Explain the internal working of Cucumber's Pickle compilation process?

Before any test actually executes, Cucumber transforms the raw, human-authored Gherkin in a feature file into a flat set of independently executable Pickles, resolving everything that requires cross-referencing other parts of the file.

The compiler first parses the feature file into an abstract syntax tree, then walks through each Scenario (or each Examples row of a Scenario Outline), merging in the Background's steps ahead of the scenario's own steps, and substituting any placeholders with that row's concrete values where applicable. Tags are also resolved and attached to each resulting Pickle at this stage, inherited from the Feature, the Scenario/Scenario Outline, and, for outline-derived pickles, the specific Examples block they came from.

The output is a list of Pickles with no remaining Gherkin-specific concepts, no Background, no placeholders, just a flat, ordered list of concrete step texts plus resolved tags, which is what lets the execution engine (and cross-language tooling built on the same Pickle format) treat every test case uniformly regardless of how much Gherkin structure originally produced it.