fish-shell/src/parse_tree.h

497 lines
20 KiB
C
Raw Normal View History

// Programmatic representation of fish code.
2013-07-26 06:24:22 +08:00
#ifndef FISH_PARSE_PRODUCTIONS_H
#define FISH_PARSE_PRODUCTIONS_H
2015-07-25 23:14:25 +08:00
#include <stddef.h>
#include <stdint.h>
#include <sys/types.h>
#include <memory>
#include <vector>
#include "common.h"
#include "maybe.h"
#include "parse_constants.h"
2018-01-08 11:07:49 +08:00
#include "parse_grammar.h"
#include "tokenizer.h"
2013-06-12 00:37:51 +08:00
class parse_node_tree_t;
typedef uint32_t node_offset_t;
2013-06-23 17:09:46 +08:00
#define NODE_OFFSET_INVALID (static_cast<node_offset_t>(-1))
typedef uint32_t source_offset_t;
constexpr source_offset_t SOURCE_OFFSET_INVALID = static_cast<source_offset_t>(-1);
/// A struct representing the token type that we use internally.
struct parse_token_t {
enum parse_token_type_t type; // The type of the token as represented by the parser
enum parse_keyword_t keyword; // Any keyword represented by this token
bool has_dash_prefix; // Hackish: whether the source contains a dash prefix
bool is_help_argument; // Hackish: whether the source looks like '-h' or '--help'
source_offset_t source_start;
source_offset_t source_length;
2013-08-11 15:35:00 +08:00
wcstring describe() const;
wcstring user_presentable_description() const;
2013-08-11 15:35:00 +08:00
};
enum {
2013-08-11 15:35:00 +08:00
parse_flag_none = 0,
/// Attempt to build a "parse tree" no matter what. This may result in a 'forest' of
/// disconnected trees. This is intended to be used by syntax highlighting.
2013-08-11 15:35:00 +08:00
parse_flag_continue_after_error = 1 << 0,
/// Include comment tokens.
parse_flag_include_comments = 1 << 1,
/// Indicate that the tokenizer should accept incomplete tokens */
parse_flag_accept_incomplete_tokens = 1 << 2,
/// Indicate that the parser should not generate the terminate token, allowing an 'unfinished'
/// tree where some nodes may have no productions.
parse_flag_leave_unterminated = 1 << 3,
/// Indicate that the parser should generate job_list entries for blank lines.
parse_flag_show_blank_lines = 1 << 4
2013-06-12 00:37:51 +08:00
};
2013-08-11 15:35:00 +08:00
typedef unsigned int parse_tree_flags_t;
wcstring parse_dump_tree(const parse_node_tree_t &tree, const wcstring &src);
2013-06-12 00:37:51 +08:00
const wchar_t *token_type_description(parse_token_type_t type);
const wchar_t *keyword_description(parse_keyword_t type);
2013-06-23 17:09:46 +08:00
// Node flags.
enum {
/// Flag indicating that the node has associated comment nodes.
parse_node_flag_has_comments = 1 << 0,
};
typedef uint8_t parse_node_flags_t;
/// Node-type specific tag value.
typedef uint8_t parse_node_tag_t;
/// Class for nodes of a parse tree. Since there's a lot of these, the size and order of the fields
/// is important.
class parse_node_t {
public:
// Start in the source code.
source_offset_t source_start;
// Length of our range in the source code.
source_offset_t source_length;
// Parent
2013-10-07 16:04:37 +08:00
node_offset_t parent;
// Children
2013-06-12 00:37:51 +08:00
node_offset_t child_start;
// Number of children.
uint8_t child_count;
// Type of the node.
enum parse_token_type_t type;
// Keyword associated with node.
enum parse_keyword_t keyword;
// Node flags.
parse_node_flags_t flags : 4;
// This is used to store e.g. the statement decoration.
parse_node_tag_t tag : 4;
// Description
2013-06-12 00:37:51 +08:00
wcstring describe(void) const;
// Constructor
explicit parse_node_t(parse_token_type_t ty)
: source_start(SOURCE_OFFSET_INVALID),
source_length(0),
parent(NODE_OFFSET_INVALID),
child_start(0),
child_count(0),
type(ty),
keyword(parse_keyword_none),
flags(0),
tag(0) {}
node_offset_t child_offset(node_offset_t which) const {
2013-06-23 17:09:46 +08:00
PARSE_ASSERT(which < child_count);
return child_start + which;
}
2013-08-11 15:35:00 +08:00
/// Indicate if this node has a range of source code associated with it.
bool has_source() const {
// Should never have a nonempty range with an invalid offset.
assert(this->source_start != SOURCE_OFFSET_INVALID || this->source_length == 0);
return this->source_length > 0;
2013-08-09 06:06:46 +08:00
}
/// Indicate if the node has comment nodes.
bool has_comments() const {
return static_cast<bool>(this->flags & parse_node_flag_has_comments);
}
/// Gets source for the node, or the empty string if it has no source.
wcstring get_source(const wcstring &str) const {
if (!has_source())
2013-10-09 06:05:30 +08:00
return wcstring();
else
return wcstring(str, this->source_start, this->source_length);
}
/// Returns whether the given location is within the source range or at its end.
bool location_in_or_at_end_of_source_range(size_t loc) const {
2013-10-13 09:17:03 +08:00
return has_source() && source_start <= loc && loc - source_start <= source_length;
}
2013-06-12 00:37:51 +08:00
};
template <typename Type>
class tnode_t;
/// The parse tree itself.
class parse_node_tree_t : public std::vector<parse_node_t> {
public:
parse_node_tree_t() {}
parse_node_tree_t(parse_node_tree_t &&) = default;
parse_node_tree_t &operator=(parse_node_tree_t &&) = default;
parse_node_tree_t(const parse_node_tree_t &) = delete; // no copying
parse_node_tree_t &operator=(const parse_node_tree_t &) = delete; // no copying
// Get the node corresponding to a child of the given node, or NULL if there is no such child.
// If expected_type is provided, assert that the node has that type.
const parse_node_t *get_child(const parse_node_t &parent, node_offset_t which,
parse_token_type_t expected_type = token_type_invalid) const;
// Find the first direct child of the given node of the given type. asserts on failure.
const parse_node_t &find_child(const parse_node_t &parent, parse_token_type_t type) const;
template <typename Type>
tnode_t<Type> find_child(const parse_node_t &parent) const;
// Get the node corresponding to the parent of the given node, or NULL if there is no such
// child. If expected_type is provided, only returns the parent if it is of that type. Note the
// asymmetry: get_child asserts since the children are known, but get_parent does not, since the
// parent may not be known.
const parse_node_t *get_parent(const parse_node_t &node,
parse_token_type_t expected_type = token_type_invalid) const;
// Find all the nodes of a given type underneath a given node, up to max_count of them.
2013-08-09 06:06:46 +08:00
typedef std::vector<const parse_node_t *> parse_node_list_t;
parse_node_list_t find_nodes(const parse_node_t &parent, parse_token_type_t type,
2018-01-12 09:19:48 +08:00
size_t max_count = size_t(-1)) const;
// Find all the nodes of a given type underneath a given node, up to max_count of them.
template <typename Type>
std::vector<tnode_t<Type>> find_nodes(const parse_node_t &parent, size_t max_count = -1) const;
// Finds the last node of a given type, or empty if it could not be found. If parent is NULL,
// this finds the last node in the tree of that type.
template <typename Type>
tnode_t<Type> find_last_node(const parse_node_t *parent = NULL) const;
// Finds a node containing the given source location. If 'parent' is not NULL, it must be an
// ancestor.
const parse_node_t *find_node_matching_source_location(parse_token_type_t type,
size_t source_loc,
const parse_node_t *parent) const;
// Indicate if the given argument_list or arguments_or_redirections_list is a root list, or has
// a parent.
bool argument_list_is_root(const parse_node_t &node) const;
// Utilities
/// Given a plain statement, get the decoration (from the parent node), or none if there is no
/// decoration.
enum parse_statement_decoration_t decoration_for_plain_statement(
const parse_node_t &node) const;
/// Given a plain statement, return true if the statement is part of a pipeline. If
/// include_first is set, the first command in a pipeline is considered part of it; otherwise
/// only the second or additional commands are.
bool statement_is_in_pipeline(const parse_node_t &node, bool include_first) const;
/// Given a redirection, get the redirection type (or TOK_NONE) and target (file path, or fd).
enum token_type type_for_redirection(const parse_node_t &node, const wcstring &src, int *out_fd,
wcstring *out_target) const;
/// If the given node is a block statement, returns the header node (for_header, while_header,
/// begin_header, or function_header). Otherwise returns NULL.
const parse_node_t *header_node_for_block_statement(const parse_node_t &node) const;
/// Given a node list (e.g. of type symbol_job_list) and a node type (e.g. symbol_job), return
/// the next element of the given type in that list, and the tail (by reference). Returns NULL
/// if we've exhausted the list.
const parse_node_t *next_node_in_node_list(const parse_node_t &node_list,
parse_token_type_t item_type,
const parse_node_t **list_tail) const;
/// Given a job, return all of its statements. These are 'specific statements' (e.g.
/// symbol_decorated_statement, not symbol_statement).
parse_node_list_t specific_statements_for_job(const parse_node_t &job) const;
2014-04-01 01:01:39 +08:00
/// Given a node, return all of its comment nodes.
parse_node_list_t comment_nodes_for_node(const parse_node_t &node) const;
/// Returns the boolean type for a boolean node.
static enum parse_bool_statement_type_t statement_boolean_type(const parse_node_t &node);
/// Given a job, return whether it should be backgrounded, because it has a & specifier.
bool job_should_be_backgrounded(const parse_node_t &job) const;
private:
// Finds the last node of a given type underneath a given node, or NULL if it could not be
// found. If parent is NULL, this finds the last node in the tree of that type.
const parse_node_t *find_last_node_of_type(parse_token_type_t type,
const parse_node_t *parent) const;
2013-10-07 16:04:37 +08:00
};
struct source_range_t {
uint32_t start;
uint32_t length;
};
2018-01-08 11:07:49 +08:00
/// A helper for type-safe manipulation of parse nodes.
/// This is a lightweight value-type class.
template <typename Type>
class tnode_t {
/// The tree containing our node.
const parse_node_tree_t *tree = nullptr;
/// The node in the tree
const parse_node_t *nodeptr = nullptr;
// Helper to get a child type at a given index.
template <class Element, uint32_t Index>
using child_at = typename std::tuple_element<Index, typename Element::type_tuple>::type;
public:
tnode_t() = default;
tnode_t(const parse_node_tree_t *t, const parse_node_t *n) : tree(t), nodeptr(n) {
assert(t && "tree cannot be null in this constructor");
assert((!n || n->type == Type::token) && "node has wrong type");
2018-01-08 11:07:49 +08:00
}
/// Temporary conversion to parse_node_t to assist in migration.
/* implicit */ operator const parse_node_t &() const {
assert(nodeptr && "Empty tnode_t");
return *nodeptr;
}
/* implicit */ operator const parse_node_t *() const { return nodeptr; }
2018-01-08 11:07:49 +08:00
/// Return the underlying (type-erased) node.
const parse_node_t *node() const { return nodeptr; }
/// Check whether we're populated.
explicit operator bool() const { return nodeptr != nullptr; }
bool operator==(const tnode_t &rhs) const { return tree == rhs.tree && nodeptr == rhs.nodeptr; }
bool operator!=(const tnode_t &rhs) const { return !(*this == rhs); }
bool has_source() const { return nodeptr && nodeptr->has_source(); }
maybe_t<source_range_t> source_range() const {
if (!has_source()) return none();
return source_range_t{nodeptr->source_start, nodeptr->source_length};
}
wcstring get_source(const wcstring &str) const {
assert(has_source() && "Source missing");
return nodeptr->get_source(str);
}
bool location_in_or_at_end_of_source_range(size_t loc) const {
return nodeptr && nodeptr->location_in_or_at_end_of_source_range(loc);
}
2018-01-12 09:19:48 +08:00
static tnode_t find_node_matching_source_location(const parse_node_tree_t *tree,
size_t source_loc,
const parse_node_t *parent) {
assert(tree && "null tree");
return tnode_t{tree,
tree->find_node_matching_source_location(Type::token, source_loc, parent)};
}
2018-01-08 11:07:49 +08:00
/// Type-safe access to a child at the given index.
template <node_offset_t Index>
tnode_t<child_at<Type, Index>> child() const {
using child_type = child_at<Type, Index>;
const parse_node_t *child = nullptr;
if (nodeptr) child = tree->get_child(*nodeptr, Index, child_type::token);
2018-01-08 11:07:49 +08:00
return tnode_t<child_type>{tree, child};
}
/// If the child at the given index has the given type, return it; otherwise return an empty
/// child.
/// This is used for e.g. alternations.
/// TODO: check that the type is possible (i.e. sum type).
template <class ChildType, node_offset_t Index>
tnode_t<ChildType> try_get_child() const {
const parse_node_t *child = nullptr;
if (nodeptr) child = tree->get_child(*nodeptr, Index);
if (child && child->type == ChildType::token) return {tree, child};
return {};
}
/// Type-safe access to a node's parent.
/// If the parent exists and has type ParentType, return it.
/// Otherwise return a missing tnode.
template <class ParentType>
tnode_t<ParentType> try_get_parent() const {
if (!nodeptr) return {};
return {tree, tree->get_parent(*nodeptr, ParentType::token)};
}
2018-01-12 09:19:48 +08:00
/// Finds all descendants (up to max_count) under this node of the given type.
template <typename DescendantType>
std::vector<tnode_t<DescendantType>> descendants(size_t max_count = -1) const {
if (!nodeptr) return {};
return tree->find_nodes<DescendantType>(*nodeptr);
}
/// Given that we are a list type, \return the next node of some Item in some node list,
/// adjusting 'this' to be the remainder of the list.
/// Returns an empty item on failure.
template <class ItemType>
tnode_t<ItemType> next_in_list() {
if (!nodeptr) return {};
const parse_node_t *next =
tree->next_node_in_node_list(*nodeptr, ItemType::token, &nodeptr);
return {tree, next};
}
2018-01-08 11:07:49 +08:00
};
template <typename Type>
tnode_t<Type> parse_node_tree_t::find_child(const parse_node_t &parent) const {
return tnode_t<Type>(this, &this->find_child(parent, Type::token));
}
template <typename Type>
tnode_t<Type> parse_node_tree_t::find_last_node(const parse_node_t *parent) const {
return tnode_t<Type>(this, this->find_last_node_of_type(Type::token, parent));
}
template <typename Type>
std::vector<tnode_t<Type>> parse_node_tree_t::find_nodes(const parse_node_t &parent,
size_t max_count) const {
auto ptrs = this->find_nodes(parent, Type::token, max_count);
std::vector<tnode_t<Type>> result;
result.reserve(ptrs.size());
for (const parse_node_t *np : ptrs) {
result.emplace_back(this, np);
}
return result;
}
/// Given a plain statement, get the command from the child node. Returns the command string on
/// success, none on failure.
maybe_t<wcstring> command_for_plain_statement(tnode_t<grammar::plain_statement> stmt,
const wcstring &src);
/// The big entry point. Parse a string, attempting to produce a tree for the given goal type.
bool parse_tree_from_string(const wcstring &str, parse_tree_flags_t flags,
parse_node_tree_t *output, parse_error_list_t *errors,
parse_token_type_t goal = symbol_job_list);
/// A type wrapping up a parse tree and the original source behind it.
struct parsed_source_t {
wcstring src;
parse_node_tree_t tree;
parsed_source_t(wcstring s, parse_node_tree_t t) : src(std::move(s)), tree(std::move(t)) {}
parsed_source_t(const parsed_source_t &) = delete;
void operator=(const parsed_source_t &) = delete;
parsed_source_t(parsed_source_t &&) = default;
parsed_source_t &operator=(parsed_source_t &&) = default;
};
/// Return a shared pointer to parsed_source_t, or null on failure.
using parsed_source_ref_t = std::shared_ptr<const parsed_source_t>;
parsed_source_ref_t parse_source(wcstring src, parse_tree_flags_t flags, parse_error_list_t *errors,
parse_token_type_t goal = symbol_job_list);
// Fish grammar:
//
// # A job_list is a list of jobs, separated by semicolons or newlines
//
// job_list = <empty> |
// job job_list |
// <TOK_END> job_list
//
// # A job is a non-empty list of statements, separated by pipes. (Non-empty is useful for cases
// like if statements, where we require a command). To represent "non-empty", we require a
// statement, followed by a possibly empty job_continuation, and then optionally a background
// specifier '&'
//
// job = statement job_continuation optional_background
// job_continuation = <empty> |
// <TOK_PIPE> statement job_continuation
//
// # A statement is a normal command, or an if / while / and etc
//
// statement = boolean_statement | block_statement | if_statement | switch_statement |
// decorated_statement
//
// # A block is a conditional, loop, or begin/end
//
// if_statement = if_clause else_clause end_command arguments_or_redirections_list
// if_clause = <IF> job <TOK_END> andor_job_list job_list
// else_clause = <empty> |
// <ELSE> else_continuation
// else_continuation = if_clause else_clause |
// <TOK_END> job_list
//
// switch_statement = SWITCH argument <TOK_END> case_item_list end_command
// arguments_or_redirections_list
// case_item_list = <empty> |
// case_item case_item_list |
// <TOK_END> case_item_list
//
// case_item = CASE argument_list <TOK_END> job_list
//
// block_statement = block_header job_list end_command arguments_or_redirections_list
// block_header = for_header | while_header | function_header | begin_header
// for_header = FOR var_name IN argument_list <TOK_END>
// while_header = WHILE job <TOK_END> andor_job_list
// begin_header = BEGIN
//
// # Functions take arguments, and require at least one (the name). No redirections allowed.
// function_header = FUNCTION argument argument_list <TOK_END>
//
// # A boolean statement is AND or OR or NOT
// boolean_statement = AND statement | OR statement | NOT statement
//
// # An andor_job_list is zero or more job lists, where each starts with an `and` or `or` boolean
// statement
// andor_job_list = <empty> |
// job andor_job_list |
// <TOK_END> andor_job_list
//
// # A decorated_statement is a command with a list of arguments_or_redirections, possibly with
// "builtin" or "command" or "exec"
//
// decorated_statement = plain_statement | COMMAND plain_statement | BUILTIN plain_statement |
// EXEC
//
// plain_statement
// plain_statement = <TOK_STRING> arguments_or_redirections_list
//
// argument_list = <empty> | argument argument_list
//
// arguments_or_redirections_list = <empty> |
// argument_or_redirection arguments_or_redirections_list
// argument_or_redirection = argument | redirection
// argument = <TOK_STRING>
//
// redirection = <TOK_REDIRECTION> <TOK_STRING>
//
// optional_background = <empty> | <TOK_BACKGROUND>
//
// end_command = END
//
// # A freestanding_argument_list is equivalent to a normal argument list, except it may contain
// TOK_END (newlines, and even semicolons, for historical reasons
//
// freestanding_argument_list = <empty> |
// argument freestanding_argument_list |
// <TOK_END> freestanding_argument_list
#endif