The abstract syntax tree Alembic.Parser produces and Alembic.Evaluator walks. Every node is a tagged tuple, chosen over structs so pattern matching in the evaluator stays exhaustive — adding a node type without handling it produces a compiler warning, not a silent no-op.

This is the AST, not the CST (concrete syntax tree): it discards punctuation ({{, }}, %}, endif, dash whitespace markers) and keeps only semantically meaningful structure. See docs/grammar.md for a worked example showing both representations of the same template.

Node catalog

Worked example

{{ user.name | upcase }} maps to:

{:output, ["user", "name"], [{:filter, "upcase", []}]}

A non-trivial template:

{% extends "base.html" %}
{% block content %}
  Hello, {{ user.name | upcase }}!
  {% if user.admin %}(admin){% endif %}
{% endblock %}

parses to:

[
  {:extends, "base.html"},
  {:block, "content", [
    {:text, "\n  Hello, "},
    {:output, ["user", "name"], [{:filter, "upcase", []}]},
    {:text, "!\n  "},
    {:if, {:variable, ["user", "admin"]}, [{:text, "(admin)"}], [], nil},
    {:text, "\n"}
  ]}
]

Expression nodes (expr())

Output tags (output_node()) require a bare variable-path base, optionally wrapped in a filter chain — a literal base like {{ "hi" | upcase }} is rejected by Alembic.Parser with {:unsupported_output_expression, _} (a constraint fixed by this module's shape since Milestone 1.1, before the parser existed).