feature
  As a user
  I need to be able to have subfolders

  Scenario: Describe a feature in subfolder
    Given I have a feature in a subfolder

In behat/gherkin:4.14 this works without problems.
However, in behat/gherkin:4.15 the lexer seems to use the language "ru" for file 3 and thus parsing the file fails.

The test becomes green when adding #language: en to file 3.

I'm pretty sure this is related to #350 because before this PR, "en" was used as second parameter when calling Lexer::analyse() inside Parser::parse().

Is the behavior change intended or does codeception just use behat/gherkin the wrong way?