Add a static analyzer for the filter language
This commit adds a class AFPSyntaxChecker which can statically analyze
a filter code to detect the following errors:
- unbound variables (which comes in two modes: conservative and liberal,
default to conservative)
- unused variables (disabled by default for compatibilty)
- assignment on built-in identifiers
- function application's arity mismatch
- function application's invalid function name
- non-string literal in the first argument of set / set_var
The existing parser and evaluator are modified as follows:
- The new (caching) evaluator no longer needs to perform variable
hoisting at runtime.
- Note that for array assignment, this changes the semantics.
- The new parser is more lenient, reducing parsing errors.
The static analyzer will catch these errors instead, allowing us
to give a much better error message and reduces the complexity of
the parser.
* The parser now allows function name to be any identifier.
* The parser now allows arity mismatch to occur.
* The parser now allows the first argument of set to be any expression.
Concretely, obvious changes that users will see are:
1. a := [1]; false & (a[] := 2); a[0] === 1
would evaluate to true, while it used to evaluate to the undefined value
due to hoisting
2. f(1)
will now error with 'f is not a valid function' as opposed to
'Unexpected "T_BRACE"'
3. length
will now error with 'Illegal use of built-in identifier "length"'
as opposed to 'Expected a ('
Appendix: conservative and liberal mode
The conservative mode is completely compatible with the current evaluator.
That is,
false & (a := 1); a
will not deem `a` as unbound, though this is actually undesirable because
`a` would then be bound to the troublesome undefined value.
The liberal mode rejects the above pattern by deeming `a` as unbound.
However, it also rejects
true & (a := 1); a
even though (a := 1) is always executed. Since there are several filters
in Wikimedia projects that rely on this behavior, we default the mode
to conservative for now.
Note that even the liberal mode doesn't really respect lexical scope
appeared in some other programming languages (see also T234690).
For instance:
(if true then (a := 1) else (a := 2) end); a
would be accepted by the liberal checker, even though under lexical scope,
`a` would be unbound. However, it is unlikely that lexical scope
will be suitable for the filter language, as most filters in
Wikimedia projects that have user-defined variable do violate lexical scope.
Bug: T260903
Bug: T238709
Bug: T237610
Bug: T234690
Bug: T231536
Change-Id: Ic6d030503e554933f8d220c6f87b680505918ae2
2019-11-20 07:09:41 +00:00
|
|
|
<?php
|
|
|
|
|
|
|
|
namespace MediaWiki\Extension\AbuseFilter\Parser;
|
|
|
|
|
2020-10-03 13:01:53 +00:00
|
|
|
use InvalidArgumentException;
|
Add a static analyzer for the filter language
This commit adds a class AFPSyntaxChecker which can statically analyze
a filter code to detect the following errors:
- unbound variables (which comes in two modes: conservative and liberal,
default to conservative)
- unused variables (disabled by default for compatibilty)
- assignment on built-in identifiers
- function application's arity mismatch
- function application's invalid function name
- non-string literal in the first argument of set / set_var
The existing parser and evaluator are modified as follows:
- The new (caching) evaluator no longer needs to perform variable
hoisting at runtime.
- Note that for array assignment, this changes the semantics.
- The new parser is more lenient, reducing parsing errors.
The static analyzer will catch these errors instead, allowing us
to give a much better error message and reduces the complexity of
the parser.
* The parser now allows function name to be any identifier.
* The parser now allows arity mismatch to occur.
* The parser now allows the first argument of set to be any expression.
Concretely, obvious changes that users will see are:
1. a := [1]; false & (a[] := 2); a[0] === 1
would evaluate to true, while it used to evaluate to the undefined value
due to hoisting
2. f(1)
will now error with 'f is not a valid function' as opposed to
'Unexpected "T_BRACE"'
3. length
will now error with 'Illegal use of built-in identifier "length"'
as opposed to 'Expected a ('
Appendix: conservative and liberal mode
The conservative mode is completely compatible with the current evaluator.
That is,
false & (a := 1); a
will not deem `a` as unbound, though this is actually undesirable because
`a` would then be bound to the troublesome undefined value.
The liberal mode rejects the above pattern by deeming `a` as unbound.
However, it also rejects
true & (a := 1); a
even though (a := 1) is always executed. Since there are several filters
in Wikimedia projects that rely on this behavior, we default the mode
to conservative for now.
Note that even the liberal mode doesn't really respect lexical scope
appeared in some other programming languages (see also T234690).
For instance:
(if true then (a := 1) else (a := 2) end); a
would be accepted by the liberal checker, even though under lexical scope,
`a` would be unbound. However, it is unlikely that lexical scope
will be suitable for the filter language, as most filters in
Wikimedia projects that have user-defined variable do violate lexical scope.
Bug: T260903
Bug: T238709
Bug: T237610
Bug: T234690
Bug: T231536
Change-Id: Ic6d030503e554933f8d220c6f87b680505918ae2
2019-11-20 07:09:41 +00:00
|
|
|
use LogicException;
|
|
|
|
use MediaWiki\Extension\AbuseFilter\KeywordsManager;
|
|
|
|
use MediaWiki\Extension\AbuseFilter\Parser\Exception\InternalException;
|
|
|
|
use MediaWiki\Extension\AbuseFilter\Parser\Exception\UserVisibleException;
|
2024-06-12 18:01:35 +00:00
|
|
|
use MediaWiki\Message\Message;
|
Add a static analyzer for the filter language
This commit adds a class AFPSyntaxChecker which can statically analyze
a filter code to detect the following errors:
- unbound variables (which comes in two modes: conservative and liberal,
default to conservative)
- unused variables (disabled by default for compatibilty)
- assignment on built-in identifiers
- function application's arity mismatch
- function application's invalid function name
- non-string literal in the first argument of set / set_var
The existing parser and evaluator are modified as follows:
- The new (caching) evaluator no longer needs to perform variable
hoisting at runtime.
- Note that for array assignment, this changes the semantics.
- The new parser is more lenient, reducing parsing errors.
The static analyzer will catch these errors instead, allowing us
to give a much better error message and reduces the complexity of
the parser.
* The parser now allows function name to be any identifier.
* The parser now allows arity mismatch to occur.
* The parser now allows the first argument of set to be any expression.
Concretely, obvious changes that users will see are:
1. a := [1]; false & (a[] := 2); a[0] === 1
would evaluate to true, while it used to evaluate to the undefined value
due to hoisting
2. f(1)
will now error with 'f is not a valid function' as opposed to
'Unexpected "T_BRACE"'
3. length
will now error with 'Illegal use of built-in identifier "length"'
as opposed to 'Expected a ('
Appendix: conservative and liberal mode
The conservative mode is completely compatible with the current evaluator.
That is,
false & (a := 1); a
will not deem `a` as unbound, though this is actually undesirable because
`a` would then be bound to the troublesome undefined value.
The liberal mode rejects the above pattern by deeming `a` as unbound.
However, it also rejects
true & (a := 1); a
even though (a := 1) is always executed. Since there are several filters
in Wikimedia projects that rely on this behavior, we default the mode
to conservative for now.
Note that even the liberal mode doesn't really respect lexical scope
appeared in some other programming languages (see also T234690).
For instance:
(if true then (a := 1) else (a := 2) end); a
would be accepted by the liberal checker, even though under lexical scope,
`a` would be unbound. However, it is unlikely that lexical scope
will be suitable for the filter language, as most filters in
Wikimedia projects that have user-defined variable do violate lexical scope.
Bug: T260903
Bug: T238709
Bug: T237610
Bug: T234690
Bug: T231536
Change-Id: Ic6d030503e554933f8d220c6f87b680505918ae2
2019-11-20 07:09:41 +00:00
|
|
|
|
|
|
|
/**
|
|
|
|
* SyntaxChecker statically analyzes the code without actually running it.
|
|
|
|
* Currently, it only checks for
|
|
|
|
*
|
|
|
|
* - unbound variables
|
|
|
|
* - unused variables: note that a := 1; a := 1; a
|
|
|
|
* is considered OK even though the first `a` seems unused
|
|
|
|
* because the pattern "a := null; if ... then (a := ...) end; ..."
|
|
|
|
* should not count first `a` as unused.
|
|
|
|
* - assignment to built-in identifiers
|
|
|
|
* - invalid function call (arity mismatch, non-valid function)
|
|
|
|
* - first-order information of `set_var` and `set`
|
|
|
|
*
|
|
|
|
* Because it doesn't cover all checks that the current Check Syntax does,
|
|
|
|
* it is currently complementary to the current Check Syntax.
|
|
|
|
* In the future, it could subsume the current Check Syntax, and could be
|
|
|
|
* extended to perform type checking or type inference.
|
|
|
|
*/
|
2020-10-03 13:01:53 +00:00
|
|
|
class SyntaxChecker {
|
Add a static analyzer for the filter language
This commit adds a class AFPSyntaxChecker which can statically analyze
a filter code to detect the following errors:
- unbound variables (which comes in two modes: conservative and liberal,
default to conservative)
- unused variables (disabled by default for compatibilty)
- assignment on built-in identifiers
- function application's arity mismatch
- function application's invalid function name
- non-string literal in the first argument of set / set_var
The existing parser and evaluator are modified as follows:
- The new (caching) evaluator no longer needs to perform variable
hoisting at runtime.
- Note that for array assignment, this changes the semantics.
- The new parser is more lenient, reducing parsing errors.
The static analyzer will catch these errors instead, allowing us
to give a much better error message and reduces the complexity of
the parser.
* The parser now allows function name to be any identifier.
* The parser now allows arity mismatch to occur.
* The parser now allows the first argument of set to be any expression.
Concretely, obvious changes that users will see are:
1. a := [1]; false & (a[] := 2); a[0] === 1
would evaluate to true, while it used to evaluate to the undefined value
due to hoisting
2. f(1)
will now error with 'f is not a valid function' as opposed to
'Unexpected "T_BRACE"'
3. length
will now error with 'Illegal use of built-in identifier "length"'
as opposed to 'Expected a ('
Appendix: conservative and liberal mode
The conservative mode is completely compatible with the current evaluator.
That is,
false & (a := 1); a
will not deem `a` as unbound, though this is actually undesirable because
`a` would then be bound to the troublesome undefined value.
The liberal mode rejects the above pattern by deeming `a` as unbound.
However, it also rejects
true & (a := 1); a
even though (a := 1) is always executed. Since there are several filters
in Wikimedia projects that rely on this behavior, we default the mode
to conservative for now.
Note that even the liberal mode doesn't really respect lexical scope
appeared in some other programming languages (see also T234690).
For instance:
(if true then (a := 1) else (a := 2) end); a
would be accepted by the liberal checker, even though under lexical scope,
`a` would be unbound. However, it is unlikely that lexical scope
will be suitable for the filter language, as most filters in
Wikimedia projects that have user-defined variable do violate lexical scope.
Bug: T260903
Bug: T238709
Bug: T237610
Bug: T234690
Bug: T231536
Change-Id: Ic6d030503e554933f8d220c6f87b680505918ae2
2019-11-20 07:09:41 +00:00
|
|
|
/**
|
|
|
|
* @var AFPTreeNode|null Root of the AST to check
|
|
|
|
*/
|
|
|
|
private $treeRoot;
|
|
|
|
|
|
|
|
/** @var KeywordsManager */
|
2023-06-23 10:28:06 +00:00
|
|
|
private $keywordsManager;
|
Add a static analyzer for the filter language
This commit adds a class AFPSyntaxChecker which can statically analyze
a filter code to detect the following errors:
- unbound variables (which comes in two modes: conservative and liberal,
default to conservative)
- unused variables (disabled by default for compatibilty)
- assignment on built-in identifiers
- function application's arity mismatch
- function application's invalid function name
- non-string literal in the first argument of set / set_var
The existing parser and evaluator are modified as follows:
- The new (caching) evaluator no longer needs to perform variable
hoisting at runtime.
- Note that for array assignment, this changes the semantics.
- The new parser is more lenient, reducing parsing errors.
The static analyzer will catch these errors instead, allowing us
to give a much better error message and reduces the complexity of
the parser.
* The parser now allows function name to be any identifier.
* The parser now allows arity mismatch to occur.
* The parser now allows the first argument of set to be any expression.
Concretely, obvious changes that users will see are:
1. a := [1]; false & (a[] := 2); a[0] === 1
would evaluate to true, while it used to evaluate to the undefined value
due to hoisting
2. f(1)
will now error with 'f is not a valid function' as opposed to
'Unexpected "T_BRACE"'
3. length
will now error with 'Illegal use of built-in identifier "length"'
as opposed to 'Expected a ('
Appendix: conservative and liberal mode
The conservative mode is completely compatible with the current evaluator.
That is,
false & (a := 1); a
will not deem `a` as unbound, though this is actually undesirable because
`a` would then be bound to the troublesome undefined value.
The liberal mode rejects the above pattern by deeming `a` as unbound.
However, it also rejects
true & (a := 1); a
even though (a := 1) is always executed. Since there are several filters
in Wikimedia projects that rely on this behavior, we default the mode
to conservative for now.
Note that even the liberal mode doesn't really respect lexical scope
appeared in some other programming languages (see also T234690).
For instance:
(if true then (a := 1) else (a := 2) end); a
would be accepted by the liberal checker, even though under lexical scope,
`a` would be unbound. However, it is unlikely that lexical scope
will be suitable for the filter language, as most filters in
Wikimedia projects that have user-defined variable do violate lexical scope.
Bug: T260903
Bug: T238709
Bug: T237610
Bug: T234690
Bug: T231536
Change-Id: Ic6d030503e554933f8d220c6f87b680505918ae2
2019-11-20 07:09:41 +00:00
|
|
|
|
|
|
|
public const MCONSERVATIVE = 'MODE_CONSERVATIVE';
|
|
|
|
public const MLIBERAL = 'MODE_LIBERAL';
|
|
|
|
public const DUMMYPOS = 0;
|
|
|
|
public const CACHE_VERSION = 1;
|
|
|
|
|
|
|
|
/**
|
|
|
|
* @var string The mode of checking. The value should be either
|
|
|
|
*
|
|
|
|
* - MLIBERAL: which guarantees that all user-defined variables
|
|
|
|
* will be bound, but incompatible with what the evaluator currently
|
|
|
|
* permits. E.g.,
|
|
|
|
*
|
|
|
|
* if true then (a := 1) else null end; a
|
|
|
|
*
|
|
|
|
* is rejected in this mode, even though `a` is in fact always bound.
|
|
|
|
*
|
|
|
|
* - MCONSERVATIVE which is compatible with what the evaluator
|
|
|
|
* currently permits, but could allow undefined variables to occur.
|
|
|
|
* E.g.,
|
|
|
|
*
|
|
|
|
* if false then (a := 1) else null end; a
|
|
|
|
*
|
|
|
|
* is accepted in this mode, even though `a` is in fact always unbound.
|
|
|
|
*/
|
|
|
|
private $mode;
|
|
|
|
|
|
|
|
/**
|
|
|
|
* @var bool Whether we want to check for unused variables
|
|
|
|
*/
|
|
|
|
private $checkUnusedVars;
|
|
|
|
|
|
|
|
/**
|
|
|
|
* @param AFPSyntaxTree $tree
|
|
|
|
* @param KeywordsManager $keywordsManager
|
|
|
|
* @param string $mode
|
|
|
|
* @param bool $checkUnusedVars
|
|
|
|
*/
|
|
|
|
public function __construct(
|
|
|
|
AFPSyntaxTree $tree,
|
|
|
|
KeywordsManager $keywordsManager,
|
|
|
|
string $mode = self::MCONSERVATIVE,
|
|
|
|
bool $checkUnusedVars = false
|
|
|
|
) {
|
|
|
|
$this->treeRoot = $tree->getRoot();
|
|
|
|
$this->keywordsManager = $keywordsManager;
|
|
|
|
$this->mode = $mode;
|
|
|
|
$this->checkUnusedVars = $checkUnusedVars;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Start the static analysis
|
|
|
|
*
|
|
|
|
* @throws UserVisibleException
|
|
|
|
*/
|
|
|
|
public function start(): void {
|
|
|
|
if ( !$this->treeRoot ) {
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
$bound = $this->check( $this->desugar( $this->treeRoot ), [] );
|
|
|
|
$unused = array_keys( array_filter( $bound, static function ( $v ) {
|
|
|
|
return !$v;
|
|
|
|
} ) );
|
|
|
|
if ( $this->checkUnusedVars && $unused ) {
|
|
|
|
throw new UserVisibleException(
|
|
|
|
'unusedvars',
|
|
|
|
self::DUMMYPOS,
|
|
|
|
[ Message::listParam( $unused, 'comma' ) ]
|
|
|
|
);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Remove syntactic sugar so that we don't need to deal with
|
|
|
|
* too many cases.
|
|
|
|
*
|
|
|
|
* This could benefit the evaluator as well, but for now, this is
|
|
|
|
* only used for static analysis.
|
|
|
|
*
|
|
|
|
* Postcondition:
|
|
|
|
* - The tree will not contain nodes of
|
|
|
|
* type ASSIGNMENT, LOGIC, COMPARE, SUM_REL, MUL_REL, POW,
|
|
|
|
* KEYWORD_OPERATOR, and ARRAY_INDEX
|
|
|
|
* - The tree may additionally contain a node of type BINOP.
|
|
|
|
* - The tree should not have set_var function application.
|
|
|
|
* - Conditionals will have both branches.
|
|
|
|
*
|
|
|
|
* @param AFPTreeNode $node
|
|
|
|
* @return AFPTreeNode
|
|
|
|
* @throws InternalException
|
|
|
|
*/
|
|
|
|
private function desugar( AFPTreeNode $node ): AFPTreeNode {
|
|
|
|
switch ( $node->type ) {
|
|
|
|
case AFPTreeNode::ATOM:
|
|
|
|
return $node;
|
|
|
|
|
|
|
|
case AFPTreeNode::FUNCTION_CALL:
|
|
|
|
if ( $node->children[0] === 'set_var' ) {
|
|
|
|
$node->children[0] = 'set';
|
|
|
|
}
|
|
|
|
return $this->newNodeMapExceptFirst( $node );
|
|
|
|
|
|
|
|
case AFPTreeNode::ARRAY_INDEX:
|
|
|
|
return $this->newNodeNamedBinop( $node, '[]' );
|
|
|
|
|
|
|
|
case AFPTreeNode::POW:
|
|
|
|
return $this->newNodeNamedBinop( $node, '**' );
|
|
|
|
|
|
|
|
case AFPTreeNode::UNARY:
|
2022-04-26 15:46:01 +00:00
|
|
|
case AFPTreeNode::INDEX_ASSIGNMENT:
|
|
|
|
case AFPTreeNode::ARRAY_APPEND:
|
Add a static analyzer for the filter language
This commit adds a class AFPSyntaxChecker which can statically analyze
a filter code to detect the following errors:
- unbound variables (which comes in two modes: conservative and liberal,
default to conservative)
- unused variables (disabled by default for compatibilty)
- assignment on built-in identifiers
- function application's arity mismatch
- function application's invalid function name
- non-string literal in the first argument of set / set_var
The existing parser and evaluator are modified as follows:
- The new (caching) evaluator no longer needs to perform variable
hoisting at runtime.
- Note that for array assignment, this changes the semantics.
- The new parser is more lenient, reducing parsing errors.
The static analyzer will catch these errors instead, allowing us
to give a much better error message and reduces the complexity of
the parser.
* The parser now allows function name to be any identifier.
* The parser now allows arity mismatch to occur.
* The parser now allows the first argument of set to be any expression.
Concretely, obvious changes that users will see are:
1. a := [1]; false & (a[] := 2); a[0] === 1
would evaluate to true, while it used to evaluate to the undefined value
due to hoisting
2. f(1)
will now error with 'f is not a valid function' as opposed to
'Unexpected "T_BRACE"'
3. length
will now error with 'Illegal use of built-in identifier "length"'
as opposed to 'Expected a ('
Appendix: conservative and liberal mode
The conservative mode is completely compatible with the current evaluator.
That is,
false & (a := 1); a
will not deem `a` as unbound, though this is actually undesirable because
`a` would then be bound to the troublesome undefined value.
The liberal mode rejects the above pattern by deeming `a` as unbound.
However, it also rejects
true & (a := 1); a
even though (a := 1) is always executed. Since there are several filters
in Wikimedia projects that rely on this behavior, we default the mode
to conservative for now.
Note that even the liberal mode doesn't really respect lexical scope
appeared in some other programming languages (see also T234690).
For instance:
(if true then (a := 1) else (a := 2) end); a
would be accepted by the liberal checker, even though under lexical scope,
`a` would be unbound. However, it is unlikely that lexical scope
will be suitable for the filter language, as most filters in
Wikimedia projects that have user-defined variable do violate lexical scope.
Bug: T260903
Bug: T238709
Bug: T237610
Bug: T234690
Bug: T231536
Change-Id: Ic6d030503e554933f8d220c6f87b680505918ae2
2019-11-20 07:09:41 +00:00
|
|
|
return $this->newNodeMapExceptFirst( $node );
|
|
|
|
|
|
|
|
case AFPTreeNode::BOOL_INVERT:
|
|
|
|
/*
|
|
|
|
* @todo this should really be combined with UNARY,
|
|
|
|
* but let's wait to change the meaning of UNARY across
|
|
|
|
* the codebase together
|
|
|
|
*/
|
|
|
|
return $this->newNodeMapAll( $node );
|
|
|
|
|
|
|
|
case AFPTreeNode::KEYWORD_OPERATOR:
|
|
|
|
case AFPTreeNode::MUL_REL:
|
|
|
|
case AFPTreeNode::SUM_REL:
|
|
|
|
case AFPTreeNode::COMPARE:
|
|
|
|
return $this->newNodeBinop( $node );
|
|
|
|
|
|
|
|
case AFPTreeNode::LOGIC:
|
|
|
|
$result = $this->newNodeBinop( $node );
|
2024-03-16 18:52:48 +00:00
|
|
|
[ $op, $left, $right ] = $result->children;
|
Add a static analyzer for the filter language
This commit adds a class AFPSyntaxChecker which can statically analyze
a filter code to detect the following errors:
- unbound variables (which comes in two modes: conservative and liberal,
default to conservative)
- unused variables (disabled by default for compatibilty)
- assignment on built-in identifiers
- function application's arity mismatch
- function application's invalid function name
- non-string literal in the first argument of set / set_var
The existing parser and evaluator are modified as follows:
- The new (caching) evaluator no longer needs to perform variable
hoisting at runtime.
- Note that for array assignment, this changes the semantics.
- The new parser is more lenient, reducing parsing errors.
The static analyzer will catch these errors instead, allowing us
to give a much better error message and reduces the complexity of
the parser.
* The parser now allows function name to be any identifier.
* The parser now allows arity mismatch to occur.
* The parser now allows the first argument of set to be any expression.
Concretely, obvious changes that users will see are:
1. a := [1]; false & (a[] := 2); a[0] === 1
would evaluate to true, while it used to evaluate to the undefined value
due to hoisting
2. f(1)
will now error with 'f is not a valid function' as opposed to
'Unexpected "T_BRACE"'
3. length
will now error with 'Illegal use of built-in identifier "length"'
as opposed to 'Expected a ('
Appendix: conservative and liberal mode
The conservative mode is completely compatible with the current evaluator.
That is,
false & (a := 1); a
will not deem `a` as unbound, though this is actually undesirable because
`a` would then be bound to the troublesome undefined value.
The liberal mode rejects the above pattern by deeming `a` as unbound.
However, it also rejects
true & (a := 1); a
even though (a := 1) is always executed. Since there are several filters
in Wikimedia projects that rely on this behavior, we default the mode
to conservative for now.
Note that even the liberal mode doesn't really respect lexical scope
appeared in some other programming languages (see also T234690).
For instance:
(if true then (a := 1) else (a := 2) end); a
would be accepted by the liberal checker, even though under lexical scope,
`a` would be unbound. However, it is unlikely that lexical scope
will be suitable for the filter language, as most filters in
Wikimedia projects that have user-defined variable do violate lexical scope.
Bug: T260903
Bug: T238709
Bug: T237610
Bug: T234690
Bug: T231536
Change-Id: Ic6d030503e554933f8d220c6f87b680505918ae2
2019-11-20 07:09:41 +00:00
|
|
|
if ( $op === '&' || $op === '|' ) {
|
|
|
|
return $this->desugarAndOr( $op, $left, $right, $node->position );
|
|
|
|
} else {
|
|
|
|
return $result;
|
|
|
|
}
|
|
|
|
|
|
|
|
case AFPTreeNode::ARRAY_DEFINITION:
|
|
|
|
case AFPTreeNode::SEMICOLON:
|
|
|
|
return $this->newNodeMapAll( $node );
|
|
|
|
|
|
|
|
case AFPTreeNode::CONDITIONAL:
|
|
|
|
if ( $node->children[2] === null ) {
|
|
|
|
$node->children[2] = new AFPTreeNode(
|
|
|
|
AFPTreeNode::ATOM,
|
|
|
|
new AFPToken(
|
2022-04-26 15:42:34 +00:00
|
|
|
AFPToken::TKEYWORD,
|
Add a static analyzer for the filter language
This commit adds a class AFPSyntaxChecker which can statically analyze
a filter code to detect the following errors:
- unbound variables (which comes in two modes: conservative and liberal,
default to conservative)
- unused variables (disabled by default for compatibilty)
- assignment on built-in identifiers
- function application's arity mismatch
- function application's invalid function name
- non-string literal in the first argument of set / set_var
The existing parser and evaluator are modified as follows:
- The new (caching) evaluator no longer needs to perform variable
hoisting at runtime.
- Note that for array assignment, this changes the semantics.
- The new parser is more lenient, reducing parsing errors.
The static analyzer will catch these errors instead, allowing us
to give a much better error message and reduces the complexity of
the parser.
* The parser now allows function name to be any identifier.
* The parser now allows arity mismatch to occur.
* The parser now allows the first argument of set to be any expression.
Concretely, obvious changes that users will see are:
1. a := [1]; false & (a[] := 2); a[0] === 1
would evaluate to true, while it used to evaluate to the undefined value
due to hoisting
2. f(1)
will now error with 'f is not a valid function' as opposed to
'Unexpected "T_BRACE"'
3. length
will now error with 'Illegal use of built-in identifier "length"'
as opposed to 'Expected a ('
Appendix: conservative and liberal mode
The conservative mode is completely compatible with the current evaluator.
That is,
false & (a := 1); a
will not deem `a` as unbound, though this is actually undesirable because
`a` would then be bound to the troublesome undefined value.
The liberal mode rejects the above pattern by deeming `a` as unbound.
However, it also rejects
true & (a := 1); a
even though (a := 1) is always executed. Since there are several filters
in Wikimedia projects that rely on this behavior, we default the mode
to conservative for now.
Note that even the liberal mode doesn't really respect lexical scope
appeared in some other programming languages (see also T234690).
For instance:
(if true then (a := 1) else (a := 2) end); a
would be accepted by the liberal checker, even though under lexical scope,
`a` would be unbound. However, it is unlikely that lexical scope
will be suitable for the filter language, as most filters in
Wikimedia projects that have user-defined variable do violate lexical scope.
Bug: T260903
Bug: T238709
Bug: T237610
Bug: T234690
Bug: T231536
Change-Id: Ic6d030503e554933f8d220c6f87b680505918ae2
2019-11-20 07:09:41 +00:00
|
|
|
"null",
|
|
|
|
$node->position
|
|
|
|
),
|
|
|
|
$node->position
|
|
|
|
);
|
|
|
|
}
|
|
|
|
return $this->newNodeMapAll( $node );
|
|
|
|
|
|
|
|
case AFPTreeNode::ASSIGNMENT:
|
2024-03-16 18:52:48 +00:00
|
|
|
[ $varname, $value ] = $node->children;
|
Add a static analyzer for the filter language
This commit adds a class AFPSyntaxChecker which can statically analyze
a filter code to detect the following errors:
- unbound variables (which comes in two modes: conservative and liberal,
default to conservative)
- unused variables (disabled by default for compatibilty)
- assignment on built-in identifiers
- function application's arity mismatch
- function application's invalid function name
- non-string literal in the first argument of set / set_var
The existing parser and evaluator are modified as follows:
- The new (caching) evaluator no longer needs to perform variable
hoisting at runtime.
- Note that for array assignment, this changes the semantics.
- The new parser is more lenient, reducing parsing errors.
The static analyzer will catch these errors instead, allowing us
to give a much better error message and reduces the complexity of
the parser.
* The parser now allows function name to be any identifier.
* The parser now allows arity mismatch to occur.
* The parser now allows the first argument of set to be any expression.
Concretely, obvious changes that users will see are:
1. a := [1]; false & (a[] := 2); a[0] === 1
would evaluate to true, while it used to evaluate to the undefined value
due to hoisting
2. f(1)
will now error with 'f is not a valid function' as opposed to
'Unexpected "T_BRACE"'
3. length
will now error with 'Illegal use of built-in identifier "length"'
as opposed to 'Expected a ('
Appendix: conservative and liberal mode
The conservative mode is completely compatible with the current evaluator.
That is,
false & (a := 1); a
will not deem `a` as unbound, though this is actually undesirable because
`a` would then be bound to the troublesome undefined value.
The liberal mode rejects the above pattern by deeming `a` as unbound.
However, it also rejects
true & (a := 1); a
even though (a := 1) is always executed. Since there are several filters
in Wikimedia projects that rely on this behavior, we default the mode
to conservative for now.
Note that even the liberal mode doesn't really respect lexical scope
appeared in some other programming languages (see also T234690).
For instance:
(if true then (a := 1) else (a := 2) end); a
would be accepted by the liberal checker, even though under lexical scope,
`a` would be unbound. However, it is unlikely that lexical scope
will be suitable for the filter language, as most filters in
Wikimedia projects that have user-defined variable do violate lexical scope.
Bug: T260903
Bug: T238709
Bug: T237610
Bug: T234690
Bug: T231536
Change-Id: Ic6d030503e554933f8d220c6f87b680505918ae2
2019-11-20 07:09:41 +00:00
|
|
|
|
|
|
|
return new AFPTreeNode(
|
|
|
|
AFPTreeNode::FUNCTION_CALL,
|
|
|
|
[
|
|
|
|
"set",
|
|
|
|
new AFPTreeNode(
|
|
|
|
AFPTreeNode::ATOM,
|
|
|
|
new AFPToken(
|
|
|
|
AFPToken::TSTRING,
|
|
|
|
$varname,
|
|
|
|
$node->position
|
|
|
|
),
|
|
|
|
$node->position
|
|
|
|
),
|
|
|
|
$this->desugar( $value )
|
|
|
|
],
|
|
|
|
$node->position
|
|
|
|
);
|
|
|
|
|
|
|
|
default:
|
|
|
|
// @codeCoverageIgnoreStart
|
|
|
|
throw new InternalException( "Unknown node type passed: {$node->type}" );
|
|
|
|
// @codeCoverageIgnoreEnd
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* @param string $op
|
|
|
|
* @param AFPTreeNode $left
|
|
|
|
* @param AFPTreeNode $right
|
|
|
|
* @param int $position
|
|
|
|
* @return AFPTreeNode
|
|
|
|
*/
|
|
|
|
private function desugarAndOr(
|
|
|
|
string $op,
|
|
|
|
AFPTreeNode $left,
|
|
|
|
AFPTreeNode $right,
|
|
|
|
int $position
|
|
|
|
): AFPTreeNode {
|
|
|
|
$trueNode = new AFPTreeNode(
|
|
|
|
AFPTreeNode::ATOM,
|
|
|
|
new AFPToken(
|
2022-04-26 15:42:34 +00:00
|
|
|
AFPToken::TKEYWORD,
|
Add a static analyzer for the filter language
This commit adds a class AFPSyntaxChecker which can statically analyze
a filter code to detect the following errors:
- unbound variables (which comes in two modes: conservative and liberal,
default to conservative)
- unused variables (disabled by default for compatibilty)
- assignment on built-in identifiers
- function application's arity mismatch
- function application's invalid function name
- non-string literal in the first argument of set / set_var
The existing parser and evaluator are modified as follows:
- The new (caching) evaluator no longer needs to perform variable
hoisting at runtime.
- Note that for array assignment, this changes the semantics.
- The new parser is more lenient, reducing parsing errors.
The static analyzer will catch these errors instead, allowing us
to give a much better error message and reduces the complexity of
the parser.
* The parser now allows function name to be any identifier.
* The parser now allows arity mismatch to occur.
* The parser now allows the first argument of set to be any expression.
Concretely, obvious changes that users will see are:
1. a := [1]; false & (a[] := 2); a[0] === 1
would evaluate to true, while it used to evaluate to the undefined value
due to hoisting
2. f(1)
will now error with 'f is not a valid function' as opposed to
'Unexpected "T_BRACE"'
3. length
will now error with 'Illegal use of built-in identifier "length"'
as opposed to 'Expected a ('
Appendix: conservative and liberal mode
The conservative mode is completely compatible with the current evaluator.
That is,
false & (a := 1); a
will not deem `a` as unbound, though this is actually undesirable because
`a` would then be bound to the troublesome undefined value.
The liberal mode rejects the above pattern by deeming `a` as unbound.
However, it also rejects
true & (a := 1); a
even though (a := 1) is always executed. Since there are several filters
in Wikimedia projects that rely on this behavior, we default the mode
to conservative for now.
Note that even the liberal mode doesn't really respect lexical scope
appeared in some other programming languages (see also T234690).
For instance:
(if true then (a := 1) else (a := 2) end); a
would be accepted by the liberal checker, even though under lexical scope,
`a` would be unbound. However, it is unlikely that lexical scope
will be suitable for the filter language, as most filters in
Wikimedia projects that have user-defined variable do violate lexical scope.
Bug: T260903
Bug: T238709
Bug: T237610
Bug: T234690
Bug: T231536
Change-Id: Ic6d030503e554933f8d220c6f87b680505918ae2
2019-11-20 07:09:41 +00:00
|
|
|
"true",
|
|
|
|
$position
|
|
|
|
),
|
|
|
|
$position
|
|
|
|
);
|
|
|
|
$falseNode = new AFPTreeNode(
|
|
|
|
AFPTreeNode::ATOM,
|
|
|
|
new AFPToken(
|
2022-04-26 15:42:34 +00:00
|
|
|
AFPToken::TKEYWORD,
|
Add a static analyzer for the filter language
This commit adds a class AFPSyntaxChecker which can statically analyze
a filter code to detect the following errors:
- unbound variables (which comes in two modes: conservative and liberal,
default to conservative)
- unused variables (disabled by default for compatibilty)
- assignment on built-in identifiers
- function application's arity mismatch
- function application's invalid function name
- non-string literal in the first argument of set / set_var
The existing parser and evaluator are modified as follows:
- The new (caching) evaluator no longer needs to perform variable
hoisting at runtime.
- Note that for array assignment, this changes the semantics.
- The new parser is more lenient, reducing parsing errors.
The static analyzer will catch these errors instead, allowing us
to give a much better error message and reduces the complexity of
the parser.
* The parser now allows function name to be any identifier.
* The parser now allows arity mismatch to occur.
* The parser now allows the first argument of set to be any expression.
Concretely, obvious changes that users will see are:
1. a := [1]; false & (a[] := 2); a[0] === 1
would evaluate to true, while it used to evaluate to the undefined value
due to hoisting
2. f(1)
will now error with 'f is not a valid function' as opposed to
'Unexpected "T_BRACE"'
3. length
will now error with 'Illegal use of built-in identifier "length"'
as opposed to 'Expected a ('
Appendix: conservative and liberal mode
The conservative mode is completely compatible with the current evaluator.
That is,
false & (a := 1); a
will not deem `a` as unbound, though this is actually undesirable because
`a` would then be bound to the troublesome undefined value.
The liberal mode rejects the above pattern by deeming `a` as unbound.
However, it also rejects
true & (a := 1); a
even though (a := 1) is always executed. Since there are several filters
in Wikimedia projects that rely on this behavior, we default the mode
to conservative for now.
Note that even the liberal mode doesn't really respect lexical scope
appeared in some other programming languages (see also T234690).
For instance:
(if true then (a := 1) else (a := 2) end); a
would be accepted by the liberal checker, even though under lexical scope,
`a` would be unbound. However, it is unlikely that lexical scope
will be suitable for the filter language, as most filters in
Wikimedia projects that have user-defined variable do violate lexical scope.
Bug: T260903
Bug: T238709
Bug: T237610
Bug: T234690
Bug: T231536
Change-Id: Ic6d030503e554933f8d220c6f87b680505918ae2
2019-11-20 07:09:41 +00:00
|
|
|
"false",
|
|
|
|
$position
|
|
|
|
),
|
|
|
|
$position
|
|
|
|
);
|
|
|
|
$conditionalNode = new AFPTreeNode(
|
|
|
|
AFPTreeNode::CONDITIONAL,
|
|
|
|
[
|
|
|
|
$right,
|
|
|
|
$trueNode,
|
|
|
|
$falseNode
|
|
|
|
],
|
|
|
|
$position
|
|
|
|
);
|
|
|
|
|
|
|
|
if ( $op === '&' ) {
|
|
|
|
// <a> & <b> is supposed to be equivalent to
|
|
|
|
// if <a> then (if <b> then true else false) else false end
|
|
|
|
// See T237336 for why this is currently not the case.
|
|
|
|
return new AFPTreeNode(
|
|
|
|
AFPTreeNode::CONDITIONAL,
|
|
|
|
[
|
|
|
|
$left,
|
|
|
|
$conditionalNode,
|
|
|
|
$falseNode
|
|
|
|
],
|
|
|
|
$position
|
|
|
|
);
|
|
|
|
} elseif ( $op === '|' ) {
|
|
|
|
// <a> | <b> is supposed to be equivalent to
|
|
|
|
// if <a> then true else (if <b> then true else false) end
|
|
|
|
// See T237336 for why this is currently not the case.
|
|
|
|
return new AFPTreeNode(
|
|
|
|
AFPTreeNode::CONDITIONAL,
|
|
|
|
[
|
|
|
|
$left,
|
|
|
|
$trueNode,
|
|
|
|
$conditionalNode
|
|
|
|
],
|
|
|
|
$position
|
|
|
|
);
|
|
|
|
} else {
|
|
|
|
// @codeCoverageIgnoreStart
|
|
|
|
throw new InternalException( "Unknown operator: {$op}" );
|
|
|
|
// @codeCoverageIgnoreEnd
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Construct a new node with information based on the old node but
|
|
|
|
* with different children
|
|
|
|
*
|
|
|
|
* @param AFPTreeNode $node
|
|
|
|
* @param AFPTreeNode[]|string[]|AFPToken $children
|
|
|
|
* @return AFPTreeNode
|
|
|
|
*/
|
|
|
|
private function newNode( AFPTreeNode $node, $children ): AFPTreeNode {
|
|
|
|
return new AFPTreeNode( $node->type, $children, $node->position );
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Construct a new node with information based on the old node but
|
|
|
|
* with different type
|
|
|
|
*
|
|
|
|
* @param AFPTreeNode $node
|
|
|
|
* @param string $type
|
|
|
|
* @return AFPTreeNode
|
|
|
|
*/
|
|
|
|
private function newNodeReplaceType(
|
|
|
|
AFPTreeNode $node,
|
|
|
|
string $type
|
|
|
|
): AFPTreeNode {
|
|
|
|
return new AFPTreeNode( $type, $node->children, $node->position );
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Recursively desugar on all children
|
|
|
|
*
|
|
|
|
* @param AFPTreeNode $node
|
|
|
|
* @return AFPTreeNode
|
|
|
|
*/
|
|
|
|
private function newNodeMapAll( AFPTreeNode $node ): AFPTreeNode {
|
|
|
|
$children = $node->children;
|
|
|
|
if ( !is_array( $children ) ) {
|
|
|
|
// @codeCoverageIgnoreStart
|
|
|
|
throw new LogicException(
|
|
|
|
"Unexpected non-array children of an AFPTreeNode of type " .
|
|
|
|
"{$node->type} at position {$node->position}"
|
|
|
|
);
|
|
|
|
// @codeCoverageIgnoreEnd
|
|
|
|
}
|
|
|
|
return $this->newNode( $node, array_map( [ $this, 'desugar' ], $children ) );
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Recursively desugar on all children except the first one
|
|
|
|
*
|
|
|
|
* @param AFPTreeNode $node
|
|
|
|
* @return AFPTreeNode
|
|
|
|
*/
|
|
|
|
private function newNodeMapExceptFirst( AFPTreeNode $node ): AFPTreeNode {
|
|
|
|
$items = [ $node->children[0] ];
|
|
|
|
$args = array_slice( $node->children, 1 );
|
|
|
|
foreach ( $args as $el ) {
|
|
|
|
$items[] = $this->desugar( $el );
|
|
|
|
}
|
|
|
|
return $this->newNode( $node, $items );
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Convert a node with an operation into a BINOP
|
|
|
|
*
|
|
|
|
* @param AFPTreeNode $node
|
|
|
|
* @return AFPTreeNode
|
|
|
|
*/
|
|
|
|
private function newNodeBinop( AFPTreeNode $node ): AFPTreeNode {
|
|
|
|
return $this->newNodeReplaceType(
|
|
|
|
$this->newNodeMapExceptFirst( $node ),
|
|
|
|
AFPTreeNode::BINOP
|
|
|
|
);
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Convert a node without an operation into a BINOP with the specified operation
|
|
|
|
*
|
|
|
|
* @param AFPTreeNode $node
|
|
|
|
* @param string $op
|
|
|
|
* @return AFPTreeNode
|
|
|
|
*/
|
|
|
|
private function newNodeNamedBinop(
|
|
|
|
AFPTreeNode $node,
|
|
|
|
string $op
|
|
|
|
): AFPTreeNode {
|
|
|
|
$items = $this->newNodeMapAll( $node )->children;
|
|
|
|
array_unshift( $items, $op );
|
|
|
|
return $this->newNodeReplaceType(
|
|
|
|
$this->newNode( $node, $items ),
|
|
|
|
AFPTreeNode::BINOP
|
|
|
|
);
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* - Statically compute what are bound after evaluating $node,
|
|
|
|
* provided that variables in $bound are already bound.
|
|
|
|
* - Similarly compute for each bound variable after evaluating $node
|
|
|
|
* whether it is used provided that we already have $bound
|
|
|
|
* that contains necessary information.
|
|
|
|
* - Ensure function application's validity.
|
|
|
|
* - Ensure that the first argument of set is a literal string.
|
|
|
|
* - Ensure that all assignment is not done on built-in identifier.
|
|
|
|
*
|
|
|
|
* Precondition:
|
|
|
|
* - The tree $node should be desugared and normalized.
|
|
|
|
*
|
|
|
|
* Postcondition:
|
|
|
|
* - $node is guaranteed to have no unbound variables
|
|
|
|
* provided that variables in $bound are already bound
|
|
|
|
* (for the definition of unbound variable indicated by $this->mode)
|
|
|
|
* - All function applications should be valid and have correct arity.
|
|
|
|
* - The set function application's first argument should be
|
|
|
|
* a literal string.
|
|
|
|
*
|
|
|
|
* @param AFPTreeNode $node
|
|
|
|
* @param bool[] $bound Map of [ variable_name => used ]
|
|
|
|
* @return bool[] Map of [ variable_name => used ]
|
|
|
|
* @throws UserVisibleException
|
|
|
|
* @throws InternalException
|
|
|
|
*/
|
|
|
|
private function check( AFPTreeNode $node, array $bound ): array {
|
|
|
|
switch ( $node->type ) {
|
|
|
|
// phpcs:ignore PSR2.ControlStructures.SwitchDeclaration.TerminatingComment
|
|
|
|
case AFPTreeNode::ATOM:
|
|
|
|
$tok = $node->children;
|
|
|
|
switch ( $tok->type ) {
|
|
|
|
case AFPToken::TID:
|
|
|
|
return $this->lookupVar(
|
|
|
|
$tok->value,
|
|
|
|
$tok->pos,
|
|
|
|
$bound
|
|
|
|
);
|
|
|
|
|
|
|
|
case AFPToken::TSTRING:
|
|
|
|
case AFPToken::TFLOAT:
|
|
|
|
case AFPToken::TINT:
|
|
|
|
case AFPToken::TKEYWORD:
|
|
|
|
return $bound;
|
|
|
|
|
|
|
|
default:
|
|
|
|
// @codeCoverageIgnoreStart
|
|
|
|
throw new InternalException( "Unknown token {$tok->type} provided in the ATOM node" );
|
|
|
|
// @codeCoverageIgnoreEnd
|
|
|
|
}
|
|
|
|
case AFPTreeNode::ARRAY_DEFINITION:
|
|
|
|
// @phan-suppress-next-line PhanTypeSuspiciousNonTraversableForeach children is array here
|
|
|
|
foreach ( $node->children as $el ) {
|
|
|
|
$bound = $this->check( $el, $bound );
|
|
|
|
}
|
|
|
|
return $bound;
|
|
|
|
|
|
|
|
case AFPTreeNode::FUNCTION_CALL:
|
|
|
|
$fname = $node->children[0];
|
|
|
|
$args = array_slice( $node->children, 1 );
|
2021-08-31 22:26:13 +00:00
|
|
|
if ( !array_key_exists( $fname, FilterEvaluator::FUNCTIONS ) ) {
|
Add a static analyzer for the filter language
This commit adds a class AFPSyntaxChecker which can statically analyze
a filter code to detect the following errors:
- unbound variables (which comes in two modes: conservative and liberal,
default to conservative)
- unused variables (disabled by default for compatibilty)
- assignment on built-in identifiers
- function application's arity mismatch
- function application's invalid function name
- non-string literal in the first argument of set / set_var
The existing parser and evaluator are modified as follows:
- The new (caching) evaluator no longer needs to perform variable
hoisting at runtime.
- Note that for array assignment, this changes the semantics.
- The new parser is more lenient, reducing parsing errors.
The static analyzer will catch these errors instead, allowing us
to give a much better error message and reduces the complexity of
the parser.
* The parser now allows function name to be any identifier.
* The parser now allows arity mismatch to occur.
* The parser now allows the first argument of set to be any expression.
Concretely, obvious changes that users will see are:
1. a := [1]; false & (a[] := 2); a[0] === 1
would evaluate to true, while it used to evaluate to the undefined value
due to hoisting
2. f(1)
will now error with 'f is not a valid function' as opposed to
'Unexpected "T_BRACE"'
3. length
will now error with 'Illegal use of built-in identifier "length"'
as opposed to 'Expected a ('
Appendix: conservative and liberal mode
The conservative mode is completely compatible with the current evaluator.
That is,
false & (a := 1); a
will not deem `a` as unbound, though this is actually undesirable because
`a` would then be bound to the troublesome undefined value.
The liberal mode rejects the above pattern by deeming `a` as unbound.
However, it also rejects
true & (a := 1); a
even though (a := 1) is always executed. Since there are several filters
in Wikimedia projects that rely on this behavior, we default the mode
to conservative for now.
Note that even the liberal mode doesn't really respect lexical scope
appeared in some other programming languages (see also T234690).
For instance:
(if true then (a := 1) else (a := 2) end); a
would be accepted by the liberal checker, even though under lexical scope,
`a` would be unbound. However, it is unlikely that lexical scope
will be suitable for the filter language, as most filters in
Wikimedia projects that have user-defined variable do violate lexical scope.
Bug: T260903
Bug: T238709
Bug: T237610
Bug: T234690
Bug: T231536
Change-Id: Ic6d030503e554933f8d220c6f87b680505918ae2
2019-11-20 07:09:41 +00:00
|
|
|
throw new UserVisibleException(
|
|
|
|
'unknownfunction',
|
|
|
|
$node->position,
|
|
|
|
[ $fname ]
|
|
|
|
);
|
|
|
|
}
|
|
|
|
$this->checkArgCount( $args, $fname, $node->position );
|
|
|
|
|
|
|
|
if ( $fname === 'set' ) {
|
|
|
|
// arity is checked, so we know $args[0] and $args[1] exist
|
|
|
|
$tok = $args[0]->children;
|
|
|
|
|
|
|
|
if (
|
|
|
|
!( $tok instanceof AFPToken ) ||
|
|
|
|
$tok->type !== AFPToken::TSTRING
|
|
|
|
) {
|
|
|
|
throw new UserVisibleException(
|
|
|
|
'variablevariable',
|
|
|
|
$node->position,
|
|
|
|
[]
|
|
|
|
);
|
|
|
|
}
|
|
|
|
|
|
|
|
$bound = $this->check( $args[1], $bound );
|
|
|
|
// set the variable as unused
|
|
|
|
return $this->assignVar(
|
|
|
|
$tok->value,
|
|
|
|
$tok->pos,
|
|
|
|
$bound
|
|
|
|
);
|
|
|
|
} else {
|
|
|
|
foreach ( $args as $arg ) {
|
|
|
|
$bound = $this->check( $arg, $bound );
|
|
|
|
}
|
|
|
|
return $bound;
|
|
|
|
}
|
|
|
|
|
|
|
|
case AFPTreeNode::BINOP:
|
2024-03-16 18:52:48 +00:00
|
|
|
[ , $left, $right ] = $node->children;
|
Add a static analyzer for the filter language
This commit adds a class AFPSyntaxChecker which can statically analyze
a filter code to detect the following errors:
- unbound variables (which comes in two modes: conservative and liberal,
default to conservative)
- unused variables (disabled by default for compatibilty)
- assignment on built-in identifiers
- function application's arity mismatch
- function application's invalid function name
- non-string literal in the first argument of set / set_var
The existing parser and evaluator are modified as follows:
- The new (caching) evaluator no longer needs to perform variable
hoisting at runtime.
- Note that for array assignment, this changes the semantics.
- The new parser is more lenient, reducing parsing errors.
The static analyzer will catch these errors instead, allowing us
to give a much better error message and reduces the complexity of
the parser.
* The parser now allows function name to be any identifier.
* The parser now allows arity mismatch to occur.
* The parser now allows the first argument of set to be any expression.
Concretely, obvious changes that users will see are:
1. a := [1]; false & (a[] := 2); a[0] === 1
would evaluate to true, while it used to evaluate to the undefined value
due to hoisting
2. f(1)
will now error with 'f is not a valid function' as opposed to
'Unexpected "T_BRACE"'
3. length
will now error with 'Illegal use of built-in identifier "length"'
as opposed to 'Expected a ('
Appendix: conservative and liberal mode
The conservative mode is completely compatible with the current evaluator.
That is,
false & (a := 1); a
will not deem `a` as unbound, though this is actually undesirable because
`a` would then be bound to the troublesome undefined value.
The liberal mode rejects the above pattern by deeming `a` as unbound.
However, it also rejects
true & (a := 1); a
even though (a := 1) is always executed. Since there are several filters
in Wikimedia projects that rely on this behavior, we default the mode
to conservative for now.
Note that even the liberal mode doesn't really respect lexical scope
appeared in some other programming languages (see also T234690).
For instance:
(if true then (a := 1) else (a := 2) end); a
would be accepted by the liberal checker, even though under lexical scope,
`a` would be unbound. However, it is unlikely that lexical scope
will be suitable for the filter language, as most filters in
Wikimedia projects that have user-defined variable do violate lexical scope.
Bug: T260903
Bug: T238709
Bug: T237610
Bug: T234690
Bug: T231536
Change-Id: Ic6d030503e554933f8d220c6f87b680505918ae2
2019-11-20 07:09:41 +00:00
|
|
|
return $this->check( $right, $this->check( $left, $bound ) );
|
|
|
|
|
|
|
|
case AFPTreeNode::UNARY:
|
2024-03-16 18:52:48 +00:00
|
|
|
[ , $argument ] = $node->children;
|
Add a static analyzer for the filter language
This commit adds a class AFPSyntaxChecker which can statically analyze
a filter code to detect the following errors:
- unbound variables (which comes in two modes: conservative and liberal,
default to conservative)
- unused variables (disabled by default for compatibilty)
- assignment on built-in identifiers
- function application's arity mismatch
- function application's invalid function name
- non-string literal in the first argument of set / set_var
The existing parser and evaluator are modified as follows:
- The new (caching) evaluator no longer needs to perform variable
hoisting at runtime.
- Note that for array assignment, this changes the semantics.
- The new parser is more lenient, reducing parsing errors.
The static analyzer will catch these errors instead, allowing us
to give a much better error message and reduces the complexity of
the parser.
* The parser now allows function name to be any identifier.
* The parser now allows arity mismatch to occur.
* The parser now allows the first argument of set to be any expression.
Concretely, obvious changes that users will see are:
1. a := [1]; false & (a[] := 2); a[0] === 1
would evaluate to true, while it used to evaluate to the undefined value
due to hoisting
2. f(1)
will now error with 'f is not a valid function' as opposed to
'Unexpected "T_BRACE"'
3. length
will now error with 'Illegal use of built-in identifier "length"'
as opposed to 'Expected a ('
Appendix: conservative and liberal mode
The conservative mode is completely compatible with the current evaluator.
That is,
false & (a := 1); a
will not deem `a` as unbound, though this is actually undesirable because
`a` would then be bound to the troublesome undefined value.
The liberal mode rejects the above pattern by deeming `a` as unbound.
However, it also rejects
true & (a := 1); a
even though (a := 1) is always executed. Since there are several filters
in Wikimedia projects that rely on this behavior, we default the mode
to conservative for now.
Note that even the liberal mode doesn't really respect lexical scope
appeared in some other programming languages (see also T234690).
For instance:
(if true then (a := 1) else (a := 2) end); a
would be accepted by the liberal checker, even though under lexical scope,
`a` would be unbound. However, it is unlikely that lexical scope
will be suitable for the filter language, as most filters in
Wikimedia projects that have user-defined variable do violate lexical scope.
Bug: T260903
Bug: T238709
Bug: T237610
Bug: T234690
Bug: T231536
Change-Id: Ic6d030503e554933f8d220c6f87b680505918ae2
2019-11-20 07:09:41 +00:00
|
|
|
return $this->check( $argument, $bound );
|
|
|
|
|
|
|
|
case AFPTreeNode::BOOL_INVERT:
|
2024-03-16 18:52:48 +00:00
|
|
|
[ $argument ] = $node->children;
|
Add a static analyzer for the filter language
This commit adds a class AFPSyntaxChecker which can statically analyze
a filter code to detect the following errors:
- unbound variables (which comes in two modes: conservative and liberal,
default to conservative)
- unused variables (disabled by default for compatibilty)
- assignment on built-in identifiers
- function application's arity mismatch
- function application's invalid function name
- non-string literal in the first argument of set / set_var
The existing parser and evaluator are modified as follows:
- The new (caching) evaluator no longer needs to perform variable
hoisting at runtime.
- Note that for array assignment, this changes the semantics.
- The new parser is more lenient, reducing parsing errors.
The static analyzer will catch these errors instead, allowing us
to give a much better error message and reduces the complexity of
the parser.
* The parser now allows function name to be any identifier.
* The parser now allows arity mismatch to occur.
* The parser now allows the first argument of set to be any expression.
Concretely, obvious changes that users will see are:
1. a := [1]; false & (a[] := 2); a[0] === 1
would evaluate to true, while it used to evaluate to the undefined value
due to hoisting
2. f(1)
will now error with 'f is not a valid function' as opposed to
'Unexpected "T_BRACE"'
3. length
will now error with 'Illegal use of built-in identifier "length"'
as opposed to 'Expected a ('
Appendix: conservative and liberal mode
The conservative mode is completely compatible with the current evaluator.
That is,
false & (a := 1); a
will not deem `a` as unbound, though this is actually undesirable because
`a` would then be bound to the troublesome undefined value.
The liberal mode rejects the above pattern by deeming `a` as unbound.
However, it also rejects
true & (a := 1); a
even though (a := 1) is always executed. Since there are several filters
in Wikimedia projects that rely on this behavior, we default the mode
to conservative for now.
Note that even the liberal mode doesn't really respect lexical scope
appeared in some other programming languages (see also T234690).
For instance:
(if true then (a := 1) else (a := 2) end); a
would be accepted by the liberal checker, even though under lexical scope,
`a` would be unbound. However, it is unlikely that lexical scope
will be suitable for the filter language, as most filters in
Wikimedia projects that have user-defined variable do violate lexical scope.
Bug: T260903
Bug: T238709
Bug: T237610
Bug: T234690
Bug: T231536
Change-Id: Ic6d030503e554933f8d220c6f87b680505918ae2
2019-11-20 07:09:41 +00:00
|
|
|
return $this->check( $argument, $bound );
|
|
|
|
// phpcs:ignore PSR2.ControlStructures.SwitchDeclaration.TerminatingComment
|
|
|
|
case AFPTreeNode::CONDITIONAL:
|
2024-03-16 18:52:48 +00:00
|
|
|
[ $condition, $exprIfTrue, $exprIfFalse ] = $node->children;
|
Add a static analyzer for the filter language
This commit adds a class AFPSyntaxChecker which can statically analyze
a filter code to detect the following errors:
- unbound variables (which comes in two modes: conservative and liberal,
default to conservative)
- unused variables (disabled by default for compatibilty)
- assignment on built-in identifiers
- function application's arity mismatch
- function application's invalid function name
- non-string literal in the first argument of set / set_var
The existing parser and evaluator are modified as follows:
- The new (caching) evaluator no longer needs to perform variable
hoisting at runtime.
- Note that for array assignment, this changes the semantics.
- The new parser is more lenient, reducing parsing errors.
The static analyzer will catch these errors instead, allowing us
to give a much better error message and reduces the complexity of
the parser.
* The parser now allows function name to be any identifier.
* The parser now allows arity mismatch to occur.
* The parser now allows the first argument of set to be any expression.
Concretely, obvious changes that users will see are:
1. a := [1]; false & (a[] := 2); a[0] === 1
would evaluate to true, while it used to evaluate to the undefined value
due to hoisting
2. f(1)
will now error with 'f is not a valid function' as opposed to
'Unexpected "T_BRACE"'
3. length
will now error with 'Illegal use of built-in identifier "length"'
as opposed to 'Expected a ('
Appendix: conservative and liberal mode
The conservative mode is completely compatible with the current evaluator.
That is,
false & (a := 1); a
will not deem `a` as unbound, though this is actually undesirable because
`a` would then be bound to the troublesome undefined value.
The liberal mode rejects the above pattern by deeming `a` as unbound.
However, it also rejects
true & (a := 1); a
even though (a := 1) is always executed. Since there are several filters
in Wikimedia projects that rely on this behavior, we default the mode
to conservative for now.
Note that even the liberal mode doesn't really respect lexical scope
appeared in some other programming languages (see also T234690).
For instance:
(if true then (a := 1) else (a := 2) end); a
would be accepted by the liberal checker, even though under lexical scope,
`a` would be unbound. However, it is unlikely that lexical scope
will be suitable for the filter language, as most filters in
Wikimedia projects that have user-defined variable do violate lexical scope.
Bug: T260903
Bug: T238709
Bug: T237610
Bug: T234690
Bug: T231536
Change-Id: Ic6d030503e554933f8d220c6f87b680505918ae2
2019-11-20 07:09:41 +00:00
|
|
|
$bound = $this->check( $condition, $bound );
|
|
|
|
$boundLeft = $this->check( $exprIfTrue, $bound );
|
|
|
|
$boundRight = $this->check( $exprIfFalse, $bound );
|
|
|
|
switch ( $this->mode ) {
|
|
|
|
case self::MCONSERVATIVE:
|
|
|
|
return $this->mapUnion( $boundLeft, $boundRight );
|
|
|
|
case self::MLIBERAL:
|
|
|
|
return $this->mapIntersect( $boundLeft, $boundRight );
|
|
|
|
default:
|
|
|
|
// @codeCoverageIgnoreStart
|
|
|
|
throw new LogicException( "Unknown mode: {$this->mode}" );
|
|
|
|
// @codeCoverageIgnoreEnd
|
|
|
|
}
|
|
|
|
|
|
|
|
case AFPTreeNode::INDEX_ASSIGNMENT:
|
2024-03-16 18:52:48 +00:00
|
|
|
[ $varName, $offset, $value ] = $node->children;
|
Add a static analyzer for the filter language
This commit adds a class AFPSyntaxChecker which can statically analyze
a filter code to detect the following errors:
- unbound variables (which comes in two modes: conservative and liberal,
default to conservative)
- unused variables (disabled by default for compatibilty)
- assignment on built-in identifiers
- function application's arity mismatch
- function application's invalid function name
- non-string literal in the first argument of set / set_var
The existing parser and evaluator are modified as follows:
- The new (caching) evaluator no longer needs to perform variable
hoisting at runtime.
- Note that for array assignment, this changes the semantics.
- The new parser is more lenient, reducing parsing errors.
The static analyzer will catch these errors instead, allowing us
to give a much better error message and reduces the complexity of
the parser.
* The parser now allows function name to be any identifier.
* The parser now allows arity mismatch to occur.
* The parser now allows the first argument of set to be any expression.
Concretely, obvious changes that users will see are:
1. a := [1]; false & (a[] := 2); a[0] === 1
would evaluate to true, while it used to evaluate to the undefined value
due to hoisting
2. f(1)
will now error with 'f is not a valid function' as opposed to
'Unexpected "T_BRACE"'
3. length
will now error with 'Illegal use of built-in identifier "length"'
as opposed to 'Expected a ('
Appendix: conservative and liberal mode
The conservative mode is completely compatible with the current evaluator.
That is,
false & (a := 1); a
will not deem `a` as unbound, though this is actually undesirable because
`a` would then be bound to the troublesome undefined value.
The liberal mode rejects the above pattern by deeming `a` as unbound.
However, it also rejects
true & (a := 1); a
even though (a := 1) is always executed. Since there are several filters
in Wikimedia projects that rely on this behavior, we default the mode
to conservative for now.
Note that even the liberal mode doesn't really respect lexical scope
appeared in some other programming languages (see also T234690).
For instance:
(if true then (a := 1) else (a := 2) end); a
would be accepted by the liberal checker, even though under lexical scope,
`a` would be unbound. However, it is unlikely that lexical scope
will be suitable for the filter language, as most filters in
Wikimedia projects that have user-defined variable do violate lexical scope.
Bug: T260903
Bug: T238709
Bug: T237610
Bug: T234690
Bug: T231536
Change-Id: Ic6d030503e554933f8d220c6f87b680505918ae2
2019-11-20 07:09:41 +00:00
|
|
|
|
|
|
|
// deal with unbound $varName
|
|
|
|
$bound = $this->lookupVar( $varName, $node->position, $bound );
|
|
|
|
$bound = $this->check( $offset, $bound );
|
|
|
|
$bound = $this->check( $value, $bound );
|
|
|
|
// deal with built-in $varName and set $varName as unused
|
|
|
|
return $this->assignVar( $varName, $node->position, $bound );
|
|
|
|
|
|
|
|
case AFPTreeNode::ARRAY_APPEND:
|
2024-03-16 18:52:48 +00:00
|
|
|
[ $varName, $value ] = $node->children;
|
Add a static analyzer for the filter language
This commit adds a class AFPSyntaxChecker which can statically analyze
a filter code to detect the following errors:
- unbound variables (which comes in two modes: conservative and liberal,
default to conservative)
- unused variables (disabled by default for compatibilty)
- assignment on built-in identifiers
- function application's arity mismatch
- function application's invalid function name
- non-string literal in the first argument of set / set_var
The existing parser and evaluator are modified as follows:
- The new (caching) evaluator no longer needs to perform variable
hoisting at runtime.
- Note that for array assignment, this changes the semantics.
- The new parser is more lenient, reducing parsing errors.
The static analyzer will catch these errors instead, allowing us
to give a much better error message and reduces the complexity of
the parser.
* The parser now allows function name to be any identifier.
* The parser now allows arity mismatch to occur.
* The parser now allows the first argument of set to be any expression.
Concretely, obvious changes that users will see are:
1. a := [1]; false & (a[] := 2); a[0] === 1
would evaluate to true, while it used to evaluate to the undefined value
due to hoisting
2. f(1)
will now error with 'f is not a valid function' as opposed to
'Unexpected "T_BRACE"'
3. length
will now error with 'Illegal use of built-in identifier "length"'
as opposed to 'Expected a ('
Appendix: conservative and liberal mode
The conservative mode is completely compatible with the current evaluator.
That is,
false & (a := 1); a
will not deem `a` as unbound, though this is actually undesirable because
`a` would then be bound to the troublesome undefined value.
The liberal mode rejects the above pattern by deeming `a` as unbound.
However, it also rejects
true & (a := 1); a
even though (a := 1) is always executed. Since there are several filters
in Wikimedia projects that rely on this behavior, we default the mode
to conservative for now.
Note that even the liberal mode doesn't really respect lexical scope
appeared in some other programming languages (see also T234690).
For instance:
(if true then (a := 1) else (a := 2) end); a
would be accepted by the liberal checker, even though under lexical scope,
`a` would be unbound. However, it is unlikely that lexical scope
will be suitable for the filter language, as most filters in
Wikimedia projects that have user-defined variable do violate lexical scope.
Bug: T260903
Bug: T238709
Bug: T237610
Bug: T234690
Bug: T231536
Change-Id: Ic6d030503e554933f8d220c6f87b680505918ae2
2019-11-20 07:09:41 +00:00
|
|
|
|
|
|
|
// deal with unbound $varName
|
|
|
|
$bound = $this->lookupVar( $varName, $node->position, $bound );
|
|
|
|
$bound = $this->check( $value, $bound );
|
|
|
|
// deal with built-in $varName and set $varName as unused
|
|
|
|
return $this->assignVar( $varName, $node->position, $bound );
|
|
|
|
|
|
|
|
case AFPTreeNode::SEMICOLON:
|
|
|
|
// @phan-suppress-next-line PhanTypeSuspiciousNonTraversableForeach children is array here
|
|
|
|
foreach ( $node->children as $statement ) {
|
|
|
|
$bound = $this->check( $statement, $bound );
|
|
|
|
}
|
|
|
|
return $bound;
|
|
|
|
|
|
|
|
default:
|
|
|
|
// @codeCoverageIgnoreStart
|
|
|
|
throw new LogicException( "Unknown type: {$node->type}" );
|
|
|
|
// @codeCoverageIgnoreEnd
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* @param array $left
|
|
|
|
* @param array $right
|
|
|
|
* @return array
|
|
|
|
*/
|
|
|
|
private function mapUnion( array $left, array $right ): array {
|
|
|
|
foreach ( $right as $key => $val ) {
|
|
|
|
if ( array_key_exists( $key, $left ) ) {
|
|
|
|
$left[ $key ] = $left[ $key ] || $val;
|
|
|
|
} else {
|
|
|
|
$left[ $key ] = $val;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return $left;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* @param array $left
|
|
|
|
* @param array $right
|
|
|
|
* @return array
|
|
|
|
*/
|
|
|
|
private function mapIntersect( array $left, array $right ): array {
|
|
|
|
$keys = array_intersect_key( $left, $right );
|
|
|
|
$result = [];
|
|
|
|
foreach ( $keys as $key => $val ) {
|
|
|
|
$result[ $key ] = $left[ $key ] || $right[ $key ];
|
|
|
|
}
|
|
|
|
return $result;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* @param string $var
|
|
|
|
* @param int $pos
|
|
|
|
* @param array $bound
|
|
|
|
* @return array
|
|
|
|
*/
|
|
|
|
private function assignVar( string $var, int $pos, array $bound ): array {
|
|
|
|
$var = strtolower( $var );
|
|
|
|
if ( $this->isReservedIdentifier( $var ) ) {
|
|
|
|
throw new UserVisibleException(
|
|
|
|
'overridebuiltin',
|
|
|
|
$pos,
|
|
|
|
[ $var ]
|
|
|
|
);
|
|
|
|
}
|
|
|
|
$bound[ $var ] = false;
|
|
|
|
return $bound;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* @param string $var
|
|
|
|
* @param int $pos
|
|
|
|
* @param array $bound
|
|
|
|
* @return array
|
|
|
|
*/
|
|
|
|
private function lookupVar( string $var, int $pos, array $bound ): array {
|
|
|
|
$var = strtolower( $var );
|
|
|
|
if ( array_key_exists( $var, $bound ) ) {
|
|
|
|
// user-defined variable
|
|
|
|
$bound[ $var ] = true;
|
|
|
|
return $bound;
|
|
|
|
} elseif ( $this->keywordsManager->isVarDisabled( $var ) ) {
|
|
|
|
// disabled built-in variables
|
|
|
|
throw new UserVisibleException(
|
|
|
|
'disabledvar',
|
|
|
|
$pos,
|
|
|
|
[ $var ]
|
|
|
|
);
|
|
|
|
} elseif ( $this->keywordsManager->varExists( $var ) ) {
|
|
|
|
// non-disabled built-in variables
|
|
|
|
return $bound;
|
|
|
|
} elseif ( $this->isReservedIdentifier( $var ) ) {
|
|
|
|
// other built-in identifiers
|
|
|
|
throw new UserVisibleException(
|
|
|
|
'usebuiltin',
|
|
|
|
$pos,
|
|
|
|
[ $var ]
|
|
|
|
);
|
|
|
|
} else {
|
|
|
|
// unbound variables
|
|
|
|
throw new UserVisibleException(
|
|
|
|
'unrecognisedvar',
|
|
|
|
$pos,
|
|
|
|
[ $var ]
|
|
|
|
);
|
|
|
|
}
|
|
|
|
}
|
2020-10-03 13:01:53 +00:00
|
|
|
|
|
|
|
/**
|
|
|
|
* Check that a built-in function has been provided the right amount of arguments
|
|
|
|
*
|
|
|
|
* @param array $args The arguments supplied to the function
|
|
|
|
* @param string $func The function name
|
|
|
|
* @param int $position
|
|
|
|
* @throws UserVisibleException
|
|
|
|
*/
|
2023-06-23 10:28:06 +00:00
|
|
|
private function checkArgCount( array $args, string $func, int $position ): void {
|
2021-08-31 22:26:13 +00:00
|
|
|
if ( !array_key_exists( $func, FilterEvaluator::FUNC_ARG_COUNT ) ) {
|
2020-10-03 13:01:53 +00:00
|
|
|
// @codeCoverageIgnoreStart
|
|
|
|
throw new InvalidArgumentException( "$func is not a valid function." );
|
|
|
|
// @codeCoverageIgnoreEnd
|
|
|
|
}
|
2024-03-16 18:52:48 +00:00
|
|
|
[ $min, $max ] = FilterEvaluator::FUNC_ARG_COUNT[ $func ];
|
2020-10-03 13:01:53 +00:00
|
|
|
if ( count( $args ) < $min ) {
|
|
|
|
throw new UserVisibleException(
|
|
|
|
$min === 1 ? 'noparams' : 'notenoughargs',
|
|
|
|
$position,
|
|
|
|
[ $func, $min, count( $args ) ]
|
|
|
|
);
|
|
|
|
} elseif ( count( $args ) > $max ) {
|
|
|
|
throw new UserVisibleException(
|
|
|
|
'toomanyargs',
|
|
|
|
$position,
|
|
|
|
[ $func, $max, count( $args ) ]
|
|
|
|
);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Check whether the given name is a reserved identifier, e.g. the name of a built-in variable,
|
|
|
|
* function, or keyword.
|
|
|
|
*
|
|
|
|
* @param string $name
|
|
|
|
* @return bool
|
|
|
|
*/
|
2023-06-23 10:28:06 +00:00
|
|
|
private function isReservedIdentifier( string $name ): bool {
|
2020-10-03 13:01:53 +00:00
|
|
|
return $this->keywordsManager->varExists( $name ) ||
|
2021-08-31 22:26:13 +00:00
|
|
|
array_key_exists( $name, FilterEvaluator::FUNCTIONS ) ||
|
2020-10-03 13:01:53 +00:00
|
|
|
// We need to check for true, false, if/then/else etc. because, even if they have a different
|
|
|
|
// AFPToken type, they may be used inside set/set_var()
|
|
|
|
in_array( $name, AbuseFilterTokenizer::KEYWORDS, true );
|
|
|
|
}
|
Add a static analyzer for the filter language
This commit adds a class AFPSyntaxChecker which can statically analyze
a filter code to detect the following errors:
- unbound variables (which comes in two modes: conservative and liberal,
default to conservative)
- unused variables (disabled by default for compatibilty)
- assignment on built-in identifiers
- function application's arity mismatch
- function application's invalid function name
- non-string literal in the first argument of set / set_var
The existing parser and evaluator are modified as follows:
- The new (caching) evaluator no longer needs to perform variable
hoisting at runtime.
- Note that for array assignment, this changes the semantics.
- The new parser is more lenient, reducing parsing errors.
The static analyzer will catch these errors instead, allowing us
to give a much better error message and reduces the complexity of
the parser.
* The parser now allows function name to be any identifier.
* The parser now allows arity mismatch to occur.
* The parser now allows the first argument of set to be any expression.
Concretely, obvious changes that users will see are:
1. a := [1]; false & (a[] := 2); a[0] === 1
would evaluate to true, while it used to evaluate to the undefined value
due to hoisting
2. f(1)
will now error with 'f is not a valid function' as opposed to
'Unexpected "T_BRACE"'
3. length
will now error with 'Illegal use of built-in identifier "length"'
as opposed to 'Expected a ('
Appendix: conservative and liberal mode
The conservative mode is completely compatible with the current evaluator.
That is,
false & (a := 1); a
will not deem `a` as unbound, though this is actually undesirable because
`a` would then be bound to the troublesome undefined value.
The liberal mode rejects the above pattern by deeming `a` as unbound.
However, it also rejects
true & (a := 1); a
even though (a := 1) is always executed. Since there are several filters
in Wikimedia projects that rely on this behavior, we default the mode
to conservative for now.
Note that even the liberal mode doesn't really respect lexical scope
appeared in some other programming languages (see also T234690).
For instance:
(if true then (a := 1) else (a := 2) end); a
would be accepted by the liberal checker, even though under lexical scope,
`a` would be unbound. However, it is unlikely that lexical scope
will be suitable for the filter language, as most filters in
Wikimedia projects that have user-defined variable do violate lexical scope.
Bug: T260903
Bug: T238709
Bug: T237610
Bug: T234690
Bug: T231536
Change-Id: Ic6d030503e554933f8d220c6f87b680505918ae2
2019-11-20 07:09:41 +00:00
|
|
|
}
|