docs

1 files changed, 14 insertions, 156 deletions

diff --git a/crates/ra_analysis/src/completion.rs b/crates/ra_analysis/src/completion.rs
index 93edcc4c2..2d61a3aef 100644
--- a/crates/ra_analysis/src/completion.rs
+++ b/crates/ra_analysis/src/completion.rs
@@ -1,4 +1,5 @@
 mod completion_item;
+mod completion_context;
 
 mod complete_fn_param;
 mod complete_keyword;
@@ -6,34 +7,33 @@ mod complete_snippet;
 mod complete_path;
 mod complete_scope;
 
-use ra_editor::find_node_at_offset;
-use ra_text_edit::AtomTextEdit;
-use ra_syntax::{
-    algo::find_leaf_at_offset,
-    ast,
-    AstNode,
-    SyntaxNodeRef,
-    SourceFileNode,
-    TextUnit,
-    SyntaxKind::*,
-};
 use ra_db::SyntaxDatabase;
-use hir::source_binder;
 
 use crate::{
     db,
     Cancelable, FilePosition,
-    completion::completion_item::{Completions, CompletionKind},
+    completion::{
+        completion_item::{Completions, CompletionKind},
+        completion_context::CompletionContext,
+    },
 };
 
 pub use crate::completion::completion_item::{CompletionItem, InsertText};
 
+/// Main entry point for copmletion. We run comletion as a two-phase process.
+///
+/// First, we look at the position and collect a so-called `CompletionContext.
+/// This is a somewhat messy process, because, during completion, syntax tree is
+/// incomplete and can look readlly weired.
+///
+/// Once the context is collected, we run a series of completion routines whihc
+/// look at the context and produce completion items.
 pub(crate) fn completions(
     db: &db::RootDatabase,
     position: FilePosition,
 ) -> Cancelable<Option<Completions>> {
     let original_file = db.source_file(position.file_id);
-    let ctx = ctry!(SyntaxContext::new(db, &original_file, position)?);
+    let ctx = ctry!(CompletionContext::new(db, &original_file, position)?);
 
     let mut acc = Completions::default();
 
@@ -47,148 +47,6 @@ pub(crate) fn completions(
     Ok(Some(acc))
 }
 
-/// `SyntaxContext` is created early during completion to figure out, where
-/// exactly is the cursor, syntax-wise.
-#[derive(Debug)]
-pub(super) struct SyntaxContext<'a> {
-    db: &'a db::RootDatabase,
-    offset: TextUnit,
-    leaf: SyntaxNodeRef<'a>,
-    module: Option<hir::Module>,
-    enclosing_fn: Option<ast::FnDef<'a>>,
-    is_param: bool,
-    /// A single-indent path, like `foo`.
-    is_trivial_path: bool,
-    /// If not a trivial, path, the prefix (qualifier).
-    path_prefix: Option<hir::Path>,
-    after_if: bool,
-    is_stmt: bool,
-    /// Something is typed at the "top" level, in module or impl/trait.
-    is_new_item: bool,
-}
-
-impl<'a> SyntaxContext<'a> {
-    pub(super) fn new(
-        db: &'a db::RootDatabase,
-        original_file: &'a SourceFileNode,
-        position: FilePosition,
-    ) -> Cancelable<Option<SyntaxContext<'a>>> {
-        let module = source_binder::module_from_position(db, position)?;
-        let leaf =
-            ctry!(find_leaf_at_offset(original_file.syntax(), position.offset).left_biased());
-        let mut ctx = SyntaxContext {
-            db,
-            leaf,
-            offset: position.offset,
-            module,
-            enclosing_fn: None,
-            is_param: false,
-            is_trivial_path: false,
-            path_prefix: None,
-            after_if: false,
-            is_stmt: false,
-            is_new_item: false,
-        };
-        ctx.fill(original_file, position.offset);
-        Ok(Some(ctx))
-    }
-
-    fn fill(&mut self, original_file: &SourceFileNode, offset: TextUnit) {
-        // Insert a fake ident to get a valid parse tree. We will use this file
-        // to determine context, though the original_file will be used for
-        // actual completion.
-        let file = {
-            let edit = AtomTextEdit::insert(offset, "intellijRulezz".to_string());
-            original_file.reparse(&edit)
-        };
-
-        // First, let's try to complete a reference to some declaration.
-        if let Some(name_ref) = find_node_at_offset::<ast::NameRef>(file.syntax(), offset) {
-            // Special case, `trait T { fn foo(i_am_a_name_ref) {} }`.
-            // See RFC#1685.
-            if is_node::<ast::Param>(name_ref.syntax()) {
-                self.is_param = true;
-                return;
-            }
-            self.classify_name_ref(&file, name_ref);
-        }
-
-        // Otherwise, see if this is a declaration. We can use heuristics to
-        // suggest declaration names, see `CompletionKind::Magic`.
-        if let Some(name) = find_node_at_offset::<ast::Name>(file.syntax(), offset) {
-            if is_node::<ast::Param>(name.syntax()) {
-                self.is_param = true;
-                return;
-            }
-        }
-    }
-    fn classify_name_ref(&mut self, file: &SourceFileNode, name_ref: ast::NameRef) {
-        let name_range = name_ref.syntax().range();
-        let top_node = name_ref
-            .syntax()
-            .ancestors()
-            .take_while(|it| it.range() == name_range)
-            .last()
-            .unwrap();
-
-        match top_node.parent().map(|it| it.kind()) {
-            Some(SOURCE_FILE) | Some(ITEM_LIST) => {
-                self.is_new_item = true;
-                return;
-            }
-            _ => (),
-        }
-
-        let parent = match name_ref.syntax().parent() {
-            Some(it) => it,
-            None => return,
-        };
-        if let Some(segment) = ast::PathSegment::cast(parent) {
-            let path = segment.parent_path();
-            if let Some(mut path) = hir::Path::from_ast(path) {
-                if !path.is_ident() {
-                    path.segments.pop().unwrap();
-                    self.path_prefix = Some(path);
-                    return;
-                }
-            }
-            if path.qualifier().is_none() {
-                self.is_trivial_path = true;
-                self.enclosing_fn = self
-                    .leaf
-                    .ancestors()
-                    .take_while(|it| it.kind() != SOURCE_FILE && it.kind() != MODULE)
-                    .find_map(ast::FnDef::cast);
-
-                self.is_stmt = match name_ref
-                    .syntax()
-                    .ancestors()
-                    .filter_map(ast::ExprStmt::cast)
-                    .next()
-                {
-                    None => false,
-                    Some(expr_stmt) => expr_stmt.syntax().range() == name_ref.syntax().range(),
-                };
-
-                if let Some(off) = name_ref.syntax().range().start().checked_sub(2.into()) {
-                    if let Some(if_expr) = find_node_at_offset::<ast::IfExpr>(file.syntax(), off) {
-                        if if_expr.syntax().range().end() < name_ref.syntax().range().start() {
-                            self.after_if = true;
-                        }
-                    }
-                }
-            }
-        }
-    }
-}
-
-fn is_node<'a, N: AstNode<'a>>(node: SyntaxNodeRef<'a>) -> bool {
-    match node.ancestors().filter_map(N::cast).next() {
-        None => false,
-        Some(n) => n.syntax().range() == node.range(),
-    }
-}
-
 #[cfg(test)]
 fn check_completion(code: &str, expected_completions: &str, kind: CompletionKind) {
     use crate::mock_analysis::{single_file_with_position, analysis_and_position};