Skip to content

Latest commit

 

History

History
493 lines (377 loc) · 11.5 KB

reizql.md

File metadata and controls

493 lines (377 loc) · 11.5 KB
Raw

ReizQL Language

ReizQL is a declarative query language for building AST matchers that work on the Reiz platform.

:::{hint} Here is an example ReizQL query that searches for an if statement where the body consists from a single assignment statement that assigns the result of requests.get(...) call's result into a variable named response

if cache.invalidated:
    response = requests.get('https://api.reiz.io/refresh')

:::

If(
    body = [
        Assign(
            targets = [
                Name('response')
            ],
            value = Call(
                Attribute(
                    Name('requests'),
                    'get'
                )
            )
        )
    ]
)

Full Grammar

start                   ::= match_pattern

pattern                 ::= negate_pattern
                         | or_pattern
                         | and_pattern
                         | match_pattern
                         | sequential_pattern
                         | reference_pattern
                         | match_string_pattern
                         | atom_pattern

negate_pattern          ::= "not" pattern
or_pattern              ::= pattern "|" pattern
and_pattern             ::= pattern "&" pattern
match_pattern           ::= NAME "(" ",".argument+ ")"
sequential_pattern      ::= "[" ",".(pattern | "*" IGNORE)+ "]"
reference_pattern       ::= "~" NAME
atom_pattern            ::= NONE
                         | STRING
                         | NUMBER
                         | IGNORE
                         | "f" STRING

argument                ::= pattern
                         | NAME "=" pattern

NONE                    ::= "None"
IGNORE                  ::= "..."
NAME                    ::= "a".."Z"
NUMBER                  ::= INTEGER | FLOAT

Match Patterns

match_pattern           ::= NAME "(" ",".argument+ ")"

Match patterns are the most fundamental part of the query expression. They consist from an identifier (matcher name) which corresponds to an AST node type, additionally they take any number of fields to be matched (values, optionally attached with the corresponding field names).

All node types and fields are described in the Abstract Grammar of Python. Here are some entries from the ASDL;

module Python
{
    ...
    stmt = FunctionDef(identifier name, arguments args,
                       stmt* body, expr* decorator_list, expr? returns,
                       string? type_comment)
          | While(expr test, stmt* body, stmt* orelse)
          | If(expr test, stmt* body, stmt* orelse)
          | With(withitem* items, stmt* body, string? type_comment)

    expr = BoolOp(boolop op, expr* values)
         | NamedExpr(expr target, expr value)
         | BinOp(expr left, operator op, expr right)
         | UnaryOp(unaryop op, expr operand)
         | Lambda(arguments args, expr body)
         | IfExp(expr test, expr body, expr orelse)
         | Dict(expr* keys, expr* values)

The left hand side is the name of the base type, stmt would be a matcher that could match all of the types in its right hand side (e.g stmt() would match FunctionDef() / While() / If() / With()). Each element on the right hand side are concrete matchers for that element in syntax. For example a BinOp() represents a binary operation (2 operands), like 2 + 2 or a % b().

Each element on the right hand side have different fields with types attached to them. So the BinOp() node has 3 fields: left, op, right (respectively they mean left hand side, operator, right hand side of an arithmetic operation). left and the right must be another matcher from the expr base type (BoolOp / NamedExpr, ...). The star (*) at the end of type declaration implies that it requires a sequential pattern where the member types inherit from that base type (e.g stmt* might be something like [If(), If(), While()]). The question mark (?) indicates the value is optional and can be None.

If the values are not named (e.g BinOp(Constant())) then the name will be positionally given (BinOp(Constant(), Add()) will be transformed to BinOp(left=Constant(), op=Add()).

Example Queries

  • Match the 1994 literal
Constant(1994)
  • Match a binary operation where both sides are literals
BinOp(left=Constant(), right=Constant())
  • Match an (ternary) if expression that checks a.b's truthness
IfExp(
    test = Attribute(
        Name('a'),
        attr = 'b'
    )
)

Sequential Patterns

sequential_pattern      ::= "[" ",".(pattern | "*" IGNORE)+ "]"

Sequential patterns represent a list of subpatterns that are combined together to match a list on the host AST. If we want to search a function definition where there are 2 statements, the first one being an if statement and the second one is a return of an identifier named status then we simply describe this query like this;

FunctionDef(
    body = [
        If(),
        Return(
            Name('status')
        )
    ]
)

Sequential patterns are ordered, and matched one-to-one unless a ignore star is seen.

Ignore Star

If any of the elements on the sequence pattern is a star (*) followed by an ignore then the matchers before the ignore-star are relative from the beginning and the matchers after the ignore-star are relative to the end of the sequence. This implies that there is no maximum limit of items (in contrast to normal sequential patterns, where the number of elements is always fixed to amount of patterns seen) and the minimum being the total amount of matchers (excluding the ignore star).

Let's say we want to find a function that starts with an if statement, and then ends with a call to fetch function.

FunctionDef(
    body = [
        If(),
        *...,
        Return(
            Call(
                Name(
                    'fetch'
                )
            )
        )
    ]
)

There might be any number of elements between the if statement and the return, and it simply won't care.

:::{note} If you need a filler value (for example you want the minimum number of statements to be 3 instead of 2 in the case above) you can use ignore atom.

FunctionDef(
    body = [
        If(),
        ...,
        *...,
        Return(
            Call(
                Name(
                    'fetch'
                )
            )
        )
    ]
)

:::

Example Queries

  • Match all functions that have 2 statements and the last being a return
FunctionDef(
    body = [
        ...,
        Return()
    ]
)
  • Match all try/except's where the last except handler is a bare except: ...
Try(
    handlers = [
        *...,
        ExceptHandler(
            type = None
        )
    ]
)

Logical Patterns

Logical patterns are different patterns connected together in the sense of some logical operation (either AND or OR)

AND patterns

and_pattern             ::= pattern "&" pattern

AND patterns chains 2 different pattern together and matches the host value if it can be matched by both of the connected patterns.

OR patterns

or_pattern              ::= pattern "|" pattern

OR patterns chains 2 different pattern together and matches the host value if it can be matched by either of the connected patterns.

Example Queries

  • Match a return statement that either returns a list literal or a tuple literal
Return(List() | Tuple())
  • Match an if statement where the first statement is an assign and the total number of statements lower/equal than 5
If(
    body = [
        Assign(),
        *...
    ] & LEN(max=5)
)

NOT Patterns

negate_pattern          ::= "not" pattern

For checking whether a certain pattern does not match on the host AST, the negation operator can be used.

:::{hint} If a value is described as an optional (?) on the ASDL, then the existence of value can be denoted via not None pattern.

Example Queries

  • Match a return statement that doesn't return a call
Return(not Call())
  • A list that doesn't start with tuples or sets
List(
    elts = [
        (not Tuple()) & (not List()),
        *...
    ]
)

Reference Patterns

reference_pattern       ::= "~" NAME

Reference patterns are query-bound variables that can be referred elsewhere and the truthness determined by checking whether all the references point to the same expression (structurally, not semantically) or not.

Example Queries

  • Match a function definition where the last statement calls another function with the same name
FunctionDef(
    ~name,
    body = [
        *...,
        Expr(Call(Name(~name)))
    ]
)
  • Match an if statement where the test is a compare operation with the same lhs/rhs (a == a / b() is b())
If(
    test = CompareOp(
        left=~comp_expr,
        comparators = [
            ~comp_expr
        ]
    )
)

Atom Patterns

atom_pattern            ::= NONE
                         | STRING
                         | NUMBER
                         | IGNORE
                         | "f" STRING

Atoms represents basic values (like integers, strings) and also some ReizQL-flavored constructs.

Ignore

IGNORE                  ::= "..."

Ignore is a construct that just omits matching that field/element (in contrary to None, where it means that value does not exist).

None

NONE                    ::= "None"

None represents the absence of the value

Match String

MATCH_STRING            ::= "f" STRING

| Pattern | Interpretation | |----------------|------------------------------------| | % | matches zero or more characters | | _ | matches exactly one character | | \%/\_ | matches the literal %/_ |

Match strings can match alike strings via denoting some basic structures (like starts/ends with some static text).

Example Queries

  • Match a string that starts with http:// or https://
Constant(f'http://%' & f'https://%')
  • Match an arg that doesn't have any type annotations
arg(annotation = None)

Builtin Matchers

There are a couple of builtin matchers (builtin functions) that can match against certain conditions.

ALL/ANY

Signature: ALL($0: pattern) / ANY($0: pattern)

Apply the given matcher ($0) to a sequence. ALL would check whether all elements can be matched through the given argument, and any would check if any of the elements would be matched.

LEN

Signature: LEN($0: Opt[INTEGER], $1: Opt[INTEGER])

Checks whether the length of the sequence fits into $0 <= <host AST> <= $1. The $0/$1 are optional values but at least one of them should be specified.

META

Signature: META(**metadata_providers)

Checks for various metadata information (like file names, project names, parent types, etc).

I

Signature: I($0: atom_pattern[MATCH_STRING])

Supports case insensitive match through match strings.

Example Queries

  • Match a tuple where all members are literals
Tuple(ALL(Constant()))
  • Match a function where one of the top level statements is an if statement
FunctionDef(
    body = ANY(
        If()
    )
)
  • Match a function call where there are minimum 3 positional arguments and 5 maximum keyword arguments
Call(
    args = LEN(min=3),
    keywords = LEN(max=5)
)
  • Match a string in an case insensitive way
Constant(I(f"foo"))