Architecture

At a high level, the behaviour of SQLFluff is divided into a few key stages. Whether calling sqlfluff lint, sqlfluff fix or sqlfluff parse, the internal flow is largely the same.

Stage 1, the templater

This stage only applies to templated SQL. Vanilla SQL is sent straight to stage 2, the lexer.

In order to lint templated SQL, SQLFluff must first convert the ‘raw’ or pre-templated code into valid SQL, which can then be parsed. The templater returns both the raw and post-templated SQL so that any rule violations which occur in templated sections can be ignored and the rest mapped to their original line location for user feedback.

SQLFluff supports multiple templating engines:

• Jinja

• SQL placeholders (e.g. SQLAlchemy parameters)

• Python format strings

• dbt (via plugin)

Under the hood dbt also uses Jinja, but in SQLFluff uses a separate mechanism which interfaces directly with the dbt python package.

For more details on how to configure the templater see Templating Configuration.

Stage 2, the lexer

The lexer takes SQL and separates it into segments of whitespace and code. Where we can impart some high level meaning to segments, we do, but the result of this operation is still a flat sequence of typed segments (all subclasses of RawSegment).

Stage 3, the parser

The parser is arguably the most complicated element of SQLFluff, and is relied on by all the other elements of the tool to do most of the heavy lifting.

1. The lexed segments are parsed using the specified dialect’s grammars. In SQLFluff, grammars describe the shape of SQL statements (or their components). The parser attempts to apply each potential grammar to the lexed segments until all the segments have been matched.

2. In SQLFluff, segments form a tree-like structure. The top-level segment is a FileSegment, which contains zero or more StatementSegments, and so on. Before the segments have been parsed and named according to their type, they are ‘raw’, meaning they have no classification other than their literal value.

3. A segment’s .match() method uses the match_grammar, on which .match() is called. SQLFluff parses in a single pass through the file, so segments will recursively match the file based on their respective grammars. In the example of a FileSegment, it first divides up the query into statements, and then the .match() method of those segments works out the structure within them.

   • Segments must implement a match_grammar. When .match()

     is called on a segment, this is the grammar which is used to decide whether there is a match.

   • Grammars combine segments or other grammars together in a

     pre-defined way. For example the OneOf grammar will match if any one of its child elements match.

   1. During the recursion, the parser eventually reaches segments which have no children (raw segments containing a single token), and so the recursion naturally finishes.

4. If no match is found for a segment, the contents will be wrapped in an UnparsableSegment which is picked up as a parsing error later. This is usually facilitated by the ParseMode on some grammars which can be set to GREEDY, allowing the grammar to capture additional segments as unparsable. As an example, bracketed sections are often configured to capture anything unexpected as unparsable rather than simply failing to match if there is more than expected (which would be the default, STRICT, behaviour).

5. The result of the .match() method is a MatchResult which contains the instructions on how to turn the flat sequence of raw segments into a nested tree of segments. Calling .apply() on this result at the end of the matching process is what finally creates the nested structure.

When working on the parser there are a couple of design principles to keep in mind.

• Grammars are contained in dialects, the root dialect being the ansi dialect. The ansi dialect is used to host logic common to all dialects, and so does not necessarily adhere to the formal ansi specification. Other SQL dialects inherit from the ansi dialect, replacing or patching any segments they need to. One reason for the Ref grammar is that it allows name resolution of grammar elements at runtime and so a patched grammar with some elements overridden can still rely on lower-level elements which haven’t been redeclared within the dialect

• All grammars and segments attempt to match as much as they can and will return partial matches where possible. It is up to the calling grammar or segment to decide whether a partial or complete match is required based on the context it is matching in.

Stage 4, the linter

Given the complete parse tree, rule classes check for linting errors by traversing the tree, looking for segments and patterns of concern. If the rule discovers a violation, it returns a LintResult pointing to the segment which caused the violation.

Some rules are able to fix the problems they find. If this is the case, the rule will return a list of fixes, which describe changes to be made to the tree. This can include edits, inserts, or deletions. Once the fixes have been applied, the updated tree is written to the original file.