From ae3abd6e575940eb1221acf26c09e96352f052fa Mon Sep 17 00:00:00 2001 From: Aleksey Kladov Date: Thu, 13 Aug 2020 16:45:10 +0200 Subject: Rename ra_ssr -> ssr --- crates/ssr/src/errors.rs | 29 ++ crates/ssr/src/lib.rs | 338 +++++++++++++ crates/ssr/src/matching.rs | 777 ++++++++++++++++++++++++++++ crates/ssr/src/nester.rs | 94 ++++ crates/ssr/src/parsing.rs | 389 ++++++++++++++ crates/ssr/src/replacing.rs | 194 +++++++ crates/ssr/src/resolving.rs | 299 +++++++++++ crates/ssr/src/search.rs | 282 +++++++++++ crates/ssr/src/tests.rs | 1174 +++++++++++++++++++++++++++++++++++++++++++ 9 files changed, 3576 insertions(+) create mode 100644 crates/ssr/src/errors.rs create mode 100644 crates/ssr/src/lib.rs create mode 100644 crates/ssr/src/matching.rs create mode 100644 crates/ssr/src/nester.rs create mode 100644 crates/ssr/src/parsing.rs create mode 100644 crates/ssr/src/replacing.rs create mode 100644 crates/ssr/src/resolving.rs create mode 100644 crates/ssr/src/search.rs create mode 100644 crates/ssr/src/tests.rs (limited to 'crates/ssr/src') diff --git a/crates/ssr/src/errors.rs b/crates/ssr/src/errors.rs new file mode 100644 index 000000000..c02bacae6 --- /dev/null +++ b/crates/ssr/src/errors.rs @@ -0,0 +1,29 @@ +//! Code relating to errors produced by SSR. + +/// Constructs an SsrError taking arguments like the format macro. +macro_rules! _error { + ($fmt:expr) => {$crate::SsrError::new(format!($fmt))}; + ($fmt:expr, $($arg:tt)+) => {$crate::SsrError::new(format!($fmt, $($arg)+))} +} +pub(crate) use _error as error; + +/// Returns from the current function with an error, supplied by arguments as for format! +macro_rules! _bail { + ($($tokens:tt)*) => {return Err(crate::errors::error!($($tokens)*))} +} +pub(crate) use _bail as bail; + +#[derive(Debug, PartialEq)] +pub struct SsrError(pub(crate) String); + +impl std::fmt::Display for SsrError { + fn fmt(&self, f: &mut core::fmt::Formatter<'_>) -> core::fmt::Result { + write!(f, "Parse error: {}", self.0) + } +} + +impl SsrError { + pub(crate) fn new(message: impl Into) -> SsrError { + SsrError(message.into()) + } +} diff --git a/crates/ssr/src/lib.rs b/crates/ssr/src/lib.rs new file mode 100644 index 000000000..292bd5b9a --- /dev/null +++ b/crates/ssr/src/lib.rs @@ -0,0 +1,338 @@ +//! Structural Search Replace +//! +//! Allows searching the AST for code that matches one or more patterns and then replacing that code +//! based on a template. + +// Feature: Structural Search and Replace +// +// Search and replace with named wildcards that will match any expression, type, path, pattern or item. +// The syntax for a structural search replace command is ` ==>> `. +// A `$` placeholder in the search pattern will match any AST node and `$` will reference it in the replacement. +// Within a macro call, a placeholder will match up until whatever token follows the placeholder. +// +// All paths in both the search pattern and the replacement template must resolve in the context +// in which this command is invoked. Paths in the search pattern will then match the code if they +// resolve to the same item, even if they're written differently. For example if we invoke the +// command in the module `foo` with a pattern of `Bar`, then code in the parent module that refers +// to `foo::Bar` will match. +// +// Paths in the replacement template will be rendered appropriately for the context in which the +// replacement occurs. For example if our replacement template is `foo::Bar` and we match some +// code in the `foo` module, we'll insert just `Bar`. +// +// Inherent method calls should generally be written in UFCS form. e.g. `foo::Bar::baz($s, $a)` will +// match `$s.baz($a)`, provided the method call `baz` resolves to the method `foo::Bar::baz`. +// +// The scope of the search / replace will be restricted to the current selection if any, otherwise +// it will apply to the whole workspace. +// +// Placeholders may be given constraints by writing them as `${::...}`. +// +// Supported constraints: +// +// |=== +// | Constraint | Restricts placeholder +// +// | kind(literal) | Is a literal (e.g. `42` or `"forty two"`) +// | not(a) | Negates the constraint `a` +// |=== +// +// Available via the command `rust-analyzer.ssr`. +// +// ```rust +// // Using structural search replace command [foo($a, $b) ==>> ($a).foo($b)] +// +// // BEFORE +// String::from(foo(y + 5, z)) +// +// // AFTER +// String::from((y + 5).foo(z)) +// ``` +// +// |=== +// | Editor | Action Name +// +// | VS Code | **Rust Analyzer: Structural Search Replace** +// |=== + +mod matching; +mod nester; +mod parsing; +mod replacing; +mod resolving; +mod search; +#[macro_use] +mod errors; +#[cfg(test)] +mod tests; + +use crate::errors::bail; +pub use crate::errors::SsrError; +pub use crate::matching::Match; +use crate::matching::MatchFailureReason; +use base_db::{FileId, FilePosition, FileRange}; +use hir::Semantics; +use ide_db::source_change::SourceFileEdit; +use resolving::ResolvedRule; +use rustc_hash::FxHashMap; +use syntax::{ast, AstNode, SyntaxNode, TextRange}; + +// A structured search replace rule. Create by calling `parse` on a str. +#[derive(Debug)] +pub struct SsrRule { + /// A structured pattern that we're searching for. + pattern: parsing::RawPattern, + /// What we'll replace it with. + template: parsing::RawPattern, + parsed_rules: Vec, +} + +#[derive(Debug)] +pub struct SsrPattern { + raw: parsing::RawPattern, + parsed_rules: Vec, +} + +#[derive(Debug, Default)] +pub struct SsrMatches { + pub matches: Vec, +} + +/// Searches a crate for pattern matches and possibly replaces them with something else. +pub struct MatchFinder<'db> { + /// Our source of information about the user's code. + sema: Semantics<'db, ide_db::RootDatabase>, + rules: Vec, + resolution_scope: resolving::ResolutionScope<'db>, + restrict_ranges: Vec, +} + +impl<'db> MatchFinder<'db> { + /// Constructs a new instance where names will be looked up as if they appeared at + /// `lookup_context`. + pub fn in_context( + db: &'db ide_db::RootDatabase, + lookup_context: FilePosition, + mut restrict_ranges: Vec, + ) -> MatchFinder<'db> { + restrict_ranges.retain(|range| !range.range.is_empty()); + let sema = Semantics::new(db); + let resolution_scope = resolving::ResolutionScope::new(&sema, lookup_context); + MatchFinder { sema, rules: Vec::new(), resolution_scope, restrict_ranges } + } + + /// Constructs an instance using the start of the first file in `db` as the lookup context. + pub fn at_first_file(db: &'db ide_db::RootDatabase) -> Result, SsrError> { + use base_db::SourceDatabaseExt; + use ide_db::symbol_index::SymbolsDatabase; + if let Some(first_file_id) = db + .local_roots() + .iter() + .next() + .and_then(|root| db.source_root(root.clone()).iter().next()) + { + Ok(MatchFinder::in_context( + db, + FilePosition { file_id: first_file_id, offset: 0.into() }, + vec![], + )) + } else { + bail!("No files to search"); + } + } + + /// Adds a rule to be applied. The order in which rules are added matters. Earlier rules take + /// precedence. If a node is matched by an earlier rule, then later rules won't be permitted to + /// match to it. + pub fn add_rule(&mut self, rule: SsrRule) -> Result<(), SsrError> { + for parsed_rule in rule.parsed_rules { + self.rules.push(ResolvedRule::new( + parsed_rule, + &self.resolution_scope, + self.rules.len(), + )?); + } + Ok(()) + } + + /// Finds matches for all added rules and returns edits for all found matches. + pub fn edits(&self) -> Vec { + use base_db::SourceDatabaseExt; + let mut matches_by_file = FxHashMap::default(); + for m in self.matches().matches { + matches_by_file + .entry(m.range.file_id) + .or_insert_with(|| SsrMatches::default()) + .matches + .push(m); + } + let mut edits = vec![]; + for (file_id, matches) in matches_by_file { + let edit = + replacing::matches_to_edit(&matches, &self.sema.db.file_text(file_id), &self.rules); + edits.push(SourceFileEdit { file_id, edit }); + } + edits + } + + /// Adds a search pattern. For use if you intend to only call `find_matches_in_file`. If you + /// intend to do replacement, use `add_rule` instead. + pub fn add_search_pattern(&mut self, pattern: SsrPattern) -> Result<(), SsrError> { + for parsed_rule in pattern.parsed_rules { + self.rules.push(ResolvedRule::new( + parsed_rule, + &self.resolution_scope, + self.rules.len(), + )?); + } + Ok(()) + } + + /// Returns matches for all added rules. + pub fn matches(&self) -> SsrMatches { + let mut matches = Vec::new(); + let mut usage_cache = search::UsageCache::default(); + for rule in &self.rules { + self.find_matches_for_rule(rule, &mut usage_cache, &mut matches); + } + nester::nest_and_remove_collisions(matches, &self.sema) + } + + /// Finds all nodes in `file_id` whose text is exactly equal to `snippet` and attempts to match + /// them, while recording reasons why they don't match. This API is useful for command + /// line-based debugging where providing a range is difficult. + pub fn debug_where_text_equal(&self, file_id: FileId, snippet: &str) -> Vec { + use base_db::SourceDatabaseExt; + let file = self.sema.parse(file_id); + let mut res = Vec::new(); + let file_text = self.sema.db.file_text(file_id); + let mut remaining_text = file_text.as_str(); + let mut base = 0; + let len = snippet.len() as u32; + while let Some(offset) = remaining_text.find(snippet) { + let start = base + offset as u32; + let end = start + len; + self.output_debug_for_nodes_at_range( + file.syntax(), + FileRange { file_id, range: TextRange::new(start.into(), end.into()) }, + &None, + &mut res, + ); + remaining_text = &remaining_text[offset + snippet.len()..]; + base = end; + } + res + } + + fn output_debug_for_nodes_at_range( + &self, + node: &SyntaxNode, + range: FileRange, + restrict_range: &Option, + out: &mut Vec, + ) { + for node in node.children() { + let node_range = self.sema.original_range(&node); + if node_range.file_id != range.file_id || !node_range.range.contains_range(range.range) + { + continue; + } + if node_range.range == range.range { + for rule in &self.rules { + // For now we ignore rules that have a different kind than our node, otherwise + // we get lots of noise. If at some point we add support for restricting rules + // to a particular kind of thing (e.g. only match type references), then we can + // relax this. We special-case expressions, since function calls can match + // method calls. + if rule.pattern.node.kind() != node.kind() + && !(ast::Expr::can_cast(rule.pattern.node.kind()) + && ast::Expr::can_cast(node.kind())) + { + continue; + } + out.push(MatchDebugInfo { + matched: matching::get_match(true, rule, &node, restrict_range, &self.sema) + .map_err(|e| MatchFailureReason { + reason: e.reason.unwrap_or_else(|| { + "Match failed, but no reason was given".to_owned() + }), + }), + pattern: rule.pattern.node.clone(), + node: node.clone(), + }); + } + } else if let Some(macro_call) = ast::MacroCall::cast(node.clone()) { + if let Some(expanded) = self.sema.expand(¯o_call) { + if let Some(tt) = macro_call.token_tree() { + self.output_debug_for_nodes_at_range( + &expanded, + range, + &Some(self.sema.original_range(tt.syntax())), + out, + ); + } + } + } + self.output_debug_for_nodes_at_range(&node, range, restrict_range, out); + } + } +} + +pub struct MatchDebugInfo { + node: SyntaxNode, + /// Our search pattern parsed as an expression or item, etc + pattern: SyntaxNode, + matched: Result, +} + +impl std::fmt::Debug for MatchDebugInfo { + fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result { + match &self.matched { + Ok(_) => writeln!(f, "Node matched")?, + Err(reason) => writeln!(f, "Node failed to match because: {}", reason.reason)?, + } + writeln!( + f, + "============ AST ===========\n\ + {:#?}", + self.node + )?; + writeln!(f, "========= PATTERN ==========")?; + writeln!(f, "{:#?}", self.pattern)?; + writeln!(f, "============================")?; + Ok(()) + } +} + +impl SsrMatches { + /// Returns `self` with any nested matches removed and made into top-level matches. + pub fn flattened(self) -> SsrMatches { + let mut out = SsrMatches::default(); + self.flatten_into(&mut out); + out + } + + fn flatten_into(self, out: &mut SsrMatches) { + for mut m in self.matches { + for p in m.placeholder_values.values_mut() { + std::mem::replace(&mut p.inner_matches, SsrMatches::default()).flatten_into(out); + } + out.matches.push(m); + } + } +} + +impl Match { + pub fn matched_text(&self) -> String { + self.matched_node.text().to_string() + } +} + +impl std::error::Error for SsrError {} + +#[cfg(test)] +impl MatchDebugInfo { + pub(crate) fn match_failure_reason(&self) -> Option<&str> { + self.matched.as_ref().err().map(|r| r.reason.as_str()) + } +} diff --git a/crates/ssr/src/matching.rs b/crates/ssr/src/matching.rs new file mode 100644 index 000000000..ffc7202ae --- /dev/null +++ b/crates/ssr/src/matching.rs @@ -0,0 +1,777 @@ +//! This module is responsible for matching a search pattern against a node in the AST. In the +//! process of matching, placeholder values are recorded. + +use crate::{ + parsing::{Constraint, NodeKind, Placeholder}, + resolving::{ResolvedPattern, ResolvedRule, UfcsCallInfo}, + SsrMatches, +}; +use base_db::FileRange; +use hir::Semantics; +use rustc_hash::FxHashMap; +use std::{cell::Cell, iter::Peekable}; +use syntax::ast::{AstNode, AstToken}; +use syntax::{ast, SyntaxElement, SyntaxElementChildren, SyntaxKind, SyntaxNode, SyntaxToken}; +use test_utils::mark; + +// Creates a match error. If we're currently attempting to match some code that we thought we were +// going to match, as indicated by the --debug-snippet flag, then populate the reason field. +macro_rules! match_error { + ($e:expr) => {{ + MatchFailed { + reason: if recording_match_fail_reasons() { + Some(format!("{}", $e)) + } else { + None + } + } + }}; + ($fmt:expr, $($arg:tt)+) => {{ + MatchFailed { + reason: if recording_match_fail_reasons() { + Some(format!($fmt, $($arg)+)) + } else { + None + } + } + }}; +} + +// Fails the current match attempt, recording the supplied reason if we're recording match fail reasons. +macro_rules! fail_match { + ($($args:tt)*) => {return Err(match_error!($($args)*))}; +} + +/// Information about a match that was found. +#[derive(Debug)] +pub struct Match { + pub(crate) range: FileRange, + pub(crate) matched_node: SyntaxNode, + pub(crate) placeholder_values: FxHashMap, + pub(crate) ignored_comments: Vec, + pub(crate) rule_index: usize, + /// The depth of matched_node. + pub(crate) depth: usize, + // Each path in the template rendered for the module in which the match was found. + pub(crate) rendered_template_paths: FxHashMap, +} + +/// Represents a `$var` in an SSR query. +#[derive(Debug, Clone, PartialEq, Eq, Hash)] +pub(crate) struct Var(pub String); + +/// Information about a placeholder bound in a match. +#[derive(Debug)] +pub(crate) struct PlaceholderMatch { + /// The node that the placeholder matched to. If set, then we'll search for further matches + /// within this node. It isn't set when we match tokens within a macro call's token tree. + pub(crate) node: Option, + pub(crate) range: FileRange, + /// More matches, found within `node`. + pub(crate) inner_matches: SsrMatches, +} + +#[derive(Debug)] +pub(crate) struct MatchFailureReason { + pub(crate) reason: String, +} + +/// An "error" indicating that matching failed. Use the fail_match! macro to create and return this. +#[derive(Clone)] +pub(crate) struct MatchFailed { + /// The reason why we failed to match. Only present when debug_active true in call to + /// `get_match`. + pub(crate) reason: Option, +} + +/// Checks if `code` matches the search pattern found in `search_scope`, returning information about +/// the match, if it does. Since we only do matching in this module and searching is done by the +/// parent module, we don't populate nested matches. +pub(crate) fn get_match( + debug_active: bool, + rule: &ResolvedRule, + code: &SyntaxNode, + restrict_range: &Option, + sema: &Semantics, +) -> Result { + record_match_fails_reasons_scope(debug_active, || { + Matcher::try_match(rule, code, restrict_range, sema) + }) +} + +/// Checks if our search pattern matches a particular node of the AST. +struct Matcher<'db, 'sema> { + sema: &'sema Semantics<'db, ide_db::RootDatabase>, + /// If any placeholders come from anywhere outside of this range, then the match will be + /// rejected. + restrict_range: Option, + rule: &'sema ResolvedRule, +} + +/// Which phase of matching we're currently performing. We do two phases because most attempted +/// matches will fail and it means we can defer more expensive checks to the second phase. +enum Phase<'a> { + /// On the first phase, we perform cheap checks. No state is mutated and nothing is recorded. + First, + /// On the second phase, we construct the `Match`. Things like what placeholders bind to is + /// recorded. + Second(&'a mut Match), +} + +impl<'db, 'sema> Matcher<'db, 'sema> { + fn try_match( + rule: &ResolvedRule, + code: &SyntaxNode, + restrict_range: &Option, + sema: &'sema Semantics<'db, ide_db::RootDatabase>, + ) -> Result { + let match_state = Matcher { sema, restrict_range: restrict_range.clone(), rule }; + // First pass at matching, where we check that node types and idents match. + match_state.attempt_match_node(&mut Phase::First, &rule.pattern.node, code)?; + match_state.validate_range(&sema.original_range(code))?; + let mut the_match = Match { + range: sema.original_range(code), + matched_node: code.clone(), + placeholder_values: FxHashMap::default(), + ignored_comments: Vec::new(), + rule_index: rule.index, + depth: 0, + rendered_template_paths: FxHashMap::default(), + }; + // Second matching pass, where we record placeholder matches, ignored comments and maybe do + // any other more expensive checks that we didn't want to do on the first pass. + match_state.attempt_match_node( + &mut Phase::Second(&mut the_match), + &rule.pattern.node, + code, + )?; + the_match.depth = sema.ancestors_with_macros(the_match.matched_node.clone()).count(); + if let Some(template) = &rule.template { + the_match.render_template_paths(template, sema)?; + } + Ok(the_match) + } + + /// Checks that `range` is within the permitted range if any. This is applicable when we're + /// processing a macro expansion and we want to fail the match if we're working with a node that + /// didn't originate from the token tree of the macro call. + fn validate_range(&self, range: &FileRange) -> Result<(), MatchFailed> { + if let Some(restrict_range) = &self.restrict_range { + if restrict_range.file_id != range.file_id + || !restrict_range.range.contains_range(range.range) + { + fail_match!("Node originated from a macro"); + } + } + Ok(()) + } + + fn attempt_match_node( + &self, + phase: &mut Phase, + pattern: &SyntaxNode, + code: &SyntaxNode, + ) -> Result<(), MatchFailed> { + // Handle placeholders. + if let Some(placeholder) = self.get_placeholder(&SyntaxElement::Node(pattern.clone())) { + for constraint in &placeholder.constraints { + self.check_constraint(constraint, code)?; + } + if let Phase::Second(matches_out) = phase { + let original_range = self.sema.original_range(code); + // We validated the range for the node when we started the match, so the placeholder + // probably can't fail range validation, but just to be safe... + self.validate_range(&original_range)?; + matches_out.placeholder_values.insert( + Var(placeholder.ident.to_string()), + PlaceholderMatch::new(code, original_range), + ); + } + return Ok(()); + } + // We allow a UFCS call to match a method call, provided they resolve to the same function. + if let Some(pattern_ufcs) = self.rule.pattern.ufcs_function_calls.get(pattern) { + if let Some(code) = ast::MethodCallExpr::cast(code.clone()) { + return self.attempt_match_ufcs_to_method_call(phase, pattern_ufcs, &code); + } + if let Some(code) = ast::CallExpr::cast(code.clone()) { + return self.attempt_match_ufcs_to_ufcs(phase, pattern_ufcs, &code); + } + } + if pattern.kind() != code.kind() { + fail_match!( + "Pattern had `{}` ({:?}), code had `{}` ({:?})", + pattern.text(), + pattern.kind(), + code.text(), + code.kind() + ); + } + // Some kinds of nodes have special handling. For everything else, we fall back to default + // matching. + match code.kind() { + SyntaxKind::RECORD_EXPR_FIELD_LIST => { + self.attempt_match_record_field_list(phase, pattern, code) + } + SyntaxKind::TOKEN_TREE => self.attempt_match_token_tree(phase, pattern, code), + SyntaxKind::PATH => self.attempt_match_path(phase, pattern, code), + _ => self.attempt_match_node_children(phase, pattern, code), + } + } + + fn attempt_match_node_children( + &self, + phase: &mut Phase, + pattern: &SyntaxNode, + code: &SyntaxNode, + ) -> Result<(), MatchFailed> { + self.attempt_match_sequences( + phase, + PatternIterator::new(pattern), + code.children_with_tokens(), + ) + } + + fn attempt_match_sequences( + &self, + phase: &mut Phase, + pattern_it: PatternIterator, + mut code_it: SyntaxElementChildren, + ) -> Result<(), MatchFailed> { + let mut pattern_it = pattern_it.peekable(); + loop { + match phase.next_non_trivial(&mut code_it) { + None => { + if let Some(p) = pattern_it.next() { + fail_match!("Part of the pattern was unmatched: {:?}", p); + } + return Ok(()); + } + Some(SyntaxElement::Token(c)) => { + self.attempt_match_token(phase, &mut pattern_it, &c)?; + } + Some(SyntaxElement::Node(c)) => match pattern_it.next() { + Some(SyntaxElement::Node(p)) => { + self.attempt_match_node(phase, &p, &c)?; + } + Some(p) => fail_match!("Pattern wanted '{}', code has {}", p, c.text()), + None => fail_match!("Pattern reached end, code has {}", c.text()), + }, + } + } + } + + fn attempt_match_token( + &self, + phase: &mut Phase, + pattern: &mut Peekable, + code: &syntax::SyntaxToken, + ) -> Result<(), MatchFailed> { + phase.record_ignored_comments(code); + // Ignore whitespace and comments. + if code.kind().is_trivia() { + return Ok(()); + } + if let Some(SyntaxElement::Token(p)) = pattern.peek() { + // If the code has a comma and the pattern is about to close something, then accept the + // comma without advancing the pattern. i.e. ignore trailing commas. + if code.kind() == SyntaxKind::COMMA && is_closing_token(p.kind()) { + return Ok(()); + } + // Conversely, if the pattern has a comma and the code doesn't, skip that part of the + // pattern and continue to match the code. + if p.kind() == SyntaxKind::COMMA && is_closing_token(code.kind()) { + pattern.next(); + } + } + // Consume an element from the pattern and make sure it matches. + match pattern.next() { + Some(SyntaxElement::Token(p)) => { + if p.kind() != code.kind() || p.text() != code.text() { + fail_match!( + "Pattern wanted token '{}' ({:?}), but code had token '{}' ({:?})", + p.text(), + p.kind(), + code.text(), + code.kind() + ) + } + } + Some(SyntaxElement::Node(p)) => { + // Not sure if this is actually reachable. + fail_match!( + "Pattern wanted {:?}, but code had token '{}' ({:?})", + p, + code.text(), + code.kind() + ); + } + None => { + fail_match!("Pattern exhausted, while code remains: `{}`", code.text()); + } + } + Ok(()) + } + + fn check_constraint( + &self, + constraint: &Constraint, + code: &SyntaxNode, + ) -> Result<(), MatchFailed> { + match constraint { + Constraint::Kind(kind) => { + kind.matches(code)?; + } + Constraint::Not(sub) => { + if self.check_constraint(&*sub, code).is_ok() { + fail_match!("Constraint {:?} failed for '{}'", constraint, code.text()); + } + } + } + Ok(()) + } + + /// Paths are matched based on whether they refer to the same thing, even if they're written + /// differently. + fn attempt_match_path( + &self, + phase: &mut Phase, + pattern: &SyntaxNode, + code: &SyntaxNode, + ) -> Result<(), MatchFailed> { + if let Some(pattern_resolved) = self.rule.pattern.resolved_paths.get(pattern) { + let pattern_path = ast::Path::cast(pattern.clone()).unwrap(); + let code_path = ast::Path::cast(code.clone()).unwrap(); + if let (Some(pattern_segment), Some(code_segment)) = + (pattern_path.segment(), code_path.segment()) + { + // Match everything within the segment except for the name-ref, which is handled + // separately via comparing what the path resolves to below. + self.attempt_match_opt( + phase, + pattern_segment.generic_arg_list(), + code_segment.generic_arg_list(), + )?; + self.attempt_match_opt( + phase, + pattern_segment.param_list(), + code_segment.param_list(), + )?; + } + if matches!(phase, Phase::Second(_)) { + let resolution = self + .sema + .resolve_path(&code_path) + .ok_or_else(|| match_error!("Failed to resolve path `{}`", code.text()))?; + if pattern_resolved.resolution != resolution { + fail_match!("Pattern had path `{}` code had `{}`", pattern.text(), code.text()); + } + } + } else { + return self.attempt_match_node_children(phase, pattern, code); + } + Ok(()) + } + + fn attempt_match_opt( + &self, + phase: &mut Phase, + pattern: Option, + code: Option, + ) -> Result<(), MatchFailed> { + match (pattern, code) { + (Some(p), Some(c)) => self.attempt_match_node(phase, &p.syntax(), &c.syntax()), + (None, None) => Ok(()), + (Some(p), None) => fail_match!("Pattern `{}` had nothing to match", p.syntax().text()), + (None, Some(c)) => { + fail_match!("Nothing in pattern to match code `{}`", c.syntax().text()) + } + } + } + + /// We want to allow the records to match in any order, so we have special matching logic for + /// them. + fn attempt_match_record_field_list( + &self, + phase: &mut Phase, + pattern: &SyntaxNode, + code: &SyntaxNode, + ) -> Result<(), MatchFailed> { + // Build a map keyed by field name. + let mut fields_by_name = FxHashMap::default(); + for child in code.children() { + if let Some(record) = ast::RecordExprField::cast(child.clone()) { + if let Some(name) = record.field_name() { + fields_by_name.insert(name.text().clone(), child.clone()); + } + } + } + for p in pattern.children_with_tokens() { + if let SyntaxElement::Node(p) = p { + if let Some(name_element) = p.first_child_or_token() { + if self.get_placeholder(&name_element).is_some() { + // If the pattern is using placeholders for field names then order + // independence doesn't make sense. Fall back to regular ordered + // matching. + return self.attempt_match_node_children(phase, pattern, code); + } + if let Some(ident) = only_ident(name_element) { + let code_record = fields_by_name.remove(ident.text()).ok_or_else(|| { + match_error!( + "Placeholder has record field '{}', but code doesn't", + ident + ) + })?; + self.attempt_match_node(phase, &p, &code_record)?; + } + } + } + } + if let Some(unmatched_fields) = fields_by_name.keys().next() { + fail_match!( + "{} field(s) of a record literal failed to match, starting with {}", + fields_by_name.len(), + unmatched_fields + ); + } + Ok(()) + } + + /// Outside of token trees, a placeholder can only match a single AST node, whereas in a token + /// tree it can match a sequence of tokens. Note, that this code will only be used when the + /// pattern matches the macro invocation. For matches within the macro call, we'll already have + /// expanded the macro. + fn attempt_match_token_tree( + &self, + phase: &mut Phase, + pattern: &SyntaxNode, + code: &syntax::SyntaxNode, + ) -> Result<(), MatchFailed> { + let mut pattern = PatternIterator::new(pattern).peekable(); + let mut children = code.children_with_tokens(); + while let Some(child) = children.next() { + if let Some(placeholder) = pattern.peek().and_then(|p| self.get_placeholder(p)) { + pattern.next(); + let next_pattern_token = pattern + .peek() + .and_then(|p| match p { + SyntaxElement::Token(t) => Some(t.clone()), + SyntaxElement::Node(n) => n.first_token(), + }) + .map(|p| p.text().to_string()); + let first_matched_token = child.clone(); + let mut last_matched_token = child; + // Read code tokens util we reach one equal to the next token from our pattern + // or we reach the end of the token tree. + while let Some(next) = children.next() { + match &next { + SyntaxElement::Token(t) => { + if Some(t.to_string()) == next_pattern_token { + pattern.next(); + break; + } + } + SyntaxElement::Node(n) => { + if let Some(first_token) = n.first_token() { + if Some(first_token.to_string()) == next_pattern_token { + if let Some(SyntaxElement::Node(p)) = pattern.next() { + // We have a subtree that starts with the next token in our pattern. + self.attempt_match_token_tree(phase, &p, &n)?; + break; + } + } + } + } + }; + last_matched_token = next; + } + if let Phase::Second(match_out) = phase { + match_out.placeholder_values.insert( + Var(placeholder.ident.to_string()), + PlaceholderMatch::from_range(FileRange { + file_id: self.sema.original_range(code).file_id, + range: first_matched_token + .text_range() + .cover(last_matched_token.text_range()), + }), + ); + } + continue; + } + // Match literal (non-placeholder) tokens. + match child { + SyntaxElement::Token(token) => { + self.attempt_match_token(phase, &mut pattern, &token)?; + } + SyntaxElement::Node(node) => match pattern.next() { + Some(SyntaxElement::Node(p)) => { + self.attempt_match_token_tree(phase, &p, &node)?; + } + Some(SyntaxElement::Token(p)) => fail_match!( + "Pattern has token '{}', code has subtree '{}'", + p.text(), + node.text() + ), + None => fail_match!("Pattern has nothing, code has '{}'", node.text()), + }, + } + } + if let Some(p) = pattern.next() { + fail_match!("Reached end of token tree in code, but pattern still has {:?}", p); + } + Ok(()) + } + + fn attempt_match_ufcs_to_method_call( + &self, + phase: &mut Phase, + pattern_ufcs: &UfcsCallInfo, + code: &ast::MethodCallExpr, + ) -> Result<(), MatchFailed> { + use ast::ArgListOwner; + let code_resolved_function = self + .sema + .resolve_method_call(code) + .ok_or_else(|| match_error!("Failed to resolve method call"))?; + if pattern_ufcs.function != code_resolved_function { + fail_match!("Method call resolved to a different function"); + } + if code_resolved_function.has_self_param(self.sema.db) { + if let (Some(pattern_type), Some(expr)) = (&pattern_ufcs.qualifier_type, &code.expr()) { + self.check_expr_type(pattern_type, expr)?; + } + } + // Check arguments. + let mut pattern_args = pattern_ufcs + .call_expr + .arg_list() + .ok_or_else(|| match_error!("Pattern function call has no args"))? + .args(); + self.attempt_match_opt(phase, pattern_args.next(), code.expr())?; + let mut code_args = + code.arg_list().ok_or_else(|| match_error!("Code method call has no args"))?.args(); + loop { + match (pattern_args.next(), code_args.next()) { + (None, None) => return Ok(()), + (p, c) => self.attempt_match_opt(phase, p, c)?, + } + } + } + + fn attempt_match_ufcs_to_ufcs( + &self, + phase: &mut Phase, + pattern_ufcs: &UfcsCallInfo, + code: &ast::CallExpr, + ) -> Result<(), MatchFailed> { + use ast::ArgListOwner; + // Check that the first argument is the expected type. + if let (Some(pattern_type), Some(expr)) = ( + &pattern_ufcs.qualifier_type, + &code.arg_list().and_then(|code_args| code_args.args().next()), + ) { + self.check_expr_type(pattern_type, expr)?; + } + self.attempt_match_node_children(phase, pattern_ufcs.call_expr.syntax(), code.syntax()) + } + + fn check_expr_type( + &self, + pattern_type: &hir::Type, + expr: &ast::Expr, + ) -> Result<(), MatchFailed> { + use hir::HirDisplay; + let code_type = self.sema.type_of_expr(&expr).ok_or_else(|| { + match_error!("Failed to get receiver type for `{}`", expr.syntax().text()) + })?; + if !code_type + .autoderef(self.sema.db) + .any(|deref_code_type| *pattern_type == deref_code_type) + { + fail_match!( + "Pattern type `{}` didn't match code type `{}`", + pattern_type.display(self.sema.db), + code_type.display(self.sema.db) + ); + } + Ok(()) + } + + fn get_placeholder(&self, element: &SyntaxElement) -> Option<&Placeholder> { + only_ident(element.clone()).and_then(|ident| self.rule.get_placeholder(&ident)) + } +} + +impl Match { + fn render_template_paths( + &mut self, + template: &ResolvedPattern, + sema: &Semantics, + ) -> Result<(), MatchFailed> { + let module = sema + .scope(&self.matched_node) + .module() + .ok_or_else(|| match_error!("Matched node isn't in a module"))?; + for (path, resolved_path) in &template.resolved_paths { + if let hir::PathResolution::Def(module_def) = resolved_path.resolution { + let mod_path = module.find_use_path(sema.db, module_def).ok_or_else(|| { + match_error!("Failed to render template path `{}` at match location") + })?; + self.rendered_template_paths.insert(path.clone(), mod_path); + } + } + Ok(()) + } +} + +impl Phase<'_> { + fn next_non_trivial(&mut self, code_it: &mut SyntaxElementChildren) -> Option { + loop { + let c = code_it.next(); + if let Some(SyntaxElement::Token(t)) = &c { + self.record_ignored_comments(t); + if t.kind().is_trivia() { + continue; + } + } + return c; + } + } + + fn record_ignored_comments(&mut self, token: &SyntaxToken) { + if token.kind() == SyntaxKind::COMMENT { + if let Phase::Second(match_out) = self { + if let Some(comment) = ast::Comment::cast(token.clone()) { + match_out.ignored_comments.push(comment); + } + } + } + } +} + +fn is_closing_token(kind: SyntaxKind) -> bool { + kind == SyntaxKind::R_PAREN || kind == SyntaxKind::R_CURLY || kind == SyntaxKind::R_BRACK +} + +pub(crate) fn record_match_fails_reasons_scope(debug_active: bool, f: F) -> T +where + F: Fn() -> T, +{ + RECORDING_MATCH_FAIL_REASONS.with(|c| c.set(debug_active)); + let res = f(); + RECORDING_MATCH_FAIL_REASONS.with(|c| c.set(false)); + res +} + +// For performance reasons, we don't want to record the reason why every match fails, only the bit +// of code that the user indicated they thought would match. We use a thread local to indicate when +// we are trying to match that bit of code. This saves us having to pass a boolean into all the bits +// of code that can make the decision to not match. +thread_local! { + pub static RECORDING_MATCH_FAIL_REASONS: Cell = Cell::new(false); +} + +fn recording_match_fail_reasons() -> bool { + RECORDING_MATCH_FAIL_REASONS.with(|c| c.get()) +} + +impl PlaceholderMatch { + fn new(node: &SyntaxNode, range: FileRange) -> Self { + Self { node: Some(node.clone()), range, inner_matches: SsrMatches::default() } + } + + fn from_range(range: FileRange) -> Self { + Self { node: None, range, inner_matches: SsrMatches::default() } + } +} + +impl NodeKind { + fn matches(&self, node: &SyntaxNode) -> Result<(), MatchFailed> { + let ok = match self { + Self::Literal => { + mark::hit!(literal_constraint); + ast::Literal::can_cast(node.kind()) + } + }; + if !ok { + fail_match!("Code '{}' isn't of kind {:?}", node.text(), self); + } + Ok(()) + } +} + +// If `node` contains nothing but an ident then return it, otherwise return None. +fn only_ident(element: SyntaxElement) -> Option { + match element { + SyntaxElement::Token(t) => { + if t.kind() == SyntaxKind::IDENT { + return Some(t); + } + } + SyntaxElement::Node(n) => { + let mut children = n.children_with_tokens(); + if let (Some(only_child), None) = (children.next(), children.next()) { + return only_ident(only_child); + } + } + } + None +} + +struct PatternIterator { + iter: SyntaxElementChildren, +} + +impl Iterator for PatternIterator { + type Item = SyntaxElement; + + fn next(&mut self) -> Option { + while let Some(element) = self.iter.next() { + if !element.kind().is_trivia() { + return Some(element); + } + } + None + } +} + +impl PatternIterator { + fn new(parent: &SyntaxNode) -> Self { + Self { iter: parent.children_with_tokens() } + } +} + +#[cfg(test)] +mod tests { + use super::*; + use crate::{MatchFinder, SsrRule}; + + #[test] + fn parse_match_replace() { + let rule: SsrRule = "foo($x) ==>> bar($x)".parse().unwrap(); + let input = "fn foo() {} fn bar() {} fn main() { foo(1+2); }"; + + let (db, position, selections) = crate::tests::single_file(input); + let mut match_finder = MatchFinder::in_context(&db, position, selections); + match_finder.add_rule(rule).unwrap(); + let matches = match_finder.matches(); + assert_eq!(matches.matches.len(), 1); + assert_eq!(matches.matches[0].matched_node.text(), "foo(1+2)"); + assert_eq!(matches.matches[0].placeholder_values.len(), 1); + assert_eq!( + matches.matches[0].placeholder_values[&Var("x".to_string())] + .node + .as_ref() + .unwrap() + .text(), + "1+2" + ); + + let edits = match_finder.edits(); + assert_eq!(edits.len(), 1); + let edit = &edits[0]; + let mut after = input.to_string(); + edit.edit.apply(&mut after); + assert_eq!(after, "fn foo() {} fn bar() {} fn main() { bar(1+2); }"); + } +} diff --git a/crates/ssr/src/nester.rs b/crates/ssr/src/nester.rs new file mode 100644 index 000000000..6ac355dfc --- /dev/null +++ b/crates/ssr/src/nester.rs @@ -0,0 +1,94 @@ +//! Converts a flat collection of matches into a nested form suitable for replacement. When there +//! are multiple matches for a node, or that overlap, priority is given to the earlier rule. Nested +//! matches are only permitted if the inner match is contained entirely within a placeholder of an +//! outer match. +//! +//! For example, if our search pattern is `foo(foo($a))` and the code had `foo(foo(foo(foo(42))))`, +//! then we'll get 3 matches, however only the outermost and innermost matches can be accepted. The +//! middle match would take the second `foo` from the outer match. + +use crate::{Match, SsrMatches}; +use rustc_hash::FxHashMap; +use syntax::SyntaxNode; + +pub(crate) fn nest_and_remove_collisions( + mut matches: Vec, + sema: &hir::Semantics, +) -> SsrMatches { + // We sort the matches by depth then by rule index. Sorting by depth means that by the time we + // see a match, any parent matches or conflicting matches will have already been seen. Sorting + // by rule_index means that if there are two matches for the same node, the rule added first + // will take precedence. + matches.sort_by(|a, b| a.depth.cmp(&b.depth).then_with(|| a.rule_index.cmp(&b.rule_index))); + let mut collector = MatchCollector::default(); + for m in matches { + collector.add_match(m, sema); + } + collector.into() +} + +#[derive(Default)] +struct MatchCollector { + matches_by_node: FxHashMap, +} + +impl MatchCollector { + /// Attempts to add `m` to matches. If it conflicts with an existing match, it is discarded. If + /// it is entirely within the a placeholder of an existing match, then it is added as a child + /// match of the existing match. + fn add_match(&mut self, m: Match, sema: &hir::Semantics) { + let matched_node = m.matched_node.clone(); + if let Some(existing) = self.matches_by_node.get_mut(&matched_node) { + try_add_sub_match(m, existing, sema); + return; + } + for ancestor in sema.ancestors_with_macros(m.matched_node.clone()) { + if let Some(existing) = self.matches_by_node.get_mut(&ancestor) { + try_add_sub_match(m, existing, sema); + return; + } + } + self.matches_by_node.insert(matched_node, m); + } +} + +/// Attempts to add `m` as a sub-match of `existing`. +fn try_add_sub_match(m: Match, existing: &mut Match, sema: &hir::Semantics) { + for p in existing.placeholder_values.values_mut() { + // Note, no need to check if p.range.file is equal to m.range.file, since we + // already know we're within `existing`. + if p.range.range.contains_range(m.range.range) { + // Convert the inner matches in `p` into a temporary MatchCollector. When + // we're done, we then convert it back into an SsrMatches. If we expected + // lots of inner matches, it might be worthwhile keeping a MatchCollector + // around for each placeholder match. However we expect most placeholder + // will have 0 and a few will have 1. More than that should hopefully be + // exceptional. + let mut collector = MatchCollector::default(); + for m in std::mem::replace(&mut p.inner_matches.matches, Vec::new()) { + collector.matches_by_node.insert(m.matched_node.clone(), m); + } + collector.add_match(m, sema); + p.inner_matches = collector.into(); + break; + } + } +} + +impl From for SsrMatches { + fn from(mut match_collector: MatchCollector) -> Self { + let mut matches = SsrMatches::default(); + for (_, m) in match_collector.matches_by_node.drain() { + matches.matches.push(m); + } + matches.matches.sort_by(|a, b| { + // Order matches by file_id then by start range. This should be sufficient since ranges + // shouldn't be overlapping. + a.range + .file_id + .cmp(&b.range.file_id) + .then_with(|| a.range.range.start().cmp(&b.range.range.start())) + }); + matches + } +} diff --git a/crates/ssr/src/parsing.rs b/crates/ssr/src/parsing.rs new file mode 100644 index 000000000..9570e96e3 --- /dev/null +++ b/crates/ssr/src/parsing.rs @@ -0,0 +1,389 @@ +//! This file contains code for parsing SSR rules, which look something like `foo($a) ==>> bar($b)`. +//! We first split everything before and after the separator `==>>`. Next, both the search pattern +//! and the replacement template get tokenized by the Rust tokenizer. Tokens are then searched for +//! placeholders, which start with `$`. For replacement templates, this is the final form. For +//! search patterns, we go further and parse the pattern as each kind of thing that we can match. +//! e.g. expressions, type references etc. + +use crate::errors::bail; +use crate::{SsrError, SsrPattern, SsrRule}; +use rustc_hash::{FxHashMap, FxHashSet}; +use std::str::FromStr; +use syntax::{ast, AstNode, SmolStr, SyntaxKind, SyntaxNode, T}; +use test_utils::mark; + +#[derive(Debug)] +pub(crate) struct ParsedRule { + pub(crate) placeholders_by_stand_in: FxHashMap, + pub(crate) pattern: SyntaxNode, + pub(crate) template: Option, +} + +#[derive(Debug)] +pub(crate) struct RawPattern { + tokens: Vec, +} + +// Part of a search or replace pattern. +#[derive(Clone, Debug, PartialEq, Eq)] +pub(crate) enum PatternElement { + Token(Token), + Placeholder(Placeholder), +} + +#[derive(Clone, Debug, PartialEq, Eq)] +pub(crate) struct Placeholder { + /// The name of this placeholder. e.g. for "$a", this would be "a" + pub(crate) ident: SmolStr, + /// A unique name used in place of this placeholder when we parse the pattern as Rust code. + stand_in_name: String, + pub(crate) constraints: Vec, +} + +#[derive(Clone, Debug, PartialEq, Eq)] +pub(crate) enum Constraint { + Kind(NodeKind), + Not(Box), +} + +#[derive(Clone, Debug, PartialEq, Eq)] +pub(crate) enum NodeKind { + Literal, +} + +#[derive(Debug, Clone, PartialEq, Eq)] +pub(crate) struct Token { + kind: SyntaxKind, + pub(crate) text: SmolStr, +} + +impl ParsedRule { + fn new( + pattern: &RawPattern, + template: Option<&RawPattern>, + ) -> Result, SsrError> { + let raw_pattern = pattern.as_rust_code(); + let raw_template = template.map(|t| t.as_rust_code()); + let raw_template = raw_template.as_ref().map(|s| s.as_str()); + let mut builder = RuleBuilder { + placeholders_by_stand_in: pattern.placeholders_by_stand_in(), + rules: Vec::new(), + }; + builder.try_add(ast::Expr::parse(&raw_pattern), raw_template.map(ast::Expr::parse)); + builder.try_add(ast::Type::parse(&raw_pattern), raw_template.map(ast::Type::parse)); + builder.try_add(ast::Item::parse(&raw_pattern), raw_template.map(ast::Item::parse)); + builder.try_add(ast::Path::parse(&raw_pattern), raw_template.map(ast::Path::parse)); + builder.try_add(ast::Pat::parse(&raw_pattern), raw_template.map(ast::Pat::parse)); + builder.build() + } +} + +struct RuleBuilder { + placeholders_by_stand_in: FxHashMap, + rules: Vec, +} + +impl RuleBuilder { + fn try_add(&mut self, pattern: Result, template: Option>) { + match (pattern, template) { + (Ok(pattern), Some(Ok(template))) => self.rules.push(ParsedRule { + placeholders_by_stand_in: self.placeholders_by_stand_in.clone(), + pattern: pattern.syntax().clone(), + template: Some(template.syntax().clone()), + }), + (Ok(pattern), None) => self.rules.push(ParsedRule { + placeholders_by_stand_in: self.placeholders_by_stand_in.clone(), + pattern: pattern.syntax().clone(), + template: None, + }), + _ => {} + } + } + + fn build(mut self) -> Result, SsrError> { + if self.rules.is_empty() { + bail!("Not a valid Rust expression, type, item, path or pattern"); + } + // If any rules contain paths, then we reject any rules that don't contain paths. Allowing a + // mix leads to strange semantics, since the path-based rules only match things where the + // path refers to semantically the same thing, whereas the non-path-based rules could match + // anything. Specifically, if we have a rule like `foo ==>> bar` we only want to match the + // `foo` that is in the current scope, not any `foo`. However "foo" can be parsed as a + // pattern (IDENT_PAT -> NAME -> IDENT). Allowing such a rule through would result in + // renaming everything called `foo` to `bar`. It'd also be slow, since without a path, we'd + // have to use the slow-scan search mechanism. + if self.rules.iter().any(|rule| contains_path(&rule.pattern)) { + let old_len = self.rules.len(); + self.rules.retain(|rule| contains_path(&rule.pattern)); + if self.rules.len() < old_len { + mark::hit!(pattern_is_a_single_segment_path); + } + } + Ok(self.rules) + } +} + +/// Returns whether there are any paths in `node`. +fn contains_path(node: &SyntaxNode) -> bool { + node.kind() == SyntaxKind::PATH + || node.descendants().any(|node| node.kind() == SyntaxKind::PATH) +} + +impl FromStr for SsrRule { + type Err = SsrError; + + fn from_str(query: &str) -> Result { + let mut it = query.split("==>>"); + let pattern = it.next().expect("at least empty string").trim(); + let template = it + .next() + .ok_or_else(|| SsrError("Cannot find delimiter `==>>`".into()))? + .trim() + .to_string(); + if it.next().is_some() { + return Err(SsrError("More than one delimiter found".into())); + } + let raw_pattern = pattern.parse()?; + let raw_template = template.parse()?; + let parsed_rules = ParsedRule::new(&raw_pattern, Some(&raw_template))?; + let rule = SsrRule { pattern: raw_pattern, template: raw_template, parsed_rules }; + validate_rule(&rule)?; + Ok(rule) + } +} + +impl FromStr for RawPattern { + type Err = SsrError; + + fn from_str(pattern_str: &str) -> Result { + Ok(RawPattern { tokens: parse_pattern(pattern_str)? }) + } +} + +impl RawPattern { + /// Returns this search pattern as Rust source code that we can feed to the Rust parser. + fn as_rust_code(&self) -> String { + let mut res = String::new(); + for t in &self.tokens { + res.push_str(match t { + PatternElement::Token(token) => token.text.as_str(), + PatternElement::Placeholder(placeholder) => placeholder.stand_in_name.as_str(), + }); + } + res + } + + pub(crate) fn placeholders_by_stand_in(&self) -> FxHashMap { + let mut res = FxHashMap::default(); + for t in &self.tokens { + if let PatternElement::Placeholder(placeholder) = t { + res.insert(SmolStr::new(placeholder.stand_in_name.clone()), placeholder.clone()); + } + } + res + } +} + +impl FromStr for SsrPattern { + type Err = SsrError; + + fn from_str(pattern_str: &str) -> Result { + let raw_pattern = pattern_str.parse()?; + let parsed_rules = ParsedRule::new(&raw_pattern, None)?; + Ok(SsrPattern { raw: raw_pattern, parsed_rules }) + } +} + +/// Returns `pattern_str`, parsed as a search or replace pattern. If `remove_whitespace` is true, +/// then any whitespace tokens will be removed, which we do for the search pattern, but not for the +/// replace pattern. +fn parse_pattern(pattern_str: &str) -> Result, SsrError> { + let mut res = Vec::new(); + let mut placeholder_names = FxHashSet::default(); + let mut tokens = tokenize(pattern_str)?.into_iter(); + while let Some(token) = tokens.next() { + if token.kind == T![$] { + let placeholder = parse_placeholder(&mut tokens)?; + if !placeholder_names.insert(placeholder.ident.clone()) { + bail!("Name `{}` repeats more than once", placeholder.ident); + } + res.push(PatternElement::Placeholder(placeholder)); + } else { + res.push(PatternElement::Token(token)); + } + } + Ok(res) +} + +/// Checks for errors in a rule. e.g. the replace pattern referencing placeholders that the search +/// pattern didn't define. +fn validate_rule(rule: &SsrRule) -> Result<(), SsrError> { + let mut defined_placeholders = FxHashSet::default(); + for p in &rule.pattern.tokens { + if let PatternElement::Placeholder(placeholder) = p { + defined_placeholders.insert(&placeholder.ident); + } + } + let mut undefined = Vec::new(); + for p in &rule.template.tokens { + if let PatternElement::Placeholder(placeholder) = p { + if !defined_placeholders.contains(&placeholder.ident) { + undefined.push(format!("${}", placeholder.ident)); + } + if !placeholder.constraints.is_empty() { + bail!("Replacement placeholders cannot have constraints"); + } + } + } + if !undefined.is_empty() { + bail!("Replacement contains undefined placeholders: {}", undefined.join(", ")); + } + Ok(()) +} + +fn tokenize(source: &str) -> Result, SsrError> { + let mut start = 0; + let (raw_tokens, errors) = syntax::tokenize(source); + if let Some(first_error) = errors.first() { + bail!("Failed to parse pattern: {}", first_error); + } + let mut tokens: Vec = Vec::new(); + for raw_token in raw_tokens { + let token_len = usize::from(raw_token.len); + tokens.push(Token { + kind: raw_token.kind, + text: SmolStr::new(&source[start..start + token_len]), + }); + start += token_len; + } + Ok(tokens) +} + +fn parse_placeholder(tokens: &mut std::vec::IntoIter) -> Result { + let mut name = None; + let mut constraints = Vec::new(); + if let Some(token) = tokens.next() { + match token.kind { + SyntaxKind::IDENT => { + name = Some(token.text); + } + T!['{'] => { + let token = + tokens.next().ok_or_else(|| SsrError::new("Unexpected end of placeholder"))?; + if token.kind == SyntaxKind::IDENT { + name = Some(token.text); + } + loop { + let token = tokens + .next() + .ok_or_else(|| SsrError::new("Placeholder is missing closing brace '}'"))?; + match token.kind { + T![:] => { + constraints.push(parse_constraint(tokens)?); + } + T!['}'] => break, + _ => bail!("Unexpected token while parsing placeholder: '{}'", token.text), + } + } + } + _ => { + bail!("Placeholders should either be $name or ${{name:constraints}}"); + } + } + } + let name = name.ok_or_else(|| SsrError::new("Placeholder ($) with no name"))?; + Ok(Placeholder::new(name, constraints)) +} + +fn parse_constraint(tokens: &mut std::vec::IntoIter) -> Result { + let constraint_type = tokens + .next() + .ok_or_else(|| SsrError::new("Found end of placeholder while looking for a constraint"))? + .text + .to_string(); + match constraint_type.as_str() { + "kind" => { + expect_token(tokens, "(")?; + let t = tokens.next().ok_or_else(|| { + SsrError::new("Unexpected end of constraint while looking for kind") + })?; + if t.kind != SyntaxKind::IDENT { + bail!("Expected ident, found {:?} while parsing kind constraint", t.kind); + } + expect_token(tokens, ")")?; + Ok(Constraint::Kind(NodeKind::from(&t.text)?)) + } + "not" => { + expect_token(tokens, "(")?; + let sub = parse_constraint(tokens)?; + expect_token(tokens, ")")?; + Ok(Constraint::Not(Box::new(sub))) + } + x => bail!("Unsupported constraint type '{}'", x), + } +} + +fn expect_token(tokens: &mut std::vec::IntoIter, expected: &str) -> Result<(), SsrError> { + if let Some(t) = tokens.next() { + if t.text == expected { + return Ok(()); + } + bail!("Expected {} found {}", expected, t.text); + } + bail!("Expected {} found end of stream", expected); +} + +impl NodeKind { + fn from(name: &SmolStr) -> Result { + Ok(match name.as_str() { + "literal" => NodeKind::Literal, + _ => bail!("Unknown node kind '{}'", name), + }) + } +} + +impl Placeholder { + fn new(name: SmolStr, constraints: Vec) -> Self { + Self { stand_in_name: format!("__placeholder_{}", name), constraints, ident: name } + } +} + +#[cfg(test)] +mod tests { + use super::*; + + #[test] + fn parser_happy_case() { + fn token(kind: SyntaxKind, text: &str) -> PatternElement { + PatternElement::Token(Token { kind, text: SmolStr::new(text) }) + } + fn placeholder(name: &str) -> PatternElement { + PatternElement::Placeholder(Placeholder::new(SmolStr::new(name), Vec::new())) + } + let result: SsrRule = "foo($a, $b) ==>> bar($b, $a)".parse().unwrap(); + assert_eq!( + result.pattern.tokens, + vec![ + token(SyntaxKind::IDENT, "foo"), + token(T!['('], "("), + placeholder("a"), + token(T![,], ","), + token(SyntaxKind::WHITESPACE, " "), + placeholder("b"), + token(T![')'], ")"), + ] + ); + assert_eq!( + result.template.tokens, + vec![ + token(SyntaxKind::IDENT, "bar"), + token(T!['('], "("), + placeholder("b"), + token(T![,], ","), + token(SyntaxKind::WHITESPACE, " "), + placeholder("a"), + token(T![')'], ")"), + ] + ); + } +} diff --git a/crates/ssr/src/replacing.rs b/crates/ssr/src/replacing.rs new file mode 100644 index 000000000..8f8fe6149 --- /dev/null +++ b/crates/ssr/src/replacing.rs @@ -0,0 +1,194 @@ +//! Code for applying replacement templates for matches that have previously been found. + +use crate::matching::Var; +use crate::{resolving::ResolvedRule, Match, SsrMatches}; +use rustc_hash::{FxHashMap, FxHashSet}; +use syntax::ast::{self, AstToken}; +use syntax::{SyntaxElement, SyntaxKind, SyntaxNode, SyntaxToken, TextRange, TextSize}; +use text_edit::TextEdit; + +/// Returns a text edit that will replace each match in `matches` with its corresponding replacement +/// template. Placeholders in the template will have been substituted with whatever they matched to +/// in the original code. +pub(crate) fn matches_to_edit( + matches: &SsrMatches, + file_src: &str, + rules: &[ResolvedRule], +) -> TextEdit { + matches_to_edit_at_offset(matches, file_src, 0.into(), rules) +} + +fn matches_to_edit_at_offset( + matches: &SsrMatches, + file_src: &str, + relative_start: TextSize, + rules: &[ResolvedRule], +) -> TextEdit { + let mut edit_builder = TextEdit::builder(); + for m in &matches.matches { + edit_builder.replace( + m.range.range.checked_sub(relative_start).unwrap(), + render_replace(m, file_src, rules), + ); + } + edit_builder.finish() +} + +struct ReplacementRenderer<'a> { + match_info: &'a Match, + file_src: &'a str, + rules: &'a [ResolvedRule], + rule: &'a ResolvedRule, + out: String, + // Map from a range within `out` to a token in `template` that represents a placeholder. This is + // used to validate that the generated source code doesn't split any placeholder expansions (see + // below). + placeholder_tokens_by_range: FxHashMap, + // Which placeholder tokens need to be wrapped in parenthesis in order to ensure that when `out` + // is parsed, placeholders don't get split. e.g. if a template of `$a.to_string()` results in `1 + // + 2.to_string()` then the placeholder value `1 + 2` was split and needs parenthesis. + placeholder_tokens_requiring_parenthesis: FxHashSet, +} + +fn render_replace(match_info: &Match, file_src: &str, rules: &[ResolvedRule]) -> String { + let rule = &rules[match_info.rule_index]; + let template = rule + .template + .as_ref() + .expect("You called MatchFinder::edits after calling MatchFinder::add_search_pattern"); + let mut renderer = ReplacementRenderer { + match_info, + file_src, + rules, + rule, + out: String::new(), + placeholder_tokens_requiring_parenthesis: FxHashSet::default(), + placeholder_tokens_by_range: FxHashMap::default(), + }; + renderer.render_node(&template.node); + renderer.maybe_rerender_with_extra_parenthesis(&template.node); + for comment in &match_info.ignored_comments { + renderer.out.push_str(&comment.syntax().to_string()); + } + renderer.out +} + +impl ReplacementRenderer<'_> { + fn render_node_children(&mut self, node: &SyntaxNode) { + for node_or_token in node.children_with_tokens() { + self.render_node_or_token(&node_or_token); + } + } + + fn render_node_or_token(&mut self, node_or_token: &SyntaxElement) { + match node_or_token { + SyntaxElement::Token(token) => { + self.render_token(&token); + } + SyntaxElement::Node(child_node) => { + self.render_node(&child_node); + } + } + } + + fn render_node(&mut self, node: &SyntaxNode) { + use syntax::ast::AstNode; + if let Some(mod_path) = self.match_info.rendered_template_paths.get(&node) { + self.out.push_str(&mod_path.to_string()); + // Emit everything except for the segment's name-ref, since we already effectively + // emitted that as part of `mod_path`. + if let Some(path) = ast::Path::cast(node.clone()) { + if let Some(segment) = path.segment() { + for node_or_token in segment.syntax().children_with_tokens() { + if node_or_token.kind() != SyntaxKind::NAME_REF { + self.render_node_or_token(&node_or_token); + } + } + } + } + } else { + self.render_node_children(&node); + } + } + + fn render_token(&mut self, token: &SyntaxToken) { + if let Some(placeholder) = self.rule.get_placeholder(&token) { + if let Some(placeholder_value) = + self.match_info.placeholder_values.get(&Var(placeholder.ident.to_string())) + { + let range = &placeholder_value.range.range; + let mut matched_text = + self.file_src[usize::from(range.start())..usize::from(range.end())].to_owned(); + let edit = matches_to_edit_at_offset( + &placeholder_value.inner_matches, + self.file_src, + range.start(), + self.rules, + ); + let needs_parenthesis = + self.placeholder_tokens_requiring_parenthesis.contains(token); + edit.apply(&mut matched_text); + if needs_parenthesis { + self.out.push('('); + } + self.placeholder_tokens_by_range.insert( + TextRange::new( + TextSize::of(&self.out), + TextSize::of(&self.out) + TextSize::of(&matched_text), + ), + token.clone(), + ); + self.out.push_str(&matched_text); + if needs_parenthesis { + self.out.push(')'); + } + } else { + // We validated that all placeholder references were valid before we + // started, so this shouldn't happen. + panic!( + "Internal error: replacement referenced unknown placeholder {}", + placeholder.ident + ); + } + } else { + self.out.push_str(token.text().as_str()); + } + } + + // Checks if the resulting code, when parsed doesn't split any placeholders due to different + // order of operations between the search pattern and the replacement template. If any do, then + // we rerender the template and wrap the problematic placeholders with parenthesis. + fn maybe_rerender_with_extra_parenthesis(&mut self, template: &SyntaxNode) { + if let Some(node) = parse_as_kind(&self.out, template.kind()) { + self.remove_node_ranges(node); + if self.placeholder_tokens_by_range.is_empty() { + return; + } + self.placeholder_tokens_requiring_parenthesis = + self.placeholder_tokens_by_range.values().cloned().collect(); + self.out.clear(); + self.render_node(template); + } + } + + fn remove_node_ranges(&mut self, node: SyntaxNode) { + self.placeholder_tokens_by_range.remove(&node.text_range()); + for child in node.children() { + self.remove_node_ranges(child); + } + } +} + +fn parse_as_kind(code: &str, kind: SyntaxKind) -> Option { + use syntax::ast::AstNode; + if ast::Expr::can_cast(kind) { + if let Ok(expr) = ast::Expr::parse(code) { + return Some(expr.syntax().clone()); + } + } else if ast::Item::can_cast(kind) { + if let Ok(item) = ast::Item::parse(code) { + return Some(item.syntax().clone()); + } + } + None +} diff --git a/crates/ssr/src/resolving.rs b/crates/ssr/src/resolving.rs new file mode 100644 index 000000000..020fd7994 --- /dev/null +++ b/crates/ssr/src/resolving.rs @@ -0,0 +1,299 @@ +//! This module is responsible for resolving paths within rules. + +use crate::errors::error; +use crate::{parsing, SsrError}; +use base_db::FilePosition; +use parsing::Placeholder; +use rustc_hash::FxHashMap; +use syntax::{ast, SmolStr, SyntaxKind, SyntaxNode, SyntaxToken}; +use test_utils::mark; + +pub(crate) struct ResolutionScope<'db> { + scope: hir::SemanticsScope<'db>, + hygiene: hir::Hygiene, + node: SyntaxNode, +} + +pub(crate) struct ResolvedRule { + pub(crate) pattern: ResolvedPattern, + pub(crate) template: Option, + pub(crate) index: usize, +} + +pub(crate) struct ResolvedPattern { + pub(crate) placeholders_by_stand_in: FxHashMap, + pub(crate) node: SyntaxNode, + // Paths in `node` that we've resolved. + pub(crate) resolved_paths: FxHashMap, + pub(crate) ufcs_function_calls: FxHashMap, + pub(crate) contains_self: bool, +} + +pub(crate) struct ResolvedPath { + pub(crate) resolution: hir::PathResolution, + /// The depth of the ast::Path that was resolved within the pattern. + pub(crate) depth: u32, +} + +pub(crate) struct UfcsCallInfo { + pub(crate) call_expr: ast::CallExpr, + pub(crate) function: hir::Function, + pub(crate) qualifier_type: Option, +} + +impl ResolvedRule { + pub(crate) fn new( + rule: parsing::ParsedRule, + resolution_scope: &ResolutionScope, + index: usize, + ) -> Result { + let resolver = + Resolver { resolution_scope, placeholders_by_stand_in: rule.placeholders_by_stand_in }; + let resolved_template = if let Some(template) = rule.template { + Some(resolver.resolve_pattern_tree(template)?) + } else { + None + }; + Ok(ResolvedRule { + pattern: resolver.resolve_pattern_tree(rule.pattern)?, + template: resolved_template, + index, + }) + } + + pub(crate) fn get_placeholder(&self, token: &SyntaxToken) -> Option<&Placeholder> { + if token.kind() != SyntaxKind::IDENT { + return None; + } + self.pattern.placeholders_by_stand_in.get(token.text()) + } +} + +struct Resolver<'a, 'db> { + resolution_scope: &'a ResolutionScope<'db>, + placeholders_by_stand_in: FxHashMap, +} + +impl Resolver<'_, '_> { + fn resolve_pattern_tree(&self, pattern: SyntaxNode) -> Result { + use syntax::ast::AstNode; + use syntax::{SyntaxElement, T}; + let mut resolved_paths = FxHashMap::default(); + self.resolve(pattern.clone(), 0, &mut resolved_paths)?; + let ufcs_function_calls = resolved_paths + .iter() + .filter_map(|(path_node, resolved)| { + if let Some(grandparent) = path_node.parent().and_then(|parent| parent.parent()) { + if let Some(call_expr) = ast::CallExpr::cast(grandparent.clone()) { + if let hir::PathResolution::AssocItem(hir::AssocItem::Function(function)) = + resolved.resolution + { + let qualifier_type = self.resolution_scope.qualifier_type(path_node); + return Some(( + grandparent, + UfcsCallInfo { call_expr, function, qualifier_type }, + )); + } + } + } + None + }) + .collect(); + let contains_self = + pattern.descendants_with_tokens().any(|node_or_token| match node_or_token { + SyntaxElement::Token(t) => t.kind() == T![self], + _ => false, + }); + Ok(ResolvedPattern { + node: pattern, + resolved_paths, + placeholders_by_stand_in: self.placeholders_by_stand_in.clone(), + ufcs_function_calls, + contains_self, + }) + } + + fn resolve( + &self, + node: SyntaxNode, + depth: u32, + resolved_paths: &mut FxHashMap, + ) -> Result<(), SsrError> { + use syntax::ast::AstNode; + if let Some(path) = ast::Path::cast(node.clone()) { + if is_self(&path) { + // Self cannot be resolved like other paths. + return Ok(()); + } + // Check if this is an appropriate place in the path to resolve. If the path is + // something like `a::B::::c` then we want to resolve `a::B`. If the path contains + // a placeholder. e.g. `a::$b::c` then we want to resolve `a`. + if !path_contains_type_arguments(path.qualifier()) + && !self.path_contains_placeholder(&path) + { + let resolution = self + .resolution_scope + .resolve_path(&path) + .ok_or_else(|| error!("Failed to resolve path `{}`", node.text()))?; + if self.ok_to_use_path_resolution(&resolution) { + resolved_paths.insert(node, ResolvedPath { resolution, depth }); + return Ok(()); + } + } + } + for node in node.children() { + self.resolve(node, depth + 1, resolved_paths)?; + } + Ok(()) + } + + /// Returns whether `path` contains a placeholder, but ignores any placeholders within type + /// arguments. + fn path_contains_placeholder(&self, path: &ast::Path) -> bool { + if let Some(segment) = path.segment() { + if let Some(name_ref) = segment.name_ref() { + if self.placeholders_by_stand_in.contains_key(name_ref.text()) { + return true; + } + } + } + if let Some(qualifier) = path.qualifier() { + return self.path_contains_placeholder(&qualifier); + } + false + } + + fn ok_to_use_path_resolution(&self, resolution: &hir::PathResolution) -> bool { + match resolution { + hir::PathResolution::AssocItem(hir::AssocItem::Function(function)) => { + if function.has_self_param(self.resolution_scope.scope.db) { + // If we don't use this path resolution, then we won't be able to match method + // calls. e.g. `Foo::bar($s)` should match `x.bar()`. + true + } else { + mark::hit!(replace_associated_trait_default_function_call); + false + } + } + hir::PathResolution::AssocItem(_) => { + // Not a function. Could be a constant or an associated type. + mark::hit!(replace_associated_trait_constant); + false + } + _ => true, + } + } +} + +impl<'db> ResolutionScope<'db> { + pub(crate) fn new( + sema: &hir::Semantics<'db, ide_db::RootDatabase>, + resolve_context: FilePosition, + ) -> ResolutionScope<'db> { + use syntax::ast::AstNode; + let file = sema.parse(resolve_context.file_id); + // Find a node at the requested position, falling back to the whole file. + let node = file + .syntax() + .token_at_offset(resolve_context.offset) + .left_biased() + .map(|token| token.parent()) + .unwrap_or_else(|| file.syntax().clone()); + let node = pick_node_for_resolution(node); + let scope = sema.scope(&node); + ResolutionScope { + scope, + hygiene: hir::Hygiene::new(sema.db, resolve_context.file_id.into()), + node, + } + } + + /// Returns the function in which SSR was invoked, if any. + pub(crate) fn current_function(&self) -> Option { + self.node.ancestors().find(|node| node.kind() == SyntaxKind::FN).map(|node| node.clone()) + } + + fn resolve_path(&self, path: &ast::Path) -> Option { + let hir_path = hir::Path::from_src(path.clone(), &self.hygiene)?; + // First try resolving the whole path. This will work for things like + // `std::collections::HashMap`, but will fail for things like + // `std::collections::HashMap::new`. + if let Some(resolution) = self.scope.resolve_hir_path(&hir_path) { + return Some(resolution); + } + // Resolution failed, try resolving the qualifier (e.g. `std::collections::HashMap` and if + // that succeeds, then iterate through the candidates on the resolved type with the provided + // name. + let resolved_qualifier = self.scope.resolve_hir_path_qualifier(&hir_path.qualifier()?)?; + if let hir::PathResolution::Def(hir::ModuleDef::Adt(adt)) = resolved_qualifier { + adt.ty(self.scope.db).iterate_path_candidates( + self.scope.db, + self.scope.module()?.krate(), + &self.scope.traits_in_scope(), + Some(hir_path.segments().last()?.name), + |_ty, assoc_item| Some(hir::PathResolution::AssocItem(assoc_item)), + ) + } else { + None + } + } + + fn qualifier_type(&self, path: &SyntaxNode) -> Option { + use syntax::ast::AstNode; + if let Some(path) = ast::Path::cast(path.clone()) { + if let Some(qualifier) = path.qualifier() { + if let Some(resolved_qualifier) = self.resolve_path(&qualifier) { + if let hir::PathResolution::Def(hir::ModuleDef::Adt(adt)) = resolved_qualifier { + return Some(adt.ty(self.scope.db)); + } + } + } + } + None + } +} + +fn is_self(path: &ast::Path) -> bool { + path.segment().map(|segment| segment.self_token().is_some()).unwrap_or(false) +} + +/// Returns a suitable node for resolving paths in the current scope. If we create a scope based on +/// a statement node, then we can't resolve local variables that were defined in the current scope +/// (only in parent scopes). So we find another node, ideally a child of the statement where local +/// variable resolution is permitted. +fn pick_node_for_resolution(node: SyntaxNode) -> SyntaxNode { + match node.kind() { + SyntaxKind::EXPR_STMT => { + if let Some(n) = node.first_child() { + mark::hit!(cursor_after_semicolon); + return n; + } + } + SyntaxKind::LET_STMT | SyntaxKind::IDENT_PAT => { + if let Some(next) = node.next_sibling() { + return pick_node_for_resolution(next); + } + } + SyntaxKind::NAME => { + if let Some(parent) = node.parent() { + return pick_node_for_resolution(parent); + } + } + _ => {} + } + node +} + +/// Returns whether `path` or any of its qualifiers contains type arguments. +fn path_contains_type_arguments(path: Option) -> bool { + if let Some(path) = path { + if let Some(segment) = path.segment() { + if segment.generic_arg_list().is_some() { + mark::hit!(type_arguments_within_path); + return true; + } + } + return path_contains_type_arguments(path.qualifier()); + } + false +} diff --git a/crates/ssr/src/search.rs b/crates/ssr/src/search.rs new file mode 100644 index 000000000..8509cfa4d --- /dev/null +++ b/crates/ssr/src/search.rs @@ -0,0 +1,282 @@ +//! Searching for matches. + +use crate::{ + matching, + resolving::{ResolvedPath, ResolvedPattern, ResolvedRule}, + Match, MatchFinder, +}; +use base_db::{FileId, FileRange}; +use ide_db::{ + defs::Definition, + search::{Reference, SearchScope}, +}; +use rustc_hash::FxHashSet; +use syntax::{ast, AstNode, SyntaxKind, SyntaxNode}; +use test_utils::mark; + +/// A cache for the results of find_usages. This is for when we have multiple patterns that have the +/// same path. e.g. if the pattern was `foo::Bar` that can parse as a path, an expression, a type +/// and as a pattern. In each, the usages of `foo::Bar` are the same and we'd like to avoid finding +/// them more than once. +#[derive(Default)] +pub(crate) struct UsageCache { + usages: Vec<(Definition, Vec)>, +} + +impl<'db> MatchFinder<'db> { + /// Adds all matches for `rule` to `matches_out`. Matches may overlap in ways that make + /// replacement impossible, so further processing is required in order to properly nest matches + /// and remove overlapping matches. This is done in the `nesting` module. + pub(crate) fn find_matches_for_rule( + &self, + rule: &ResolvedRule, + usage_cache: &mut UsageCache, + matches_out: &mut Vec, + ) { + if rule.pattern.contains_self { + // If the pattern contains `self` we restrict the scope of the search to just the + // current method. No other method can reference the same `self`. This makes the + // behavior of `self` consistent with other variables. + if let Some(current_function) = self.resolution_scope.current_function() { + self.slow_scan_node(¤t_function, rule, &None, matches_out); + } + return; + } + if pick_path_for_usages(&rule.pattern).is_none() { + self.slow_scan(rule, matches_out); + return; + } + self.find_matches_for_pattern_tree(rule, &rule.pattern, usage_cache, matches_out); + } + + fn find_matches_for_pattern_tree( + &self, + rule: &ResolvedRule, + pattern: &ResolvedPattern, + usage_cache: &mut UsageCache, + matches_out: &mut Vec, + ) { + if let Some(resolved_path) = pick_path_for_usages(pattern) { + let definition: Definition = resolved_path.resolution.clone().into(); + for reference in self.find_usages(usage_cache, definition) { + if let Some(node_to_match) = self.find_node_to_match(resolved_path, reference) { + if !is_search_permitted_ancestors(&node_to_match) { + mark::hit!(use_declaration_with_braces); + continue; + } + self.try_add_match(rule, &node_to_match, &None, matches_out); + } + } + } + } + + fn find_node_to_match( + &self, + resolved_path: &ResolvedPath, + reference: &Reference, + ) -> Option { + let file = self.sema.parse(reference.file_range.file_id); + let depth = resolved_path.depth as usize; + let offset = reference.file_range.range.start(); + if let Some(path) = + self.sema.find_node_at_offset_with_descend::(file.syntax(), offset) + { + self.sema.ancestors_with_macros(path.syntax().clone()).skip(depth).next() + } else if let Some(path) = + self.sema.find_node_at_offset_with_descend::(file.syntax(), offset) + { + // If the pattern contained a path and we found a reference to that path that wasn't + // itself a path, but was a method call, then we need to adjust how far up to try + // matching by how deep the path was within a CallExpr. The structure would have been + // CallExpr, PathExpr, Path - i.e. a depth offset of 2. We don't need to check if the + // path was part of a CallExpr because if it wasn't then all that will happen is we'll + // fail to match, which is the desired behavior. + const PATH_DEPTH_IN_CALL_EXPR: usize = 2; + if depth < PATH_DEPTH_IN_CALL_EXPR { + return None; + } + self.sema + .ancestors_with_macros(path.syntax().clone()) + .skip(depth - PATH_DEPTH_IN_CALL_EXPR) + .next() + } else { + None + } + } + + fn find_usages<'a>( + &self, + usage_cache: &'a mut UsageCache, + definition: Definition, + ) -> &'a [Reference] { + // Logically if a lookup succeeds we should just return it. Unfortunately returning it would + // extend the lifetime of the borrow, then we wouldn't be able to do the insertion on a + // cache miss. This is a limitation of NLL and is fixed with Polonius. For now we do two + // lookups in the case of a cache hit. + if usage_cache.find(&definition).is_none() { + let usages = definition.find_usages(&self.sema, Some(self.search_scope())); + usage_cache.usages.push((definition, usages)); + return &usage_cache.usages.last().unwrap().1; + } + usage_cache.find(&definition).unwrap() + } + + /// Returns the scope within which we want to search. We don't want un unrestricted search + /// scope, since we don't want to find references in external dependencies. + fn search_scope(&self) -> SearchScope { + // FIXME: We should ideally have a test that checks that we edit local roots and not library + // roots. This probably would require some changes to fixtures, since currently everything + // seems to get put into a single source root. + let mut files = Vec::new(); + self.search_files_do(|file_id| { + files.push(file_id); + }); + SearchScope::files(&files) + } + + fn slow_scan(&self, rule: &ResolvedRule, matches_out: &mut Vec) { + self.search_files_do(|file_id| { + let file = self.sema.parse(file_id); + let code = file.syntax(); + self.slow_scan_node(code, rule, &None, matches_out); + }) + } + + fn search_files_do(&self, mut callback: impl FnMut(FileId)) { + if self.restrict_ranges.is_empty() { + // Unrestricted search. + use base_db::SourceDatabaseExt; + use ide_db::symbol_index::SymbolsDatabase; + for &root in self.sema.db.local_roots().iter() { + let sr = self.sema.db.source_root(root); + for file_id in sr.iter() { + callback(file_id); + } + } + } else { + // Search is restricted, deduplicate file IDs (generally only one). + let mut files = FxHashSet::default(); + for range in &self.restrict_ranges { + if files.insert(range.file_id) { + callback(range.file_id); + } + } + } + } + + fn slow_scan_node( + &self, + code: &SyntaxNode, + rule: &ResolvedRule, + restrict_range: &Option, + matches_out: &mut Vec, + ) { + if !is_search_permitted(code) { + return; + } + self.try_add_match(rule, &code, restrict_range, matches_out); + // If we've got a macro call, we already tried matching it pre-expansion, which is the only + // way to match the whole macro, now try expanding it and matching the expansion. + if let Some(macro_call) = ast::MacroCall::cast(code.clone()) { + if let Some(expanded) = self.sema.expand(¯o_call) { + if let Some(tt) = macro_call.token_tree() { + // When matching within a macro expansion, we only want to allow matches of + // nodes that originated entirely from within the token tree of the macro call. + // i.e. we don't want to match something that came from the macro itself. + self.slow_scan_node( + &expanded, + rule, + &Some(self.sema.original_range(tt.syntax())), + matches_out, + ); + } + } + } + for child in code.children() { + self.slow_scan_node(&child, rule, restrict_range, matches_out); + } + } + + fn try_add_match( + &self, + rule: &ResolvedRule, + code: &SyntaxNode, + restrict_range: &Option, + matches_out: &mut Vec, + ) { + if !self.within_range_restrictions(code) { + mark::hit!(replace_nonpath_within_selection); + return; + } + if let Ok(m) = matching::get_match(false, rule, code, restrict_range, &self.sema) { + matches_out.push(m); + } + } + + /// Returns whether `code` is within one of our range restrictions if we have any. No range + /// restrictions is considered unrestricted and always returns true. + fn within_range_restrictions(&self, code: &SyntaxNode) -> bool { + if self.restrict_ranges.is_empty() { + // There is no range restriction. + return true; + } + let node_range = self.sema.original_range(code); + for range in &self.restrict_ranges { + if range.file_id == node_range.file_id && range.range.contains_range(node_range.range) { + return true; + } + } + false + } +} + +/// Returns whether we support matching within `node` and all of its ancestors. +fn is_search_permitted_ancestors(node: &SyntaxNode) -> bool { + if let Some(parent) = node.parent() { + if !is_search_permitted_ancestors(&parent) { + return false; + } + } + is_search_permitted(node) +} + +/// Returns whether we support matching within this kind of node. +fn is_search_permitted(node: &SyntaxNode) -> bool { + // FIXME: Properly handle use declarations. At the moment, if our search pattern is `foo::bar` + // and the code is `use foo::{baz, bar}`, we'll match `bar`, since it resolves to `foo::bar`. + // However we'll then replace just the part we matched `bar`. We probably need to instead remove + // `bar` and insert a new use declaration. + node.kind() != SyntaxKind::USE +} + +impl UsageCache { + fn find(&mut self, definition: &Definition) -> Option<&[Reference]> { + // We expect a very small number of cache entries (generally 1), so a linear scan should be + // fast enough and avoids the need to implement Hash for Definition. + for (d, refs) in &self.usages { + if d == definition { + return Some(refs); + } + } + None + } +} + +/// Returns a path that's suitable for path resolution. We exclude builtin types, since they aren't +/// something that we can find references to. We then somewhat arbitrarily pick the path that is the +/// longest as this is hopefully more likely to be less common, making it faster to find. +fn pick_path_for_usages(pattern: &ResolvedPattern) -> Option<&ResolvedPath> { + // FIXME: Take the scope of the resolved path into account. e.g. if there are any paths that are + // private to the current module, then we definitely would want to pick them over say a path + // from std. Possibly we should go further than this and intersect the search scopes for all + // resolved paths then search only in that scope. + pattern + .resolved_paths + .iter() + .filter(|(_, p)| { + !matches!(p.resolution, hir::PathResolution::Def(hir::ModuleDef::BuiltinType(_))) + }) + .map(|(node, resolved)| (node.text().len(), resolved)) + .max_by(|(a, _), (b, _)| a.cmp(b)) + .map(|(_, resolved)| resolved) +} diff --git a/crates/ssr/src/tests.rs b/crates/ssr/src/tests.rs new file mode 100644 index 000000000..0d0a00090 --- /dev/null +++ b/crates/ssr/src/tests.rs @@ -0,0 +1,1174 @@ +use crate::{MatchFinder, SsrRule}; +use base_db::{salsa::Durability, FileId, FilePosition, FileRange, SourceDatabaseExt}; +use expect::{expect, Expect}; +use rustc_hash::FxHashSet; +use std::sync::Arc; +use test_utils::{mark, RangeOrOffset}; + +fn parse_error_text(query: &str) -> String { + format!("{}", query.parse::().unwrap_err()) +} + +#[test] +fn parser_empty_query() { + assert_eq!(parse_error_text(""), "Parse error: Cannot find delimiter `==>>`"); +} + +#[test] +fn parser_no_delimiter() { + assert_eq!(parse_error_text("foo()"), "Parse error: Cannot find delimiter `==>>`"); +} + +#[test] +fn parser_two_delimiters() { + assert_eq!( + parse_error_text("foo() ==>> a ==>> b "), + "Parse error: More than one delimiter found" + ); +} + +#[test] +fn parser_repeated_name() { + assert_eq!( + parse_error_text("foo($a, $a) ==>>"), + "Parse error: Name `a` repeats more than once" + ); +} + +#[test] +fn parser_invalid_pattern() { + assert_eq!( + parse_error_text(" ==>> ()"), + "Parse error: Not a valid Rust expression, type, item, path or pattern" + ); +} + +#[test] +fn parser_invalid_template() { + assert_eq!( + parse_error_text("() ==>> )"), + "Parse error: Not a valid Rust expression, type, item, path or pattern" + ); +} + +#[test] +fn parser_undefined_placeholder_in_replacement() { + assert_eq!( + parse_error_text("42 ==>> $a"), + "Parse error: Replacement contains undefined placeholders: $a" + ); +} + +/// `code` may optionally contain a cursor marker `<|>`. If it doesn't, then the position will be +/// the start of the file. If there's a second cursor marker, then we'll return a single range. +pub(crate) fn single_file(code: &str) -> (ide_db::RootDatabase, FilePosition, Vec) { + use base_db::fixture::WithFixture; + use ide_db::symbol_index::SymbolsDatabase; + let (mut db, file_id, range_or_offset) = if code.contains(test_utils::CURSOR_MARKER) { + ide_db::RootDatabase::with_range_or_offset(code) + } else { + let (db, file_id) = ide_db::RootDatabase::with_single_file(code); + (db, file_id, RangeOrOffset::Offset(0.into())) + }; + let selections; + let position; + match range_or_offset { + RangeOrOffset::Range(range) => { + position = FilePosition { file_id, offset: range.start() }; + selections = vec![FileRange { file_id, range: range }]; + } + RangeOrOffset::Offset(offset) => { + position = FilePosition { file_id, offset }; + selections = vec![]; + } + } + let mut local_roots = FxHashSet::default(); + local_roots.insert(base_db::fixture::WORKSPACE); + db.set_local_roots_with_durability(Arc::new(local_roots), Durability::HIGH); + (db, position, selections) +} + +fn assert_ssr_transform(rule: &str, input: &str, expected: Expect) { + assert_ssr_transforms(&[rule], input, expected); +} + +fn assert_ssr_transforms(rules: &[&str], input: &str, expected: Expect) { + let (db, position, selections) = single_file(input); + let mut match_finder = MatchFinder::in_context(&db, position, selections); + for rule in rules { + let rule: SsrRule = rule.parse().unwrap(); + match_finder.add_rule(rule).unwrap(); + } + let edits = match_finder.edits(); + if edits.is_empty() { + panic!("No edits were made"); + } + assert_eq!(edits[0].file_id, position.file_id); + // Note, db.file_text is not necessarily the same as `input`, since fixture parsing alters + // stuff. + let mut actual = db.file_text(position.file_id).to_string(); + edits[0].edit.apply(&mut actual); + expected.assert_eq(&actual); +} + +fn print_match_debug_info(match_finder: &MatchFinder, file_id: FileId, snippet: &str) { + let debug_info = match_finder.debug_where_text_equal(file_id, snippet); + println!( + "Match debug info: {} nodes had text exactly equal to '{}'", + debug_info.len(), + snippet + ); + for (index, d) in debug_info.iter().enumerate() { + println!("Node #{}\n{:#?}\n", index, d); + } +} + +fn assert_matches(pattern: &str, code: &str, expected: &[&str]) { + let (db, position, selections) = single_file(code); + let mut match_finder = MatchFinder::in_context(&db, position, selections); + match_finder.add_search_pattern(pattern.parse().unwrap()).unwrap(); + let matched_strings: Vec = + match_finder.matches().flattened().matches.iter().map(|m| m.matched_text()).collect(); + if matched_strings != expected && !expected.is_empty() { + print_match_debug_info(&match_finder, position.file_id, &expected[0]); + } + assert_eq!(matched_strings, expected); +} + +fn assert_no_match(pattern: &str, code: &str) { + let (db, position, selections) = single_file(code); + let mut match_finder = MatchFinder::in_context(&db, position, selections); + match_finder.add_search_pattern(pattern.parse().unwrap()).unwrap(); + let matches = match_finder.matches().flattened().matches; + if !matches.is_empty() { + print_match_debug_info(&match_finder, position.file_id, &matches[0].matched_text()); + panic!("Got {} matches when we expected none: {:#?}", matches.len(), matches); + } +} + +fn assert_match_failure_reason(pattern: &str, code: &str, snippet: &str, expected_reason: &str) { + let (db, position, selections) = single_file(code); + let mut match_finder = MatchFinder::in_context(&db, position, selections); + match_finder.add_search_pattern(pattern.parse().unwrap()).unwrap(); + let mut reasons = Vec::new(); + for d in match_finder.debug_where_text_equal(position.file_id, snippet) { + if let Some(reason) = d.match_failure_reason() { + reasons.push(reason.to_owned()); + } + } + assert_eq!(reasons, vec![expected_reason]); +} + +#[test] +fn ssr_function_to_method() { + assert_ssr_transform( + "my_function($a, $b) ==>> ($a).my_method($b)", + "fn my_function() {} fn main() { loop { my_function( other_func(x, y), z + w) } }", + expect![["fn my_function() {} fn main() { loop { (other_func(x, y)).my_method(z + w) } }"]], + ) +} + +#[test] +fn ssr_nested_function() { + assert_ssr_transform( + "foo($a, $b, $c) ==>> bar($c, baz($a, $b))", + r#" + //- /lib.rs crate:foo + fn foo() {} + fn bar() {} + fn baz() {} + fn main { foo (x + value.method(b), x+y-z, true && false) } + "#, + expect![[r#" + fn foo() {} + fn bar() {} + fn baz() {} + fn main { bar(true && false, baz(x + value.method(b), x+y-z)) } + "#]], + ) +} + +#[test] +fn ssr_expected_spacing() { + assert_ssr_transform( + "foo($x) + bar() ==>> bar($x)", + "fn foo() {} fn bar() {} fn main() { foo(5) + bar() }", + expect![["fn foo() {} fn bar() {} fn main() { bar(5) }"]], + ); +} + +#[test] +fn ssr_with_extra_space() { + assert_ssr_transform( + "foo($x ) + bar() ==>> bar($x)", + "fn foo() {} fn bar() {} fn main() { foo( 5 ) +bar( ) }", + expect![["fn foo() {} fn bar() {} fn main() { bar(5) }"]], + ); +} + +#[test] +fn ssr_keeps_nested_comment() { + assert_ssr_transform( + "foo($x) ==>> bar($x)", + "fn foo() {} fn bar() {} fn main() { foo(other(5 /* using 5 */)) }", + expect![["fn foo() {} fn bar() {} fn main() { bar(other(5 /* using 5 */)) }"]], + ) +} + +#[test] +fn ssr_keeps_comment() { + assert_ssr_transform( + "foo($x) ==>> bar($x)", + "fn foo() {} fn bar() {} fn main() { foo(5 /* using 5 */) }", + expect![["fn foo() {} fn bar() {} fn main() { bar(5)/* using 5 */ }"]], + ) +} + +#[test] +fn ssr_struct_lit() { + assert_ssr_transform( + "Foo{a: $a, b: $b} ==>> Foo::new($a, $b)", + r#" + struct Foo() {} + impl Foo { fn new() {} } + fn main() { Foo{b:2, a:1} } + "#, + expect![[r#" + struct Foo() {} + impl Foo { fn new() {} } + fn main() { Foo::new(1, 2) } + "#]], + ) +} + +#[test] +fn ignores_whitespace() { + assert_matches("1+2", "fn f() -> i32 {1 + 2}", &["1 + 2"]); + assert_matches("1 + 2", "fn f() -> i32 {1+2}", &["1+2"]); +} + +#[test] +fn no_match() { + assert_no_match("1 + 3", "fn f() -> i32 {1 + 2}"); +} + +#[test] +fn match_fn_definition() { + assert_matches("fn $a($b: $t) {$c}", "fn f(a: i32) {bar()}", &["fn f(a: i32) {bar()}"]); +} + +#[test] +fn match_struct_definition() { + let code = r#" + struct Option {} + struct Bar {} + struct Foo {name: Option}"#; + assert_matches("struct $n {$f: Option}", code, &["struct Foo {name: Option}"]); +} + +#[test] +fn match_expr() { + let code = r#" + fn foo() {} + fn f() -> i32 {foo(40 + 2, 42)}"#; + assert_matches("foo($a, $b)", code, &["foo(40 + 2, 42)"]); + assert_no_match("foo($a, $b, $c)", code); + assert_no_match("foo($a)", code); +} + +#[test] +fn match_nested_method_calls() { + assert_matches( + "$a.z().z().z()", + "fn f() {h().i().j().z().z().z().d().e()}", + &["h().i().j().z().z().z()"], + ); +} + +// Make sure that our node matching semantics don't differ within macro calls. +#[test] +fn match_nested_method_calls_with_macro_call() { + assert_matches( + "$a.z().z().z()", + r#" + macro_rules! m1 { ($a:expr) => {$a}; } + fn f() {m1!(h().i().j().z().z().z().d().e())}"#, + &["h().i().j().z().z().z()"], + ); +} + +#[test] +fn match_complex_expr() { + let code = r#" + fn foo() {} fn bar() {} + fn f() -> i32 {foo(bar(40, 2), 42)}"#; + assert_matches("foo($a, $b)", code, &["foo(bar(40, 2), 42)"]); + assert_no_match("foo($a, $b, $c)", code); + assert_no_match("foo($a)", code); + assert_matches("bar($a, $b)", code, &["bar(40, 2)"]); +} + +// Trailing commas in the code should be ignored. +#[test] +fn match_with_trailing_commas() { + // Code has comma, pattern doesn't. + assert_matches("foo($a, $b)", "fn foo() {} fn f() {foo(1, 2,);}", &["foo(1, 2,)"]); + assert_matches("Foo{$a, $b}", "struct Foo {} fn f() {Foo{1, 2,};}", &["Foo{1, 2,}"]); + + // Pattern has comma, code doesn't. + assert_matches("foo($a, $b,)", "fn foo() {} fn f() {foo(1, 2);}", &["foo(1, 2)"]); + assert_matches("Foo{$a, $b,}", "struct Foo {} fn f() {Foo{1, 2};}", &["Foo{1, 2}"]); +} + +#[test] +fn match_type() { + assert_matches("i32", "fn f() -> i32 {1 + 2}", &["i32"]); + assert_matches( + "Option<$a>", + "struct Option {} fn f() -> Option {42}", + &["Option"], + ); + assert_no_match( + "Option<$a>", + "struct Option {} struct Result {} fn f() -> Result {42}", + ); +} + +#[test] +fn match_struct_instantiation() { + let code = r#" + struct Foo {bar: i32, baz: i32} + fn f() {Foo {bar: 1, baz: 2}}"#; + assert_matches("Foo {bar: 1, baz: 2}", code, &["Foo {bar: 1, baz: 2}"]); + // Now with placeholders for all parts of the struct. + assert_matches("Foo {$a: $b, $c: $d}", code, &["Foo {bar: 1, baz: 2}"]); + assert_matches("Foo {}", "struct Foo {} fn f() {Foo {}}", &["Foo {}"]); +} + +#[test] +fn match_path() { + let code = r#" + mod foo { + pub fn bar() {} + } + fn f() {foo::bar(42)}"#; + assert_matches("foo::bar", code, &["foo::bar"]); + assert_matches("$a::bar", code, &["foo::bar"]); + assert_matches("foo::$b", code, &["foo::bar"]); +} + +#[test] +fn match_pattern() { + assert_matches("Some($a)", "struct Some(); fn f() {if let Some(x) = foo() {}}", &["Some(x)"]); +} + +// If our pattern has a full path, e.g. a::b::c() and the code has c(), but c resolves to +// a::b::c, then we should match. +#[test] +fn match_fully_qualified_fn_path() { + let code = r#" + mod a { + pub mod b { + pub fn c(_: i32) {} + } + } + use a::b::c; + fn f1() { + c(42); + } + "#; + assert_matches("a::b::c($a)", code, &["c(42)"]); +} + +#[test] +fn match_resolved_type_name() { + let code = r#" + mod m1 { + pub mod m2 { + pub trait Foo {} + } + } + mod m3 { + trait Foo {} + fn f1(f: Option<&dyn Foo>) {} + } + mod m4 { + use crate::m1::m2::Foo; + fn f1(f: Option<&dyn Foo>) {} + } + "#; + assert_matches("m1::m2::Foo<$t>", code, &["Foo"]); +} + +#[test] +fn type_arguments_within_path() { + mark::check!(type_arguments_within_path); + let code = r#" + mod foo { + pub struct Bar {t: T} + impl Bar { + pub fn baz() {} + } + } + fn f1() {foo::Bar::::baz();} + "#; + assert_no_match("foo::Bar::::baz()", code); + assert_matches("foo::Bar::::baz()", code, &["foo::Bar::::baz()"]); +} + +#[test] +fn literal_constraint() { + mark::check!(literal_constraint); + let code = r#" + enum Option { Some(T), None } + use Option::Some; + fn f1() { + let x1 = Some(42); + let x2 = Some("foo"); + let x3 = Some(x1); + let x4 = Some(40 + 2); + let x5 = Some(true); + } + "#; + assert_matches("Some(${a:kind(literal)})", code, &["Some(42)", "Some(\"foo\")", "Some(true)"]); + assert_matches("Some(${a:not(kind(literal))})", code, &["Some(x1)", "Some(40 + 2)"]); +} + +#[test] +fn match_reordered_struct_instantiation() { + assert_matches( + "Foo {aa: 1, b: 2, ccc: 3}", + "struct Foo {} fn f() {Foo {b: 2, ccc: 3, aa: 1}}", + &["Foo {b: 2, ccc: 3, aa: 1}"], + ); + assert_no_match("Foo {a: 1}", "struct Foo {} fn f() {Foo {b: 1}}"); + assert_no_match("Foo {a: 1}", "struct Foo {} fn f() {Foo {a: 2}}"); + assert_no_match("Foo {a: 1, b: 2}", "struct Foo {} fn f() {Foo {a: 1}}"); + assert_no_match("Foo {a: 1, b: 2}", "struct Foo {} fn f() {Foo {b: 2}}"); + assert_no_match("Foo {a: 1, }", "struct Foo {} fn f() {Foo {a: 1, b: 2}}"); + assert_no_match("Foo {a: 1, z: 9}", "struct Foo {} fn f() {Foo {a: 1}}"); +} + +#[test] +fn match_macro_invocation() { + assert_matches( + "foo!($a)", + "macro_rules! foo {() => {}} fn() {foo(foo!(foo()))}", + &["foo!(foo())"], + ); + assert_matches( + "foo!(41, $a, 43)", + "macro_rules! foo {() => {}} fn() {foo!(41, 42, 43)}", + &["foo!(41, 42, 43)"], + ); + assert_no_match("foo!(50, $a, 43)", "macro_rules! foo {() => {}} fn() {foo!(41, 42, 43}"); + assert_no_match("foo!(41, $a, 50)", "macro_rules! foo {() => {}} fn() {foo!(41, 42, 43}"); + assert_matches( + "foo!($a())", + "macro_rules! foo {() => {}} fn() {foo!(bar())}", + &["foo!(bar())"], + ); +} + +// When matching within a macro expansion, we only allow matches of nodes that originated from +// the macro call, not from the macro definition. +#[test] +fn no_match_expression_from_macro() { + assert_no_match( + "$a.clone()", + r#" + macro_rules! m1 { + () => {42.clone()} + } + fn f1() {m1!()} + "#, + ); +} + +// We definitely don't want to allow matching of an expression that part originates from the +// macro call `42` and part from the macro definition `.clone()`. +#[test] +fn no_match_split_expression() { + assert_no_match( + "$a.clone()", + r#" + macro_rules! m1 { + ($x:expr) => {$x.clone()} + } + fn f1() {m1!(42)} + "#, + ); +} + +#[test] +fn replace_function_call() { + // This test also makes sure that we ignore empty-ranges. + assert_ssr_transform( + "foo() ==>> bar()", + "fn foo() {<|><|>} fn bar() {} fn f1() {foo(); foo();}", + expect![["fn foo() {} fn bar() {} fn f1() {bar(); bar();}"]], + ); +} + +#[test] +fn replace_function_call_with_placeholders() { + assert_ssr_transform( + "foo($a, $b) ==>> bar($b, $a)", + "fn foo() {} fn bar() {} fn f1() {foo(5, 42)}", + expect![["fn foo() {} fn bar() {} fn f1() {bar(42, 5)}"]], + ); +} + +#[test] +fn replace_nested_function_calls() { + assert_ssr_transform( + "foo($a) ==>> bar($a)", + "fn foo() {} fn bar() {} fn f1() {foo(foo(42))}", + expect![["fn foo() {} fn bar() {} fn f1() {bar(bar(42))}"]], + ); +} + +#[test] +fn replace_associated_function_call() { + assert_ssr_transform( + "Foo::new() ==>> Bar::new()", + r#" + struct Foo {} + impl Foo { fn new() {} } + struct Bar {} + impl Bar { fn new() {} } + fn f1() {Foo::new();} + "#, + expect![[r#" + struct Foo {} + impl Foo { fn new() {} } + struct Bar {} + impl Bar { fn new() {} } + fn f1() {Bar::new();} + "#]], + ); +} + +#[test] +fn replace_associated_trait_default_function_call() { + mark::check!(replace_associated_trait_default_function_call); + assert_ssr_transform( + "Bar2::foo() ==>> Bar2::foo2()", + r#" + trait Foo { fn foo() {} } + pub struct Bar {} + impl Foo for Bar {} + pub struct Bar2 {} + impl Foo for Bar2 {} + impl Bar2 { fn foo2() {} } + fn main() { + Bar::foo(); + Bar2::foo(); + } + "#, + expect![[r#" + trait Foo { fn foo() {} } + pub struct Bar {} + impl Foo for Bar {} + pub struct Bar2 {} + impl Foo for Bar2 {} + impl Bar2 { fn foo2() {} } + fn main() { + Bar::foo(); + Bar2::foo2(); + } + "#]], + ); +} + +#[test] +fn replace_associated_trait_constant() { + mark::check!(replace_associated_trait_constant); + assert_ssr_transform( + "Bar2::VALUE ==>> Bar2::VALUE_2222", + r#" + trait Foo { const VALUE: i32; const VALUE_2222: i32; } + pub struct Bar {} + impl Foo for Bar { const VALUE: i32 = 1; const VALUE_2222: i32 = 2; } + pub struct Bar2 {} + impl Foo for Bar2 { const VALUE: i32 = 1; const VALUE_2222: i32 = 2; } + impl Bar2 { fn foo2() {} } + fn main() { + Bar::VALUE; + Bar2::VALUE; + } + "#, + expect![[r#" + trait Foo { const VALUE: i32; const VALUE_2222: i32; } + pub struct Bar {} + impl Foo for Bar { const VALUE: i32 = 1; const VALUE_2222: i32 = 2; } + pub struct Bar2 {} + impl Foo for Bar2 { const VALUE: i32 = 1; const VALUE_2222: i32 = 2; } + impl Bar2 { fn foo2() {} } + fn main() { + Bar::VALUE; + Bar2::VALUE_2222; + } + "#]], + ); +} + +#[test] +fn replace_path_in_different_contexts() { + // Note the <|> inside module a::b which marks the point where the rule is interpreted. We + // replace foo with bar, but both need different path qualifiers in different contexts. In f4, + // foo is unqualified because of a use statement, however the replacement needs to be fully + // qualified. + assert_ssr_transform( + "c::foo() ==>> c::bar()", + r#" + mod a { + pub mod b {<|> + pub mod c { + pub fn foo() {} + pub fn bar() {} + fn f1() { foo() } + } + fn f2() { c::foo() } + } + fn f3() { b::c::foo() } + } + use a::b::c::foo; + fn f4() { foo() } + "#, + expect![[r#" + mod a { + pub mod b { + pub mod c { + pub fn foo() {} + pub fn bar() {} + fn f1() { bar() } + } + fn f2() { c::bar() } + } + fn f3() { b::c::bar() } + } + use a::b::c::foo; + fn f4() { a::b::c::bar() } + "#]], + ); +} + +#[test] +fn replace_associated_function_with_generics() { + assert_ssr_transform( + "c::Foo::<$a>::new() ==>> d::Bar::<$a>::default()", + r#" + mod c { + pub struct Foo {v: T} + impl Foo { pub fn new() {} } + fn f1() { + Foo::::new(); + } + } + mod d { + pub struct Bar {v: T} + impl Bar { pub fn default() {} } + fn f1() { + super::c::Foo::::new(); + } + } + "#, + expect![[r#" + mod c { + pub struct Foo {v: T} + impl Foo { pub fn new() {} } + fn f1() { + crate::d::Bar::::default(); + } + } + mod d { + pub struct Bar {v: T} + impl Bar { pub fn default() {} } + fn f1() { + Bar::::default(); + } + } + "#]], + ); +} + +#[test] +fn replace_type() { + assert_ssr_transform( + "Result<(), $a> ==>> Option<$a>", + "struct Result {} struct Option {} fn f1() -> Result<(), Vec> {foo()}", + expect![[ + "struct Result {} struct Option {} fn f1() -> Option> {foo()}" + ]], + ); +} + +#[test] +fn replace_macro_invocations() { + assert_ssr_transform( + "try!($a) ==>> $a?", + "macro_rules! try {() => {}} fn f1() -> Result<(), E> {bar(try!(foo()));}", + expect![["macro_rules! try {() => {}} fn f1() -> Result<(), E> {bar(foo()?);}"]], + ); + assert_ssr_transform( + "foo!($a($b)) ==>> foo($b, $a)", + "macro_rules! foo {() => {}} fn f1() {foo!(abc(def() + 2));}", + expect![["macro_rules! foo {() => {}} fn f1() {foo(def() + 2, abc);}"]], + ); +} + +#[test] +fn replace_binary_op() { + assert_ssr_transform( + "$a + $b ==>> $b + $a", + "fn f() {2 * 3 + 4 * 5}", + expect![["fn f() {4 * 5 + 2 * 3}"]], + ); + assert_ssr_transform( + "$a + $b ==>> $b + $a", + "fn f() {1 + 2 + 3 + 4}", + expect![[r#"fn f() {4 + (3 + (2 + 1))}"#]], + ); +} + +#[test] +fn match_binary_op() { + assert_matches("$a + $b", "fn f() {1 + 2 + 3 + 4}", &["1 + 2", "1 + 2 + 3", "1 + 2 + 3 + 4"]); +} + +#[test] +fn multiple_rules() { + assert_ssr_transforms( + &["$a + 1 ==>> add_one($a)", "$a + $b ==>> add($a, $b)"], + "fn add() {} fn add_one() {} fn f() -> i32 {3 + 2 + 1}", + expect![["fn add() {} fn add_one() {} fn f() -> i32 {add_one(add(3, 2))}"]], + ) +} + +#[test] +fn multiple_rules_with_nested_matches() { + assert_ssr_transforms( + &["foo1($a) ==>> bar1($a)", "foo2($a) ==>> bar2($a)"], + r#" + fn foo1() {} fn foo2() {} fn bar1() {} fn bar2() {} + fn f() {foo1(foo2(foo1(foo2(foo1(42)))))} + "#, + expect![[r#" + fn foo1() {} fn foo2() {} fn bar1() {} fn bar2() {} + fn f() {bar1(bar2(bar1(bar2(bar1(42)))))} + "#]], + ) +} + +#[test] +fn match_within_macro_invocation() { + let code = r#" + macro_rules! foo { + ($a:stmt; $b:expr) => { + $b + }; + } + struct A {} + impl A { + fn bar() {} + } + fn f1() { + let aaa = A {}; + foo!(macro_ignores_this(); aaa.bar()); + } + "#; + assert_matches("$a.bar()", code, &["aaa.bar()"]); +} + +#[test] +fn replace_within_macro_expansion() { + assert_ssr_transform( + "$a.foo() ==>> bar($a)", + r#" + macro_rules! macro1 { + ($a:expr) => {$a} + } + fn bar() {} + fn f() {macro1!(5.x().foo().o2())} + "#, + expect![[r#" + macro_rules! macro1 { + ($a:expr) => {$a} + } + fn bar() {} + fn f() {macro1!(bar(5.x()).o2())} + "#]], + ) +} + +#[test] +fn replace_outside_and_within_macro_expansion() { + assert_ssr_transform( + "foo($a) ==>> bar($a)", + r#" + fn foo() {} fn bar() {} + macro_rules! macro1 { + ($a:expr) => {$a} + } + fn f() {foo(foo(macro1!(foo(foo(42)))))} + "#, + expect![[r#" + fn foo() {} fn bar() {} + macro_rules! macro1 { + ($a:expr) => {$a} + } + fn f() {bar(bar(macro1!(bar(bar(42)))))} + "#]], + ) +} + +#[test] +fn preserves_whitespace_within_macro_expansion() { + assert_ssr_transform( + "$a + $b ==>> $b - $a", + r#" + macro_rules! macro1 { + ($a:expr) => {$a} + } + fn f() {macro1!(1 * 2 + 3 + 4} + "#, + expect![[r#" + macro_rules! macro1 { + ($a:expr) => {$a} + } + fn f() {macro1!(4 - (3 - 1 * 2)} + "#]], + ) +} + +#[test] +fn add_parenthesis_when_necessary() { + assert_ssr_transform( + "foo($a) ==>> $a.to_string()", + r#" + fn foo(_: i32) {} + fn bar3(v: i32) { + foo(1 + 2); + foo(-v); + } + "#, + expect![[r#" + fn foo(_: i32) {} + fn bar3(v: i32) { + (1 + 2).to_string(); + (-v).to_string(); + } + "#]], + ) +} + +#[test] +fn match_failure_reasons() { + let code = r#" + fn bar() {} + macro_rules! foo { + ($a:expr) => { + 1 + $a + 2 + }; + } + fn f1() { + bar(1, 2); + foo!(5 + 43.to_string() + 5); + } + "#; + assert_match_failure_reason( + "bar($a, 3)", + code, + "bar(1, 2)", + r#"Pattern wanted token '3' (INT_NUMBER), but code had token '2' (INT_NUMBER)"#, + ); + assert_match_failure_reason( + "42.to_string()", + code, + "43.to_string()", + r#"Pattern wanted token '42' (INT_NUMBER), but code had token '43' (INT_NUMBER)"#, + ); +} + +#[test] +fn overlapping_possible_matches() { + // There are three possible matches here, however the middle one, `foo(foo(foo(42)))` shouldn't + // match because it overlaps with the outer match. The inner match is permitted since it's is + // contained entirely within the placeholder of the outer match. + assert_matches( + "foo(foo($a))", + "fn foo() {} fn main() {foo(foo(foo(foo(42))))}", + &["foo(foo(42))", "foo(foo(foo(foo(42))))"], + ); +} + +#[test] +fn use_declaration_with_braces() { + // It would be OK for a path rule to match and alter a use declaration. We shouldn't mess it up + // though. In particular, we must not change `use foo::{baz, bar}` to `use foo::{baz, + // foo2::bar2}`. + mark::check!(use_declaration_with_braces); + assert_ssr_transform( + "foo::bar ==>> foo2::bar2", + r#" + mod foo { pub fn bar() {} pub fn baz() {} } + mod foo2 { pub fn bar2() {} } + use foo::{baz, bar}; + fn main() { bar() } + "#, + expect![[" + mod foo { pub fn bar() {} pub fn baz() {} } + mod foo2 { pub fn bar2() {} } + use foo::{baz, bar}; + fn main() { foo2::bar2() } + "]], + ) +} + +#[test] +fn ufcs_matches_method_call() { + let code = r#" + struct Foo {} + impl Foo { + fn new(_: i32) -> Foo { Foo {} } + fn do_stuff(&self, _: i32) {} + } + struct Bar {} + impl Bar { + fn new(_: i32) -> Bar { Bar {} } + fn do_stuff(&self, v: i32) {} + } + fn main() { + let b = Bar {}; + let f = Foo {}; + b.do_stuff(1); + f.do_stuff(2); + Foo::new(4).do_stuff(3); + // Too many / too few args - should never match + f.do_stuff(2, 10); + f.do_stuff(); + } + "#; + assert_matches("Foo::do_stuff($a, $b)", code, &["f.do_stuff(2)", "Foo::new(4).do_stuff(3)"]); + // The arguments needs special handling in the case of a function call matching a method call + // and the first argument is different. + assert_matches("Foo::do_stuff($a, 2)", code, &["f.do_stuff(2)"]); + assert_matches("Foo::do_stuff(Foo::new(4), $b)", code, &["Foo::new(4).do_stuff(3)"]); + + assert_ssr_transform( + "Foo::do_stuff(Foo::new($a), $b) ==>> Bar::new($b).do_stuff($a)", + code, + expect![[r#" + struct Foo {} + impl Foo { + fn new(_: i32) -> Foo { Foo {} } + fn do_stuff(&self, _: i32) {} + } + struct Bar {} + impl Bar { + fn new(_: i32) -> Bar { Bar {} } + fn do_stuff(&self, v: i32) {} + } + fn main() { + let b = Bar {}; + let f = Foo {}; + b.do_stuff(1); + f.do_stuff(2); + Bar::new(3).do_stuff(4); + // Too many / too few args - should never match + f.do_stuff(2, 10); + f.do_stuff(); + } + "#]], + ); +} + +#[test] +fn pattern_is_a_single_segment_path() { + mark::check!(pattern_is_a_single_segment_path); + // The first function should not be altered because the `foo` in scope at the cursor position is + // a different `foo`. This case is special because "foo" can be parsed as a pattern (IDENT_PAT -> + // NAME -> IDENT), which contains no path. If we're not careful we'll end up matching the `foo` + // in `let foo` from the first function. Whether we should match the `let foo` in the second + // function is less clear. At the moment, we don't. Doing so sounds like a rename operation, + // which isn't really what SSR is for, especially since the replacement `bar` must be able to be + // resolved, which means if we rename `foo` we'll get a name collision. + assert_ssr_transform( + "foo ==>> bar", + r#" + fn f1() -> i32 { + let foo = 1; + let bar = 2; + foo + } + fn f1() -> i32 { + let foo = 1; + let bar = 2; + foo<|> + } + "#, + expect![[r#" + fn f1() -> i32 { + let foo = 1; + let bar = 2; + foo + } + fn f1() -> i32 { + let foo = 1; + let bar = 2; + bar + } + "#]], + ); +} + +#[test] +fn replace_local_variable_reference() { + // The pattern references a local variable `foo` in the block containing the cursor. We should + // only replace references to this variable `foo`, not other variables that just happen to have + // the same name. + mark::check!(cursor_after_semicolon); + assert_ssr_transform( + "foo + $a ==>> $a - foo", + r#" + fn bar1() -> i32 { + let mut res = 0; + let foo = 5; + res += foo + 1; + let foo = 10; + res += foo + 2;<|> + res += foo + 3; + let foo = 15; + res += foo + 4; + res + } + "#, + expect![[r#" + fn bar1() -> i32 { + let mut res = 0; + let foo = 5; + res += foo + 1; + let foo = 10; + res += 2 - foo; + res += 3 - foo; + let foo = 15; + res += foo + 4; + res + } + "#]], + ) +} + +#[test] +fn replace_path_within_selection() { + assert_ssr_transform( + "foo ==>> bar", + r#" + fn main() { + let foo = 41; + let bar = 42; + do_stuff(foo); + do_stuff(foo);<|> + do_stuff(foo); + do_stuff(foo);<|> + do_stuff(foo); + }"#, + expect![[r#" + fn main() { + let foo = 41; + let bar = 42; + do_stuff(foo); + do_stuff(foo); + do_stuff(bar); + do_stuff(bar); + do_stuff(foo); + }"#]], + ); +} + +#[test] +fn replace_nonpath_within_selection() { + mark::check!(replace_nonpath_within_selection); + assert_ssr_transform( + "$a + $b ==>> $b * $a", + r#" + fn main() { + let v = 1 + 2;<|> + let v2 = 3 + 3; + let v3 = 4 + 5;<|> + let v4 = 6 + 7; + }"#, + expect![[r#" + fn main() { + let v = 1 + 2; + let v2 = 3 * 3; + let v3 = 5 * 4; + let v4 = 6 + 7; + }"#]], + ); +} + +#[test] +fn replace_self() { + // `foo(self)` occurs twice in the code, however only the first occurrence is the `self` that's + // in scope where the rule is invoked. + assert_ssr_transform( + "foo(self) ==>> bar(self)", + r#" + struct S1 {} + fn foo(_: &S1) {} + fn bar(_: &S1) {} + impl S1 { + fn f1(&self) { + foo(self)<|> + } + fn f2(&self) { + foo(self) + } + } + "#, + expect![[r#" + struct S1 {} + fn foo(_: &S1) {} + fn bar(_: &S1) {} + impl S1 { + fn f1(&self) { + bar(self) + } + fn f2(&self) { + foo(self) + } + } + "#]], + ); +} + +#[test] +fn match_trait_method_call() { + // `Bar::foo` and `Bar2::foo` resolve to the same function. Make sure we only match if the type + // matches what's in the pattern. Also checks that we handle autoderef. + let code = r#" + pub struct Bar {} + pub struct Bar2 {} + pub trait Foo { + fn foo(&self, _: i32) {} + } + impl Foo for Bar {} + impl Foo for Bar2 {} + fn main() { + let v1 = Bar {}; + let v2 = Bar2 {}; + let v1_ref = &v1; + let v2_ref = &v2; + v1.foo(1); + v2.foo(2); + Bar::foo(&v1, 3); + Bar2::foo(&v2, 4); + v1_ref.foo(5); + v2_ref.foo(6); + } + "#; + assert_matches("Bar::foo($a, $b)", code, &["v1.foo(1)", "Bar::foo(&v1, 3)", "v1_ref.foo(5)"]); + assert_matches("Bar2::foo($a, $b)", code, &["v2.foo(2)", "Bar2::foo(&v2, 4)", "v2_ref.foo(6)"]); +} -- cgit v1.2.3 From 9664c57e60ec5662b3e8b063324d9ab7879d5570 Mon Sep 17 00:00:00 2001 From: Aleksey Kladov Date: Thu, 13 Aug 2020 23:52:14 +0200 Subject: Make hygiene private to hir --- crates/ssr/src/resolving.rs | 24 +++++++++++++----------- 1 file changed, 13 insertions(+), 11 deletions(-) (limited to 'crates/ssr/src') diff --git a/crates/ssr/src/resolving.rs b/crates/ssr/src/resolving.rs index 020fd7994..4441fb426 100644 --- a/crates/ssr/src/resolving.rs +++ b/crates/ssr/src/resolving.rs @@ -10,7 +10,6 @@ use test_utils::mark; pub(crate) struct ResolutionScope<'db> { scope: hir::SemanticsScope<'db>, - hygiene: hir::Hygiene, node: SyntaxNode, } @@ -201,11 +200,7 @@ impl<'db> ResolutionScope<'db> { .unwrap_or_else(|| file.syntax().clone()); let node = pick_node_for_resolution(node); let scope = sema.scope(&node); - ResolutionScope { - scope, - hygiene: hir::Hygiene::new(sema.db, resolve_context.file_id.into()), - node, - } + ResolutionScope { scope, node } } /// Returns the function in which SSR was invoked, if any. @@ -214,24 +209,31 @@ impl<'db> ResolutionScope<'db> { } fn resolve_path(&self, path: &ast::Path) -> Option { - let hir_path = hir::Path::from_src(path.clone(), &self.hygiene)?; // First try resolving the whole path. This will work for things like // `std::collections::HashMap`, but will fail for things like // `std::collections::HashMap::new`. - if let Some(resolution) = self.scope.resolve_hir_path(&hir_path) { + if let Some(resolution) = self.scope.resolve_hypothetical(&path) { return Some(resolution); } // Resolution failed, try resolving the qualifier (e.g. `std::collections::HashMap` and if // that succeeds, then iterate through the candidates on the resolved type with the provided // name. - let resolved_qualifier = self.scope.resolve_hir_path_qualifier(&hir_path.qualifier()?)?; + let resolved_qualifier = self.scope.resolve_hypothetical(&path.qualifier()?)?; if let hir::PathResolution::Def(hir::ModuleDef::Adt(adt)) = resolved_qualifier { + let name = path.segment()?.name_ref()?; adt.ty(self.scope.db).iterate_path_candidates( self.scope.db, self.scope.module()?.krate(), &self.scope.traits_in_scope(), - Some(hir_path.segments().last()?.name), - |_ty, assoc_item| Some(hir::PathResolution::AssocItem(assoc_item)), + None, + |_ty, assoc_item| { + let item_name = assoc_item.name(self.scope.db)?; + if item_name.to_string().as_str() == name.text().as_str() { + Some(hir::PathResolution::AssocItem(assoc_item)) + } else { + None + } + }, ) } else { None -- cgit v1.2.3