diff options
author | bors[bot] <26634292+bors[bot]@users.noreply.github.com> | 2020-11-28 14:33:57 +0000 |
---|---|---|
committer | GitHub <[email protected]> | 2020-11-28 14:33:57 +0000 |
commit | b7ece77af49ce59762fc3246a4c721411efe637e (patch) | |
tree | 81c617c2df781e27b0c16c10515ebcf588cb376e /crates/ide_db/src/helpers | |
parent | fe2ac4480bd71c9cf25d2c21bda82b33558d27a4 (diff) | |
parent | 3f612d37c68a6e4c09e407b7cd2ad8a1d17ab4e6 (diff) |
Merge #6650
6650: Make completion and assists module independent r=matklad a=SomeoneToIgnore
A follow-up of https://github.com/rust-analyzer/rust-analyzer/pull/6553#discussion_r524402907
Move the common code for both assists and completion modules into a separate crate.
Co-authored-by: Kirill Bulatov <[email protected]>
Diffstat (limited to 'crates/ide_db/src/helpers')
-rw-r--r-- | crates/ide_db/src/helpers/insert_use.rs | 1204 |
1 files changed, 1204 insertions, 0 deletions
diff --git a/crates/ide_db/src/helpers/insert_use.rs b/crates/ide_db/src/helpers/insert_use.rs new file mode 100644 index 000000000..67e800fad --- /dev/null +++ b/crates/ide_db/src/helpers/insert_use.rs | |||
@@ -0,0 +1,1204 @@ | |||
1 | //! Handle syntactic aspects of inserting a new `use`. | ||
2 | use std::{cmp::Ordering, iter::successors}; | ||
3 | |||
4 | use crate::RootDatabase; | ||
5 | use hir::Semantics; | ||
6 | use itertools::{EitherOrBoth, Itertools}; | ||
7 | use syntax::{ | ||
8 | algo::SyntaxRewriter, | ||
9 | ast::{ | ||
10 | self, | ||
11 | edit::{AstNodeEdit, IndentLevel}, | ||
12 | make, AstNode, PathSegmentKind, VisibilityOwner, | ||
13 | }, | ||
14 | AstToken, InsertPosition, NodeOrToken, SyntaxElement, SyntaxNode, SyntaxToken, | ||
15 | }; | ||
16 | use test_utils::mark; | ||
17 | |||
18 | #[derive(Debug, Clone)] | ||
19 | pub enum ImportScope { | ||
20 | File(ast::SourceFile), | ||
21 | Module(ast::ItemList), | ||
22 | } | ||
23 | |||
24 | impl ImportScope { | ||
25 | pub fn from(syntax: SyntaxNode) -> Option<Self> { | ||
26 | if let Some(module) = ast::Module::cast(syntax.clone()) { | ||
27 | module.item_list().map(ImportScope::Module) | ||
28 | } else if let this @ Some(_) = ast::SourceFile::cast(syntax.clone()) { | ||
29 | this.map(ImportScope::File) | ||
30 | } else { | ||
31 | ast::ItemList::cast(syntax).map(ImportScope::Module) | ||
32 | } | ||
33 | } | ||
34 | |||
35 | /// Determines the containing syntax node in which to insert a `use` statement affecting `position`. | ||
36 | pub fn find_insert_use_container( | ||
37 | position: &SyntaxNode, | ||
38 | sema: &Semantics<'_, RootDatabase>, | ||
39 | ) -> Option<Self> { | ||
40 | sema.ancestors_with_macros(position.clone()).find_map(Self::from) | ||
41 | } | ||
42 | |||
43 | pub fn as_syntax_node(&self) -> &SyntaxNode { | ||
44 | match self { | ||
45 | ImportScope::File(file) => file.syntax(), | ||
46 | ImportScope::Module(item_list) => item_list.syntax(), | ||
47 | } | ||
48 | } | ||
49 | |||
50 | fn indent_level(&self) -> IndentLevel { | ||
51 | match self { | ||
52 | ImportScope::File(file) => file.indent_level(), | ||
53 | ImportScope::Module(item_list) => item_list.indent_level() + 1, | ||
54 | } | ||
55 | } | ||
56 | |||
57 | fn first_insert_pos(&self) -> (InsertPosition<SyntaxElement>, AddBlankLine) { | ||
58 | match self { | ||
59 | ImportScope::File(_) => (InsertPosition::First, AddBlankLine::AfterTwice), | ||
60 | // don't insert the imports before the item list's opening curly brace | ||
61 | ImportScope::Module(item_list) => item_list | ||
62 | .l_curly_token() | ||
63 | .map(|b| (InsertPosition::After(b.into()), AddBlankLine::Around)) | ||
64 | .unwrap_or((InsertPosition::First, AddBlankLine::AfterTwice)), | ||
65 | } | ||
66 | } | ||
67 | |||
68 | fn insert_pos_after_last_inner_element(&self) -> (InsertPosition<SyntaxElement>, AddBlankLine) { | ||
69 | self.as_syntax_node() | ||
70 | .children_with_tokens() | ||
71 | .filter(|child| match child { | ||
72 | NodeOrToken::Node(node) => is_inner_attribute(node.clone()), | ||
73 | NodeOrToken::Token(token) => is_inner_comment(token.clone()), | ||
74 | }) | ||
75 | .last() | ||
76 | .map(|last_inner_element| { | ||
77 | (InsertPosition::After(last_inner_element.into()), AddBlankLine::BeforeTwice) | ||
78 | }) | ||
79 | .unwrap_or_else(|| self.first_insert_pos()) | ||
80 | } | ||
81 | } | ||
82 | |||
83 | fn is_inner_attribute(node: SyntaxNode) -> bool { | ||
84 | ast::Attr::cast(node).map(|attr| attr.kind()) == Some(ast::AttrKind::Inner) | ||
85 | } | ||
86 | |||
87 | fn is_inner_comment(token: SyntaxToken) -> bool { | ||
88 | ast::Comment::cast(token).and_then(|comment| comment.kind().doc) | ||
89 | == Some(ast::CommentPlacement::Inner) | ||
90 | } | ||
91 | |||
92 | /// Insert an import path into the given file/node. A `merge` value of none indicates that no import merging is allowed to occur. | ||
93 | pub fn insert_use<'a>( | ||
94 | scope: &ImportScope, | ||
95 | path: ast::Path, | ||
96 | merge: Option<MergeBehaviour>, | ||
97 | ) -> SyntaxRewriter<'a> { | ||
98 | let _p = profile::span("insert_use"); | ||
99 | let mut rewriter = SyntaxRewriter::default(); | ||
100 | let use_item = make::use_(make::use_tree(path.clone(), None, None, false)); | ||
101 | // merge into existing imports if possible | ||
102 | if let Some(mb) = merge { | ||
103 | for existing_use in scope.as_syntax_node().children().filter_map(ast::Use::cast) { | ||
104 | if let Some(merged) = try_merge_imports(&existing_use, &use_item, mb) { | ||
105 | rewriter.replace(existing_use.syntax(), merged.syntax()); | ||
106 | return rewriter; | ||
107 | } | ||
108 | } | ||
109 | } | ||
110 | |||
111 | // either we weren't allowed to merge or there is no import that fits the merge conditions | ||
112 | // so look for the place we have to insert to | ||
113 | let (insert_position, add_blank) = find_insert_position(scope, path); | ||
114 | |||
115 | let indent = if let ident_level @ 1..=usize::MAX = scope.indent_level().0 as usize { | ||
116 | Some(make::tokens::whitespace(&" ".repeat(4 * ident_level)).into()) | ||
117 | } else { | ||
118 | None | ||
119 | }; | ||
120 | |||
121 | let to_insert: Vec<SyntaxElement> = { | ||
122 | let mut buf = Vec::new(); | ||
123 | |||
124 | match add_blank { | ||
125 | AddBlankLine::Before | AddBlankLine::Around => { | ||
126 | buf.push(make::tokens::single_newline().into()) | ||
127 | } | ||
128 | AddBlankLine::BeforeTwice => buf.push(make::tokens::blank_line().into()), | ||
129 | _ => (), | ||
130 | } | ||
131 | |||
132 | if add_blank.has_before() { | ||
133 | if let Some(indent) = indent.clone() { | ||
134 | mark::hit!(insert_use_indent_before); | ||
135 | buf.push(indent); | ||
136 | } | ||
137 | } | ||
138 | |||
139 | buf.push(use_item.syntax().clone().into()); | ||
140 | |||
141 | match add_blank { | ||
142 | AddBlankLine::After | AddBlankLine::Around => { | ||
143 | buf.push(make::tokens::single_newline().into()) | ||
144 | } | ||
145 | AddBlankLine::AfterTwice => buf.push(make::tokens::blank_line().into()), | ||
146 | _ => (), | ||
147 | } | ||
148 | |||
149 | // only add indentation *after* our stuff if there's another node directly after it | ||
150 | if add_blank.has_after() && matches!(insert_position, InsertPosition::Before(_)) { | ||
151 | if let Some(indent) = indent { | ||
152 | mark::hit!(insert_use_indent_after); | ||
153 | buf.push(indent); | ||
154 | } | ||
155 | } else if add_blank.has_after() && matches!(insert_position, InsertPosition::After(_)) { | ||
156 | mark::hit!(insert_use_no_indent_after); | ||
157 | } | ||
158 | |||
159 | buf | ||
160 | }; | ||
161 | |||
162 | match insert_position { | ||
163 | InsertPosition::First => { | ||
164 | rewriter.insert_many_as_first_children(scope.as_syntax_node(), to_insert) | ||
165 | } | ||
166 | InsertPosition::Last => return rewriter, // actually unreachable | ||
167 | InsertPosition::Before(anchor) => rewriter.insert_many_before(&anchor, to_insert), | ||
168 | InsertPosition::After(anchor) => rewriter.insert_many_after(&anchor, to_insert), | ||
169 | } | ||
170 | rewriter | ||
171 | } | ||
172 | |||
173 | fn eq_visibility(vis0: Option<ast::Visibility>, vis1: Option<ast::Visibility>) -> bool { | ||
174 | match (vis0, vis1) { | ||
175 | (None, None) => true, | ||
176 | // FIXME: Don't use the string representation to check for equality | ||
177 | // spaces inside of the node would break this comparison | ||
178 | (Some(vis0), Some(vis1)) => vis0.to_string() == vis1.to_string(), | ||
179 | _ => false, | ||
180 | } | ||
181 | } | ||
182 | |||
183 | pub fn try_merge_imports( | ||
184 | lhs: &ast::Use, | ||
185 | rhs: &ast::Use, | ||
186 | merge_behaviour: MergeBehaviour, | ||
187 | ) -> Option<ast::Use> { | ||
188 | // don't merge imports with different visibilities | ||
189 | if !eq_visibility(lhs.visibility(), rhs.visibility()) { | ||
190 | return None; | ||
191 | } | ||
192 | let lhs_tree = lhs.use_tree()?; | ||
193 | let rhs_tree = rhs.use_tree()?; | ||
194 | let merged = try_merge_trees(&lhs_tree, &rhs_tree, merge_behaviour)?; | ||
195 | Some(lhs.with_use_tree(merged)) | ||
196 | } | ||
197 | |||
198 | pub fn try_merge_trees( | ||
199 | lhs: &ast::UseTree, | ||
200 | rhs: &ast::UseTree, | ||
201 | merge: MergeBehaviour, | ||
202 | ) -> Option<ast::UseTree> { | ||
203 | let lhs_path = lhs.path()?; | ||
204 | let rhs_path = rhs.path()?; | ||
205 | |||
206 | let (lhs_prefix, rhs_prefix) = common_prefix(&lhs_path, &rhs_path)?; | ||
207 | let (lhs, rhs) = if is_simple_path(lhs) | ||
208 | && is_simple_path(rhs) | ||
209 | && lhs_path == lhs_prefix | ||
210 | && rhs_path == rhs_prefix | ||
211 | { | ||
212 | (lhs.clone(), rhs.clone()) | ||
213 | } else { | ||
214 | (lhs.split_prefix(&lhs_prefix), rhs.split_prefix(&rhs_prefix)) | ||
215 | }; | ||
216 | recursive_merge(&lhs, &rhs, merge) | ||
217 | } | ||
218 | |||
219 | /// Recursively "zips" together lhs and rhs. | ||
220 | fn recursive_merge( | ||
221 | lhs: &ast::UseTree, | ||
222 | rhs: &ast::UseTree, | ||
223 | merge: MergeBehaviour, | ||
224 | ) -> Option<ast::UseTree> { | ||
225 | let mut use_trees = lhs | ||
226 | .use_tree_list() | ||
227 | .into_iter() | ||
228 | .flat_map(|list| list.use_trees()) | ||
229 | // we use Option here to early return from this function(this is not the same as a `filter` op) | ||
230 | .map(|tree| match merge.is_tree_allowed(&tree) { | ||
231 | true => Some(tree), | ||
232 | false => None, | ||
233 | }) | ||
234 | .collect::<Option<Vec<_>>>()?; | ||
235 | use_trees.sort_unstable_by(|a, b| path_cmp_for_sort(a.path(), b.path())); | ||
236 | for rhs_t in rhs.use_tree_list().into_iter().flat_map(|list| list.use_trees()) { | ||
237 | if !merge.is_tree_allowed(&rhs_t) { | ||
238 | return None; | ||
239 | } | ||
240 | let rhs_path = rhs_t.path(); | ||
241 | match use_trees.binary_search_by(|lhs_t| { | ||
242 | let (lhs_t, rhs_t) = match lhs_t | ||
243 | .path() | ||
244 | .zip(rhs_path.clone()) | ||
245 | .and_then(|(lhs, rhs)| common_prefix(&lhs, &rhs)) | ||
246 | { | ||
247 | Some((lhs_p, rhs_p)) => (lhs_t.split_prefix(&lhs_p), rhs_t.split_prefix(&rhs_p)), | ||
248 | None => (lhs_t.clone(), rhs_t.clone()), | ||
249 | }; | ||
250 | |||
251 | path_cmp_bin_search(lhs_t.path(), rhs_t.path()) | ||
252 | }) { | ||
253 | Ok(idx) => { | ||
254 | let lhs_t = &mut use_trees[idx]; | ||
255 | let lhs_path = lhs_t.path()?; | ||
256 | let rhs_path = rhs_path?; | ||
257 | let (lhs_prefix, rhs_prefix) = common_prefix(&lhs_path, &rhs_path)?; | ||
258 | if lhs_prefix == lhs_path && rhs_prefix == rhs_path { | ||
259 | let tree_is_self = |tree: ast::UseTree| { | ||
260 | tree.path().as_ref().map(path_is_self).unwrap_or(false) | ||
261 | }; | ||
262 | // check if only one of the two trees has a tree list, and whether that then contains `self` or not. | ||
263 | // If this is the case we can skip this iteration since the path without the list is already included in the other one via `self` | ||
264 | let tree_contains_self = |tree: &ast::UseTree| { | ||
265 | tree.use_tree_list() | ||
266 | .map(|tree_list| tree_list.use_trees().any(tree_is_self)) | ||
267 | .unwrap_or(false) | ||
268 | }; | ||
269 | match (tree_contains_self(&lhs_t), tree_contains_self(&rhs_t)) { | ||
270 | (true, false) => continue, | ||
271 | (false, true) => { | ||
272 | *lhs_t = rhs_t; | ||
273 | continue; | ||
274 | } | ||
275 | _ => (), | ||
276 | } | ||
277 | |||
278 | // glob imports arent part of the use-tree lists so we need to special handle them here as well | ||
279 | // this special handling is only required for when we merge a module import into a glob import of said module | ||
280 | // see the `merge_self_glob` or `merge_mod_into_glob` tests | ||
281 | if lhs_t.star_token().is_some() || rhs_t.star_token().is_some() { | ||
282 | *lhs_t = make::use_tree( | ||
283 | make::path_unqualified(make::path_segment_self()), | ||
284 | None, | ||
285 | None, | ||
286 | false, | ||
287 | ); | ||
288 | use_trees.insert(idx, make::glob_use_tree()); | ||
289 | continue; | ||
290 | } | ||
291 | |||
292 | if lhs_t.use_tree_list().is_none() && rhs_t.use_tree_list().is_none() { | ||
293 | continue; | ||
294 | } | ||
295 | } | ||
296 | let lhs = lhs_t.split_prefix(&lhs_prefix); | ||
297 | let rhs = rhs_t.split_prefix(&rhs_prefix); | ||
298 | match recursive_merge(&lhs, &rhs, merge) { | ||
299 | Some(use_tree) => use_trees[idx] = use_tree, | ||
300 | None => return None, | ||
301 | } | ||
302 | } | ||
303 | Err(_) | ||
304 | if merge == MergeBehaviour::Last | ||
305 | && use_trees.len() > 0 | ||
306 | && rhs_t.use_tree_list().is_some() => | ||
307 | { | ||
308 | return None | ||
309 | } | ||
310 | Err(idx) => { | ||
311 | use_trees.insert(idx, rhs_t); | ||
312 | } | ||
313 | } | ||
314 | } | ||
315 | Some(lhs.with_use_tree_list(make::use_tree_list(use_trees))) | ||
316 | } | ||
317 | |||
318 | /// Traverses both paths until they differ, returning the common prefix of both. | ||
319 | fn common_prefix(lhs: &ast::Path, rhs: &ast::Path) -> Option<(ast::Path, ast::Path)> { | ||
320 | let mut res = None; | ||
321 | let mut lhs_curr = first_path(&lhs); | ||
322 | let mut rhs_curr = first_path(&rhs); | ||
323 | loop { | ||
324 | match (lhs_curr.segment(), rhs_curr.segment()) { | ||
325 | (Some(lhs), Some(rhs)) if lhs.syntax().text() == rhs.syntax().text() => (), | ||
326 | _ => break res, | ||
327 | } | ||
328 | res = Some((lhs_curr.clone(), rhs_curr.clone())); | ||
329 | |||
330 | match lhs_curr.parent_path().zip(rhs_curr.parent_path()) { | ||
331 | Some((lhs, rhs)) => { | ||
332 | lhs_curr = lhs; | ||
333 | rhs_curr = rhs; | ||
334 | } | ||
335 | _ => break res, | ||
336 | } | ||
337 | } | ||
338 | } | ||
339 | |||
340 | fn is_simple_path(use_tree: &ast::UseTree) -> bool { | ||
341 | use_tree.use_tree_list().is_none() && use_tree.star_token().is_none() | ||
342 | } | ||
343 | |||
344 | fn path_is_self(path: &ast::Path) -> bool { | ||
345 | path.segment().and_then(|seg| seg.self_token()).is_some() && path.qualifier().is_none() | ||
346 | } | ||
347 | |||
348 | #[inline] | ||
349 | fn first_segment(path: &ast::Path) -> Option<ast::PathSegment> { | ||
350 | first_path(path).segment() | ||
351 | } | ||
352 | |||
353 | fn first_path(path: &ast::Path) -> ast::Path { | ||
354 | successors(Some(path.clone()), ast::Path::qualifier).last().unwrap() | ||
355 | } | ||
356 | |||
357 | fn segment_iter(path: &ast::Path) -> impl Iterator<Item = ast::PathSegment> + Clone { | ||
358 | // cant make use of SyntaxNode::siblings, because the returned Iterator is not clone | ||
359 | successors(first_segment(path), |p| p.parent_path().parent_path().and_then(|p| p.segment())) | ||
360 | } | ||
361 | |||
362 | fn path_len(path: ast::Path) -> usize { | ||
363 | segment_iter(&path).count() | ||
364 | } | ||
365 | |||
366 | /// Orders paths in the following way: | ||
367 | /// the sole self token comes first, after that come uppercase identifiers, then lowercase identifiers | ||
368 | // FIXME: rustfmt sorts lowercase idents before uppercase, in general we want to have the same ordering rustfmt has | ||
369 | // which is `self` and `super` first, then identifier imports with lowercase ones first, then glob imports and at last list imports. | ||
370 | // Example foo::{self, foo, baz, Baz, Qux, *, {Bar}} | ||
371 | fn path_cmp_for_sort(a: Option<ast::Path>, b: Option<ast::Path>) -> Ordering { | ||
372 | match (a, b) { | ||
373 | (None, None) => Ordering::Equal, | ||
374 | (None, Some(_)) => Ordering::Less, | ||
375 | (Some(_), None) => Ordering::Greater, | ||
376 | (Some(ref a), Some(ref b)) => match (path_is_self(a), path_is_self(b)) { | ||
377 | (true, true) => Ordering::Equal, | ||
378 | (true, false) => Ordering::Less, | ||
379 | (false, true) => Ordering::Greater, | ||
380 | (false, false) => path_cmp_short(a, b), | ||
381 | }, | ||
382 | } | ||
383 | } | ||
384 | |||
385 | /// Path comparison func for binary searching for merging. | ||
386 | fn path_cmp_bin_search(lhs: Option<ast::Path>, rhs: Option<ast::Path>) -> Ordering { | ||
387 | match (lhs.and_then(|path| path.segment()), rhs.and_then(|path| path.segment())) { | ||
388 | (None, None) => Ordering::Equal, | ||
389 | (None, Some(_)) => Ordering::Less, | ||
390 | (Some(_), None) => Ordering::Greater, | ||
391 | (Some(ref a), Some(ref b)) => path_segment_cmp(a, b), | ||
392 | } | ||
393 | } | ||
394 | |||
395 | /// Short circuiting comparison, if both paths are equal until one of them ends they are considered | ||
396 | /// equal | ||
397 | fn path_cmp_short(a: &ast::Path, b: &ast::Path) -> Ordering { | ||
398 | let a = segment_iter(a); | ||
399 | let b = segment_iter(b); | ||
400 | // cmp_by would be useful for us here but that is currently unstable | ||
401 | // cmp doesnt work due the lifetimes on text's return type | ||
402 | a.zip(b) | ||
403 | .find_map(|(a, b)| match path_segment_cmp(&a, &b) { | ||
404 | Ordering::Equal => None, | ||
405 | ord => Some(ord), | ||
406 | }) | ||
407 | .unwrap_or(Ordering::Equal) | ||
408 | } | ||
409 | |||
410 | /// Compares to paths, if one ends earlier than the other the has_tl parameters decide which is | ||
411 | /// greater as a a path that has a tree list should be greater, while one that just ends without | ||
412 | /// a tree list should be considered less. | ||
413 | fn use_tree_path_cmp(a: &ast::Path, a_has_tl: bool, b: &ast::Path, b_has_tl: bool) -> Ordering { | ||
414 | let a_segments = segment_iter(a); | ||
415 | let b_segments = segment_iter(b); | ||
416 | // cmp_by would be useful for us here but that is currently unstable | ||
417 | // cmp doesnt work due the lifetimes on text's return type | ||
418 | a_segments | ||
419 | .zip_longest(b_segments) | ||
420 | .find_map(|zipped| match zipped { | ||
421 | EitherOrBoth::Both(ref a, ref b) => match path_segment_cmp(a, b) { | ||
422 | Ordering::Equal => None, | ||
423 | ord => Some(ord), | ||
424 | }, | ||
425 | EitherOrBoth::Left(_) if !b_has_tl => Some(Ordering::Greater), | ||
426 | EitherOrBoth::Left(_) => Some(Ordering::Less), | ||
427 | EitherOrBoth::Right(_) if !a_has_tl => Some(Ordering::Less), | ||
428 | EitherOrBoth::Right(_) => Some(Ordering::Greater), | ||
429 | }) | ||
430 | .unwrap_or(Ordering::Equal) | ||
431 | } | ||
432 | |||
433 | fn path_segment_cmp(a: &ast::PathSegment, b: &ast::PathSegment) -> Ordering { | ||
434 | let a = a.name_ref(); | ||
435 | let b = b.name_ref(); | ||
436 | a.as_ref().map(ast::NameRef::text).cmp(&b.as_ref().map(ast::NameRef::text)) | ||
437 | } | ||
438 | |||
439 | /// What type of merges are allowed. | ||
440 | #[derive(Copy, Clone, Debug, PartialEq, Eq)] | ||
441 | pub enum MergeBehaviour { | ||
442 | /// Merge everything together creating deeply nested imports. | ||
443 | Full, | ||
444 | /// Only merge the last import level, doesn't allow import nesting. | ||
445 | Last, | ||
446 | } | ||
447 | |||
448 | impl MergeBehaviour { | ||
449 | #[inline] | ||
450 | fn is_tree_allowed(&self, tree: &ast::UseTree) -> bool { | ||
451 | match self { | ||
452 | MergeBehaviour::Full => true, | ||
453 | // only simple single segment paths are allowed | ||
454 | MergeBehaviour::Last => { | ||
455 | tree.use_tree_list().is_none() && tree.path().map(path_len) <= Some(1) | ||
456 | } | ||
457 | } | ||
458 | } | ||
459 | } | ||
460 | |||
461 | #[derive(Eq, PartialEq, PartialOrd, Ord)] | ||
462 | enum ImportGroup { | ||
463 | // the order here defines the order of new group inserts | ||
464 | Std, | ||
465 | ExternCrate, | ||
466 | ThisCrate, | ||
467 | ThisModule, | ||
468 | SuperModule, | ||
469 | } | ||
470 | |||
471 | impl ImportGroup { | ||
472 | fn new(path: &ast::Path) -> ImportGroup { | ||
473 | let default = ImportGroup::ExternCrate; | ||
474 | |||
475 | let first_segment = match first_segment(path) { | ||
476 | Some(it) => it, | ||
477 | None => return default, | ||
478 | }; | ||
479 | |||
480 | let kind = first_segment.kind().unwrap_or(PathSegmentKind::SelfKw); | ||
481 | match kind { | ||
482 | PathSegmentKind::SelfKw => ImportGroup::ThisModule, | ||
483 | PathSegmentKind::SuperKw => ImportGroup::SuperModule, | ||
484 | PathSegmentKind::CrateKw => ImportGroup::ThisCrate, | ||
485 | PathSegmentKind::Name(name) => match name.text().as_str() { | ||
486 | "std" => ImportGroup::Std, | ||
487 | "core" => ImportGroup::Std, | ||
488 | _ => ImportGroup::ExternCrate, | ||
489 | }, | ||
490 | PathSegmentKind::Type { .. } => unreachable!(), | ||
491 | } | ||
492 | } | ||
493 | } | ||
494 | |||
495 | #[derive(PartialEq, Eq)] | ||
496 | enum AddBlankLine { | ||
497 | Before, | ||
498 | BeforeTwice, | ||
499 | Around, | ||
500 | After, | ||
501 | AfterTwice, | ||
502 | } | ||
503 | |||
504 | impl AddBlankLine { | ||
505 | fn has_before(&self) -> bool { | ||
506 | matches!(self, AddBlankLine::Before | AddBlankLine::BeforeTwice | AddBlankLine::Around) | ||
507 | } | ||
508 | fn has_after(&self) -> bool { | ||
509 | matches!(self, AddBlankLine::After | AddBlankLine::AfterTwice | AddBlankLine::Around) | ||
510 | } | ||
511 | } | ||
512 | |||
513 | fn find_insert_position( | ||
514 | scope: &ImportScope, | ||
515 | insert_path: ast::Path, | ||
516 | ) -> (InsertPosition<SyntaxElement>, AddBlankLine) { | ||
517 | let group = ImportGroup::new(&insert_path); | ||
518 | let path_node_iter = scope | ||
519 | .as_syntax_node() | ||
520 | .children() | ||
521 | .filter_map(|node| ast::Use::cast(node.clone()).zip(Some(node))) | ||
522 | .flat_map(|(use_, node)| { | ||
523 | let tree = use_.use_tree()?; | ||
524 | let path = tree.path()?; | ||
525 | let has_tl = tree.use_tree_list().is_some(); | ||
526 | Some((path, has_tl, node)) | ||
527 | }); | ||
528 | // Iterator that discards anything thats not in the required grouping | ||
529 | // This implementation allows the user to rearrange their import groups as this only takes the first group that fits | ||
530 | let group_iter = path_node_iter | ||
531 | .clone() | ||
532 | .skip_while(|(path, ..)| ImportGroup::new(path) != group) | ||
533 | .take_while(|(path, ..)| ImportGroup::new(path) == group); | ||
534 | |||
535 | // track the last element we iterated over, if this is still None after the iteration then that means we never iterated in the first place | ||
536 | let mut last = None; | ||
537 | // find the element that would come directly after our new import | ||
538 | let post_insert = group_iter.inspect(|(.., node)| last = Some(node.clone())).find( | ||
539 | |&(ref path, has_tl, _)| { | ||
540 | use_tree_path_cmp(&insert_path, false, path, has_tl) != Ordering::Greater | ||
541 | }, | ||
542 | ); | ||
543 | match post_insert { | ||
544 | // insert our import before that element | ||
545 | Some((.., node)) => (InsertPosition::Before(node.into()), AddBlankLine::After), | ||
546 | // there is no element after our new import, so append it to the end of the group | ||
547 | None => match last { | ||
548 | Some(node) => (InsertPosition::After(node.into()), AddBlankLine::Before), | ||
549 | // the group we were looking for actually doesnt exist, so insert | ||
550 | None => { | ||
551 | // similar concept here to the `last` from above | ||
552 | let mut last = None; | ||
553 | // find the group that comes after where we want to insert | ||
554 | let post_group = path_node_iter | ||
555 | .inspect(|(.., node)| last = Some(node.clone())) | ||
556 | .find(|(p, ..)| ImportGroup::new(p) > group); | ||
557 | match post_group { | ||
558 | Some((.., node)) => { | ||
559 | (InsertPosition::Before(node.into()), AddBlankLine::AfterTwice) | ||
560 | } | ||
561 | // there is no such group, so append after the last one | ||
562 | None => match last { | ||
563 | Some(node) => { | ||
564 | (InsertPosition::After(node.into()), AddBlankLine::BeforeTwice) | ||
565 | } | ||
566 | // there are no imports in this file at all | ||
567 | None => scope.insert_pos_after_last_inner_element(), | ||
568 | }, | ||
569 | } | ||
570 | } | ||
571 | }, | ||
572 | } | ||
573 | } | ||
574 | |||
575 | #[cfg(test)] | ||
576 | mod tests { | ||
577 | use super::*; | ||
578 | |||
579 | use test_utils::assert_eq_text; | ||
580 | |||
581 | #[test] | ||
582 | fn insert_existing() { | ||
583 | check_full("std::fs", "use std::fs;", "use std::fs;") | ||
584 | } | ||
585 | |||
586 | #[test] | ||
587 | fn insert_start() { | ||
588 | check_none( | ||
589 | "std::bar::AA", | ||
590 | r" | ||
591 | use std::bar::B; | ||
592 | use std::bar::D; | ||
593 | use std::bar::F; | ||
594 | use std::bar::G;", | ||
595 | r" | ||
596 | use std::bar::AA; | ||
597 | use std::bar::B; | ||
598 | use std::bar::D; | ||
599 | use std::bar::F; | ||
600 | use std::bar::G;", | ||
601 | ) | ||
602 | } | ||
603 | |||
604 | #[test] | ||
605 | fn insert_start_indent() { | ||
606 | mark::check!(insert_use_indent_after); | ||
607 | check_none( | ||
608 | "std::bar::AA", | ||
609 | r" | ||
610 | use std::bar::B; | ||
611 | use std::bar::D;", | ||
612 | r" | ||
613 | use std::bar::AA; | ||
614 | use std::bar::B; | ||
615 | use std::bar::D;", | ||
616 | ) | ||
617 | } | ||
618 | |||
619 | #[test] | ||
620 | fn insert_middle() { | ||
621 | check_none( | ||
622 | "std::bar::EE", | ||
623 | r" | ||
624 | use std::bar::A; | ||
625 | use std::bar::D; | ||
626 | use std::bar::F; | ||
627 | use std::bar::G;", | ||
628 | r" | ||
629 | use std::bar::A; | ||
630 | use std::bar::D; | ||
631 | use std::bar::EE; | ||
632 | use std::bar::F; | ||
633 | use std::bar::G;", | ||
634 | ) | ||
635 | } | ||
636 | |||
637 | #[test] | ||
638 | fn insert_middle_indent() { | ||
639 | check_none( | ||
640 | "std::bar::EE", | ||
641 | r" | ||
642 | use std::bar::A; | ||
643 | use std::bar::D; | ||
644 | use std::bar::F; | ||
645 | use std::bar::G;", | ||
646 | r" | ||
647 | use std::bar::A; | ||
648 | use std::bar::D; | ||
649 | use std::bar::EE; | ||
650 | use std::bar::F; | ||
651 | use std::bar::G;", | ||
652 | ) | ||
653 | } | ||
654 | |||
655 | #[test] | ||
656 | fn insert_end() { | ||
657 | check_none( | ||
658 | "std::bar::ZZ", | ||
659 | r" | ||
660 | use std::bar::A; | ||
661 | use std::bar::D; | ||
662 | use std::bar::F; | ||
663 | use std::bar::G;", | ||
664 | r" | ||
665 | use std::bar::A; | ||
666 | use std::bar::D; | ||
667 | use std::bar::F; | ||
668 | use std::bar::G; | ||
669 | use std::bar::ZZ;", | ||
670 | ) | ||
671 | } | ||
672 | |||
673 | #[test] | ||
674 | fn insert_end_indent() { | ||
675 | mark::check!(insert_use_indent_before); | ||
676 | check_none( | ||
677 | "std::bar::ZZ", | ||
678 | r" | ||
679 | use std::bar::A; | ||
680 | use std::bar::D; | ||
681 | use std::bar::F; | ||
682 | use std::bar::G;", | ||
683 | r" | ||
684 | use std::bar::A; | ||
685 | use std::bar::D; | ||
686 | use std::bar::F; | ||
687 | use std::bar::G; | ||
688 | use std::bar::ZZ;", | ||
689 | ) | ||
690 | } | ||
691 | |||
692 | #[test] | ||
693 | fn insert_middle_nested() { | ||
694 | check_none( | ||
695 | "std::bar::EE", | ||
696 | r" | ||
697 | use std::bar::A; | ||
698 | use std::bar::{D, Z}; // example of weird imports due to user | ||
699 | use std::bar::F; | ||
700 | use std::bar::G;", | ||
701 | r" | ||
702 | use std::bar::A; | ||
703 | use std::bar::EE; | ||
704 | use std::bar::{D, Z}; // example of weird imports due to user | ||
705 | use std::bar::F; | ||
706 | use std::bar::G;", | ||
707 | ) | ||
708 | } | ||
709 | |||
710 | #[test] | ||
711 | fn insert_middle_groups() { | ||
712 | check_none( | ||
713 | "foo::bar::GG", | ||
714 | r" | ||
715 | use std::bar::A; | ||
716 | use std::bar::D; | ||
717 | |||
718 | use foo::bar::F; | ||
719 | use foo::bar::H;", | ||
720 | r" | ||
721 | use std::bar::A; | ||
722 | use std::bar::D; | ||
723 | |||
724 | use foo::bar::F; | ||
725 | use foo::bar::GG; | ||
726 | use foo::bar::H;", | ||
727 | ) | ||
728 | } | ||
729 | |||
730 | #[test] | ||
731 | fn insert_first_matching_group() { | ||
732 | check_none( | ||
733 | "foo::bar::GG", | ||
734 | r" | ||
735 | use foo::bar::A; | ||
736 | use foo::bar::D; | ||
737 | |||
738 | use std; | ||
739 | |||
740 | use foo::bar::F; | ||
741 | use foo::bar::H;", | ||
742 | r" | ||
743 | use foo::bar::A; | ||
744 | use foo::bar::D; | ||
745 | use foo::bar::GG; | ||
746 | |||
747 | use std; | ||
748 | |||
749 | use foo::bar::F; | ||
750 | use foo::bar::H;", | ||
751 | ) | ||
752 | } | ||
753 | |||
754 | #[test] | ||
755 | fn insert_missing_group_std() { | ||
756 | check_none( | ||
757 | "std::fmt", | ||
758 | r" | ||
759 | use foo::bar::A; | ||
760 | use foo::bar::D;", | ||
761 | r" | ||
762 | use std::fmt; | ||
763 | |||
764 | use foo::bar::A; | ||
765 | use foo::bar::D;", | ||
766 | ) | ||
767 | } | ||
768 | |||
769 | #[test] | ||
770 | fn insert_missing_group_self() { | ||
771 | check_none( | ||
772 | "self::fmt", | ||
773 | r" | ||
774 | use foo::bar::A; | ||
775 | use foo::bar::D;", | ||
776 | r" | ||
777 | use foo::bar::A; | ||
778 | use foo::bar::D; | ||
779 | |||
780 | use self::fmt;", | ||
781 | ) | ||
782 | } | ||
783 | |||
784 | #[test] | ||
785 | fn insert_no_imports() { | ||
786 | check_full( | ||
787 | "foo::bar", | ||
788 | "fn main() {}", | ||
789 | r"use foo::bar; | ||
790 | |||
791 | fn main() {}", | ||
792 | ) | ||
793 | } | ||
794 | |||
795 | #[test] | ||
796 | fn insert_empty_file() { | ||
797 | // empty files will get two trailing newlines | ||
798 | // this is due to the test case insert_no_imports above | ||
799 | check_full( | ||
800 | "foo::bar", | ||
801 | "", | ||
802 | r"use foo::bar; | ||
803 | |||
804 | ", | ||
805 | ) | ||
806 | } | ||
807 | |||
808 | #[test] | ||
809 | fn insert_empty_module() { | ||
810 | mark::check!(insert_use_no_indent_after); | ||
811 | check( | ||
812 | "foo::bar", | ||
813 | "mod x {}", | ||
814 | r"{ | ||
815 | use foo::bar; | ||
816 | }", | ||
817 | None, | ||
818 | true, | ||
819 | ) | ||
820 | } | ||
821 | |||
822 | #[test] | ||
823 | fn insert_after_inner_attr() { | ||
824 | check_full( | ||
825 | "foo::bar", | ||
826 | r"#![allow(unused_imports)]", | ||
827 | r"#![allow(unused_imports)] | ||
828 | |||
829 | use foo::bar;", | ||
830 | ) | ||
831 | } | ||
832 | |||
833 | #[test] | ||
834 | fn insert_after_inner_attr2() { | ||
835 | check_full( | ||
836 | "foo::bar", | ||
837 | r"#![allow(unused_imports)] | ||
838 | |||
839 | #![no_std] | ||
840 | fn main() {}", | ||
841 | r"#![allow(unused_imports)] | ||
842 | |||
843 | #![no_std] | ||
844 | |||
845 | use foo::bar; | ||
846 | fn main() {}", | ||
847 | ); | ||
848 | } | ||
849 | |||
850 | #[test] | ||
851 | fn inserts_after_single_line_inner_comments() { | ||
852 | check_none( | ||
853 | "foo::bar::Baz", | ||
854 | "//! Single line inner comments do not allow any code before them.", | ||
855 | r#"//! Single line inner comments do not allow any code before them. | ||
856 | |||
857 | use foo::bar::Baz;"#, | ||
858 | ); | ||
859 | } | ||
860 | |||
861 | #[test] | ||
862 | fn inserts_after_multiline_inner_comments() { | ||
863 | check_none( | ||
864 | "foo::bar::Baz", | ||
865 | r#"/*! Multiline inner comments do not allow any code before them. */ | ||
866 | |||
867 | /*! Still an inner comment, cannot place any code before. */ | ||
868 | fn main() {}"#, | ||
869 | r#"/*! Multiline inner comments do not allow any code before them. */ | ||
870 | |||
871 | /*! Still an inner comment, cannot place any code before. */ | ||
872 | |||
873 | use foo::bar::Baz; | ||
874 | fn main() {}"#, | ||
875 | ) | ||
876 | } | ||
877 | |||
878 | #[test] | ||
879 | fn inserts_after_all_inner_items() { | ||
880 | check_none( | ||
881 | "foo::bar::Baz", | ||
882 | r#"#![allow(unused_imports)] | ||
883 | /*! Multiline line comment 2 */ | ||
884 | |||
885 | |||
886 | //! Single line comment 1 | ||
887 | #![no_std] | ||
888 | //! Single line comment 2 | ||
889 | fn main() {}"#, | ||
890 | r#"#![allow(unused_imports)] | ||
891 | /*! Multiline line comment 2 */ | ||
892 | |||
893 | |||
894 | //! Single line comment 1 | ||
895 | #![no_std] | ||
896 | //! Single line comment 2 | ||
897 | |||
898 | use foo::bar::Baz; | ||
899 | fn main() {}"#, | ||
900 | ) | ||
901 | } | ||
902 | |||
903 | #[test] | ||
904 | fn merge_groups() { | ||
905 | check_last("std::io", r"use std::fmt;", r"use std::{fmt, io};") | ||
906 | } | ||
907 | |||
908 | #[test] | ||
909 | fn merge_groups_last() { | ||
910 | check_last( | ||
911 | "std::io", | ||
912 | r"use std::fmt::{Result, Display};", | ||
913 | r"use std::fmt::{Result, Display}; | ||
914 | use std::io;", | ||
915 | ) | ||
916 | } | ||
917 | |||
918 | #[test] | ||
919 | fn merge_last_into_self() { | ||
920 | check_last("foo::bar::baz", r"use foo::bar;", r"use foo::bar::{self, baz};"); | ||
921 | } | ||
922 | |||
923 | #[test] | ||
924 | fn merge_groups_full() { | ||
925 | check_full( | ||
926 | "std::io", | ||
927 | r"use std::fmt::{Result, Display};", | ||
928 | r"use std::{fmt::{Result, Display}, io};", | ||
929 | ) | ||
930 | } | ||
931 | |||
932 | #[test] | ||
933 | fn merge_groups_long_full() { | ||
934 | check_full( | ||
935 | "std::foo::bar::Baz", | ||
936 | r"use std::foo::bar::Qux;", | ||
937 | r"use std::foo::bar::{Baz, Qux};", | ||
938 | ) | ||
939 | } | ||
940 | |||
941 | #[test] | ||
942 | fn merge_groups_long_last() { | ||
943 | check_last( | ||
944 | "std::foo::bar::Baz", | ||
945 | r"use std::foo::bar::Qux;", | ||
946 | r"use std::foo::bar::{Baz, Qux};", | ||
947 | ) | ||
948 | } | ||
949 | |||
950 | #[test] | ||
951 | fn merge_groups_long_full_list() { | ||
952 | check_full( | ||
953 | "std::foo::bar::Baz", | ||
954 | r"use std::foo::bar::{Qux, Quux};", | ||
955 | r"use std::foo::bar::{Baz, Quux, Qux};", | ||
956 | ) | ||
957 | } | ||
958 | |||
959 | #[test] | ||
960 | fn merge_groups_long_last_list() { | ||
961 | check_last( | ||
962 | "std::foo::bar::Baz", | ||
963 | r"use std::foo::bar::{Qux, Quux};", | ||
964 | r"use std::foo::bar::{Baz, Quux, Qux};", | ||
965 | ) | ||
966 | } | ||
967 | |||
968 | #[test] | ||
969 | fn merge_groups_long_full_nested() { | ||
970 | check_full( | ||
971 | "std::foo::bar::Baz", | ||
972 | r"use std::foo::bar::{Qux, quux::{Fez, Fizz}};", | ||
973 | r"use std::foo::bar::{Baz, Qux, quux::{Fez, Fizz}};", | ||
974 | ) | ||
975 | } | ||
976 | |||
977 | #[test] | ||
978 | fn merge_groups_long_last_nested() { | ||
979 | check_last( | ||
980 | "std::foo::bar::Baz", | ||
981 | r"use std::foo::bar::{Qux, quux::{Fez, Fizz}};", | ||
982 | r"use std::foo::bar::Baz; | ||
983 | use std::foo::bar::{Qux, quux::{Fez, Fizz}};", | ||
984 | ) | ||
985 | } | ||
986 | |||
987 | #[test] | ||
988 | fn merge_groups_full_nested_deep() { | ||
989 | check_full( | ||
990 | "std::foo::bar::quux::Baz", | ||
991 | r"use std::foo::bar::{Qux, quux::{Fez, Fizz}};", | ||
992 | r"use std::foo::bar::{Qux, quux::{Baz, Fez, Fizz}};", | ||
993 | ) | ||
994 | } | ||
995 | |||
996 | #[test] | ||
997 | fn merge_groups_full_nested_long() { | ||
998 | check_full( | ||
999 | "std::foo::bar::Baz", | ||
1000 | r"use std::{foo::bar::Qux};", | ||
1001 | r"use std::{foo::bar::{Baz, Qux}};", | ||
1002 | ); | ||
1003 | } | ||
1004 | |||
1005 | #[test] | ||
1006 | fn merge_groups_last_nested_long() { | ||
1007 | check_full( | ||
1008 | "std::foo::bar::Baz", | ||
1009 | r"use std::{foo::bar::Qux};", | ||
1010 | r"use std::{foo::bar::{Baz, Qux}};", | ||
1011 | ); | ||
1012 | } | ||
1013 | |||
1014 | #[test] | ||
1015 | fn merge_groups_skip_pub() { | ||
1016 | check_full( | ||
1017 | "std::io", | ||
1018 | r"pub use std::fmt::{Result, Display};", | ||
1019 | r"pub use std::fmt::{Result, Display}; | ||
1020 | use std::io;", | ||
1021 | ) | ||
1022 | } | ||
1023 | |||
1024 | #[test] | ||
1025 | fn merge_groups_skip_pub_crate() { | ||
1026 | check_full( | ||
1027 | "std::io", | ||
1028 | r"pub(crate) use std::fmt::{Result, Display};", | ||
1029 | r"pub(crate) use std::fmt::{Result, Display}; | ||
1030 | use std::io;", | ||
1031 | ) | ||
1032 | } | ||
1033 | |||
1034 | #[test] | ||
1035 | #[ignore] // FIXME: Support this | ||
1036 | fn split_out_merge() { | ||
1037 | check_last( | ||
1038 | "std::fmt::Result", | ||
1039 | r"use std::{fmt, io};", | ||
1040 | r"use std::fmt::{self, Result}; | ||
1041 | use std::io;", | ||
1042 | ) | ||
1043 | } | ||
1044 | |||
1045 | #[test] | ||
1046 | fn merge_into_module_import() { | ||
1047 | check_full( | ||
1048 | "std::fmt::Result", | ||
1049 | r"use std::{fmt, io};", | ||
1050 | r"use std::{fmt::{self, Result}, io};", | ||
1051 | ) | ||
1052 | } | ||
1053 | |||
1054 | #[test] | ||
1055 | fn merge_groups_self() { | ||
1056 | check_full("std::fmt::Debug", r"use std::fmt;", r"use std::fmt::{self, Debug};") | ||
1057 | } | ||
1058 | |||
1059 | #[test] | ||
1060 | fn merge_mod_into_glob() { | ||
1061 | check_full( | ||
1062 | "token::TokenKind", | ||
1063 | r"use token::TokenKind::*;", | ||
1064 | r"use token::TokenKind::{*, self};", | ||
1065 | ) | ||
1066 | // FIXME: have it emit `use token::TokenKind::{self, *}`? | ||
1067 | } | ||
1068 | |||
1069 | #[test] | ||
1070 | fn merge_self_glob() { | ||
1071 | check_full("self", r"use self::*;", r"use self::{*, self};") | ||
1072 | // FIXME: have it emit `use {self, *}`? | ||
1073 | } | ||
1074 | |||
1075 | #[test] | ||
1076 | fn merge_glob_nested() { | ||
1077 | check_full( | ||
1078 | "foo::bar::quux::Fez", | ||
1079 | r"use foo::bar::{Baz, quux::*};", | ||
1080 | r"use foo::bar::{Baz, quux::{self::*, Fez}};", | ||
1081 | ) | ||
1082 | } | ||
1083 | |||
1084 | #[test] | ||
1085 | fn skip_merge_last_too_long() { | ||
1086 | check_last( | ||
1087 | "foo::bar", | ||
1088 | r"use foo::bar::baz::Qux;", | ||
1089 | r"use foo::bar; | ||
1090 | use foo::bar::baz::Qux;", | ||
1091 | ); | ||
1092 | } | ||
1093 | |||
1094 | #[test] | ||
1095 | fn skip_merge_last_too_long2() { | ||
1096 | check_last( | ||
1097 | "foo::bar::baz::Qux", | ||
1098 | r"use foo::bar;", | ||
1099 | r"use foo::bar; | ||
1100 | use foo::bar::baz::Qux;", | ||
1101 | ); | ||
1102 | } | ||
1103 | |||
1104 | #[test] | ||
1105 | fn insert_short_before_long() { | ||
1106 | check_none( | ||
1107 | "foo::bar", | ||
1108 | r"use foo::bar::baz::Qux;", | ||
1109 | r"use foo::bar; | ||
1110 | use foo::bar::baz::Qux;", | ||
1111 | ); | ||
1112 | } | ||
1113 | |||
1114 | #[test] | ||
1115 | fn merge_last_fail() { | ||
1116 | check_merge_only_fail( | ||
1117 | r"use foo::bar::{baz::{Qux, Fez}};", | ||
1118 | r"use foo::bar::{baaz::{Quux, Feez}};", | ||
1119 | MergeBehaviour::Last, | ||
1120 | ); | ||
1121 | } | ||
1122 | |||
1123 | #[test] | ||
1124 | fn merge_last_fail1() { | ||
1125 | check_merge_only_fail( | ||
1126 | r"use foo::bar::{baz::{Qux, Fez}};", | ||
1127 | r"use foo::bar::baaz::{Quux, Feez};", | ||
1128 | MergeBehaviour::Last, | ||
1129 | ); | ||
1130 | } | ||
1131 | |||
1132 | #[test] | ||
1133 | fn merge_last_fail2() { | ||
1134 | check_merge_only_fail( | ||
1135 | r"use foo::bar::baz::{Qux, Fez};", | ||
1136 | r"use foo::bar::{baaz::{Quux, Feez}};", | ||
1137 | MergeBehaviour::Last, | ||
1138 | ); | ||
1139 | } | ||
1140 | |||
1141 | #[test] | ||
1142 | fn merge_last_fail3() { | ||
1143 | check_merge_only_fail( | ||
1144 | r"use foo::bar::baz::{Qux, Fez};", | ||
1145 | r"use foo::bar::baaz::{Quux, Feez};", | ||
1146 | MergeBehaviour::Last, | ||
1147 | ); | ||
1148 | } | ||
1149 | |||
1150 | fn check( | ||
1151 | path: &str, | ||
1152 | ra_fixture_before: &str, | ||
1153 | ra_fixture_after: &str, | ||
1154 | mb: Option<MergeBehaviour>, | ||
1155 | module: bool, | ||
1156 | ) { | ||
1157 | let mut syntax = ast::SourceFile::parse(ra_fixture_before).tree().syntax().clone(); | ||
1158 | if module { | ||
1159 | syntax = syntax.descendants().find_map(ast::Module::cast).unwrap().syntax().clone(); | ||
1160 | } | ||
1161 | let file = super::ImportScope::from(syntax).unwrap(); | ||
1162 | let path = ast::SourceFile::parse(&format!("use {};", path)) | ||
1163 | .tree() | ||
1164 | .syntax() | ||
1165 | .descendants() | ||
1166 | .find_map(ast::Path::cast) | ||
1167 | .unwrap(); | ||
1168 | |||
1169 | let rewriter = insert_use(&file, path, mb); | ||
1170 | let result = rewriter.rewrite(file.as_syntax_node()).to_string(); | ||
1171 | assert_eq_text!(&result, ra_fixture_after); | ||
1172 | } | ||
1173 | |||
1174 | fn check_full(path: &str, ra_fixture_before: &str, ra_fixture_after: &str) { | ||
1175 | check(path, ra_fixture_before, ra_fixture_after, Some(MergeBehaviour::Full), false) | ||
1176 | } | ||
1177 | |||
1178 | fn check_last(path: &str, ra_fixture_before: &str, ra_fixture_after: &str) { | ||
1179 | check(path, ra_fixture_before, ra_fixture_after, Some(MergeBehaviour::Last), false) | ||
1180 | } | ||
1181 | |||
1182 | fn check_none(path: &str, ra_fixture_before: &str, ra_fixture_after: &str) { | ||
1183 | check(path, ra_fixture_before, ra_fixture_after, None, false) | ||
1184 | } | ||
1185 | |||
1186 | fn check_merge_only_fail(ra_fixture0: &str, ra_fixture1: &str, mb: MergeBehaviour) { | ||
1187 | let use0 = ast::SourceFile::parse(ra_fixture0) | ||
1188 | .tree() | ||
1189 | .syntax() | ||
1190 | .descendants() | ||
1191 | .find_map(ast::Use::cast) | ||
1192 | .unwrap(); | ||
1193 | |||
1194 | let use1 = ast::SourceFile::parse(ra_fixture1) | ||
1195 | .tree() | ||
1196 | .syntax() | ||
1197 | .descendants() | ||
1198 | .find_map(ast::Use::cast) | ||
1199 | .unwrap(); | ||
1200 | |||
1201 | let result = try_merge_imports(&use0, &use1, mb); | ||
1202 | assert_eq!(result.map(|u| u.to_string()), None); | ||
1203 | } | ||
1204 | } | ||