1 //===--- RewriteRule.h - RewriteRule class ----------------------*- C++ -*-===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 ///
9 /// \file
10 /// Defines the RewriteRule class and related functions for creating,
11 /// modifying and interpreting RewriteRules.
12 ///
13 //===----------------------------------------------------------------------===//
14
15 #ifndef LLVM_CLANG_TOOLING_TRANSFORMER_REWRITE_RULE_H_
16 #define LLVM_CLANG_TOOLING_TRANSFORMER_REWRITE_RULE_H_
17
18 #include "clang/ASTMatchers/ASTMatchFinder.h"
19 #include "clang/ASTMatchers/ASTMatchers.h"
20 #include "clang/ASTMatchers/ASTMatchersInternal.h"
21 #include "clang/Tooling/Refactoring/AtomicChange.h"
22 #include "clang/Tooling/Transformer/MatchConsumer.h"
23 #include "clang/Tooling/Transformer/RangeSelector.h"
24 #include "llvm/ADT/Any.h"
25 #include "llvm/ADT/STLExtras.h"
26 #include "llvm/ADT/SmallVector.h"
27 #include "llvm/Support/Error.h"
28 #include <functional>
29 #include <string>
30 #include <utility>
31
32 namespace clang {
33 namespace transformer {
34 // Specifies how to interpret an edit.
35 enum class EditKind {
36 // Edits a source range in the file.
37 Range,
38 // Inserts an include in the file. The `Replacement` field is the name of the
39 // newly included file.
40 AddInclude,
41 };
42
43 /// A concrete description of a source edit, represented by a character range in
44 /// the source to be replaced and a corresponding replacement string.
45 struct Edit {
46 EditKind Kind = EditKind::Range;
47 CharSourceRange Range;
48 std::string Replacement;
49 llvm::Any Metadata;
50 };
51
52 /// Format of the path in an include directive -- angle brackets or quotes.
53 enum class IncludeFormat {
54 Quoted,
55 Angled,
56 };
57
58 /// Maps a match result to a list of concrete edits (with possible
59 /// failure). This type is a building block of rewrite rules, but users will
60 /// generally work in terms of `ASTEdit`s (below) rather than directly in terms
61 /// of `EditGenerator`.
62 using EditGenerator = MatchConsumer<llvm::SmallVector<Edit, 1>>;
63
64 using TextGenerator = std::shared_ptr<MatchComputation<std::string>>;
65
66 using AnyGenerator = MatchConsumer<llvm::Any>;
67
68 // Description of a source-code edit, expressed in terms of an AST node.
69 // Includes: an ID for the (bound) node, a selector for source related to the
70 // node, a replacement and, optionally, an explanation for the edit.
71 //
72 // * Target: the source code impacted by the rule. This identifies an AST node,
73 // or part thereof (\c Part), whose source range indicates the extent of the
74 // replacement applied by the replacement term. By default, the extent is the
75 // node matched by the pattern term (\c NodePart::Node). Target's are typed
76 // (\c Kind), which guides the determination of the node extent.
77 //
78 // * Replacement: a function that produces a replacement string for the target,
79 // based on the match result.
80 //
81 // * Note: (optional) a note specifically for this edit, potentially referencing
82 // elements of the match. This will be displayed to the user, where possible;
83 // for example, in clang-tidy diagnostics. Use of notes should be rare --
84 // explanations of the entire rewrite should be set in the rule
85 // (`RewriteRule::Explanation`) instead. Notes serve the rare cases wherein
86 // edit-specific diagnostics are required.
87 //
88 // `ASTEdit` should be built using the `change` convenience functions. For
89 // example,
90 // \code
91 // changeTo(name(fun), cat("Frodo"))
92 // \endcode
93 // Or, if we use Stencil for the TextGenerator:
94 // \code
95 // using stencil::cat;
96 // changeTo(statement(thenNode), cat("{", thenNode, "}"))
97 // changeTo(callArgs(call), cat(x, ",", y))
98 // \endcode
99 // Or, if you are changing the node corresponding to the rule's matcher, you can
100 // use the single-argument override of \c change:
101 // \code
102 // changeTo(cat("different_expr"))
103 // \endcode
104 struct ASTEdit {
105 EditKind Kind = EditKind::Range;
106 RangeSelector TargetRange;
107 TextGenerator Replacement;
108 TextGenerator Note;
109 // Not all transformations will want or need to attach metadata and therefore
110 // should not be required to do so.
111 AnyGenerator Metadata = [](const ast_matchers::MatchFinder::MatchResult &)
112 -> llvm::Expected<llvm::Any> {
113 return llvm::Expected<llvm::Any>(llvm::Any());
114 };
115 };
116
117 /// Generates a single (specified) edit.
118 EditGenerator edit(ASTEdit E);
119
120 /// Lifts a list of `ASTEdit`s into an `EditGenerator`.
121 ///
122 /// The `EditGenerator` will return an empty vector if any of the edits apply to
123 /// portions of the source that are ineligible for rewriting (certain
124 /// interactions with macros, for example) and it will fail if any invariants
125 /// are violated relating to bound nodes in the match. However, it does not
126 /// fail in the case of conflicting edits -- conflict handling is left to
127 /// clients. We recommend use of the \c AtomicChange or \c Replacements classes
128 /// for assistance in detecting such conflicts.
129 EditGenerator editList(llvm::SmallVector<ASTEdit, 1> Edits);
130
131 /// Generates no edits.
noEdits()132 inline EditGenerator noEdits() { return editList({}); }
133
134 /// Generates a single, no-op edit anchored at the start location of the
135 /// specified range. A `noopEdit` may be preferred over `noEdits` to associate a
136 /// diagnostic `Explanation` with the rule.
137 EditGenerator noopEdit(RangeSelector Anchor);
138
139 /// Version of `ifBound` specialized to `ASTEdit`.
ifBound(std::string ID,ASTEdit TrueEdit,ASTEdit FalseEdit)140 inline EditGenerator ifBound(std::string ID, ASTEdit TrueEdit,
141 ASTEdit FalseEdit) {
142 return ifBound(std::move(ID), edit(std::move(TrueEdit)),
143 edit(std::move(FalseEdit)));
144 }
145
146 /// Version of `ifBound` that has no "False" branch. If the node is not bound,
147 /// then no edits are produced.
ifBound(std::string ID,ASTEdit TrueEdit)148 inline EditGenerator ifBound(std::string ID, ASTEdit TrueEdit) {
149 return ifBound(std::move(ID), edit(std::move(TrueEdit)), noEdits());
150 }
151
152 /// Flattens a list of generators into a single generator whose elements are the
153 /// concatenation of the results of the argument generators.
154 EditGenerator flattenVector(SmallVector<EditGenerator, 2> Generators);
155
156 namespace detail {
157 /// Helper function to construct an \c EditGenerator. Overloaded for common
158 /// cases so that user doesn't need to specify which factory function to
159 /// use. This pattern gives benefits similar to implicit constructors, while
160 /// maintaing a higher degree of explicitness.
injectEdits(ASTEdit E)161 inline EditGenerator injectEdits(ASTEdit E) { return edit(std::move(E)); }
injectEdits(EditGenerator G)162 inline EditGenerator injectEdits(EditGenerator G) { return G; }
163 } // namespace detail
164
flatten(Ts &&...Edits)165 template <typename... Ts> EditGenerator flatten(Ts &&...Edits) {
166 return flattenVector({detail::injectEdits(std::forward<Ts>(Edits))...});
167 }
168
169 // Every rewrite rule is triggered by a match against some AST node.
170 // Transformer guarantees that this ID is bound to the triggering node whenever
171 // a rewrite rule is applied.
172 extern const char RootID[];
173
174 /// Replaces a portion of the source text with \p Replacement.
175 ASTEdit changeTo(RangeSelector Target, TextGenerator Replacement);
176 /// DEPRECATED: use \c changeTo.
change(RangeSelector Target,TextGenerator Replacement)177 inline ASTEdit change(RangeSelector Target, TextGenerator Replacement) {
178 return changeTo(std::move(Target), std::move(Replacement));
179 }
180
181 /// Replaces the entirety of a RewriteRule's match with \p Replacement. For
182 /// example, to replace a function call, one could write:
183 /// \code
184 /// makeRule(callExpr(callee(functionDecl(hasName("foo")))),
185 /// changeTo(cat("bar()")))
186 /// \endcode
changeTo(TextGenerator Replacement)187 inline ASTEdit changeTo(TextGenerator Replacement) {
188 return changeTo(node(RootID), std::move(Replacement));
189 }
190 /// DEPRECATED: use \c changeTo.
change(TextGenerator Replacement)191 inline ASTEdit change(TextGenerator Replacement) {
192 return changeTo(std::move(Replacement));
193 }
194
195 /// Inserts \p Replacement before \p S, leaving the source selected by \S
196 /// unchanged.
insertBefore(RangeSelector S,TextGenerator Replacement)197 inline ASTEdit insertBefore(RangeSelector S, TextGenerator Replacement) {
198 return changeTo(before(std::move(S)), std::move(Replacement));
199 }
200
201 /// Inserts \p Replacement after \p S, leaving the source selected by \S
202 /// unchanged.
insertAfter(RangeSelector S,TextGenerator Replacement)203 inline ASTEdit insertAfter(RangeSelector S, TextGenerator Replacement) {
204 return changeTo(after(std::move(S)), std::move(Replacement));
205 }
206
207 /// Removes the source selected by \p S.
208 ASTEdit remove(RangeSelector S);
209
210 /// Adds an include directive for the given header to the file of `Target`. The
211 /// particular location specified by `Target` is ignored.
212 ASTEdit addInclude(RangeSelector Target, StringRef Header,
213 IncludeFormat Format = IncludeFormat::Quoted);
214
215 /// Adds an include directive for the given header to the file associated with
216 /// `RootID`.
217 inline ASTEdit addInclude(StringRef Header,
218 IncludeFormat Format = IncludeFormat::Quoted) {
219 return addInclude(node(RootID), Header, Format);
220 }
221
222 // FIXME: If `Metadata` returns an `llvm::Expected<T>` the `AnyGenerator` will
223 // construct an `llvm::Expected<llvm::Any>` where no error is present but the
224 // `llvm::Any` holds the error. This is unlikely but potentially surprising.
225 // Perhaps the `llvm::Expected` should be unwrapped, or perhaps this should be a
226 // compile-time error. No solution here is perfect.
227 //
228 // Note: This function template accepts any type callable with a MatchResult
229 // rather than a `std::function` because the return-type needs to be deduced. If
230 // it accepted a `std::function<R(MatchResult)>`, lambdas or other callable
231 // types would not be able to deduce `R`, and users would be forced to specify
232 // explicitly the type they intended to return by wrapping the lambda at the
233 // call-site.
234 template <typename Callable>
withMetadata(ASTEdit Edit,Callable Metadata)235 inline ASTEdit withMetadata(ASTEdit Edit, Callable Metadata) {
236 Edit.Metadata =
237 [Gen = std::move(Metadata)](
238 const ast_matchers::MatchFinder::MatchResult &R) -> llvm::Any {
239 return Gen(R);
240 };
241
242 return Edit;
243 }
244
245 /// Assuming that the inner range is enclosed by the outer range, creates
246 /// precision edits to remove the parts of the outer range that are not included
247 /// in the inner range.
shrinkTo(RangeSelector outer,RangeSelector inner)248 inline EditGenerator shrinkTo(RangeSelector outer, RangeSelector inner) {
249 return editList({remove(enclose(before(outer), before(inner))),
250 remove(enclose(after(inner), after(outer)))});
251 }
252
253 /// Description of a source-code transformation.
254 //
255 // A *rewrite rule* describes a transformation of source code. A simple rule
256 // contains each of the following components:
257 //
258 // * Matcher: the pattern term, expressed as clang matchers (with Transformer
259 // extensions).
260 //
261 // * Edits: a set of Edits to the source code, described with ASTEdits.
262 //
263 // * Explanation: explanation of the rewrite. This will be displayed to the
264 // user, where possible; for example, in clang-tidy diagnostics.
265 //
266 // However, rules can also consist of (sub)rules, where the first that matches
267 // is applied and the rest are ignored. So, the above components are gathered
268 // as a `Case` and a rule is a list of cases.
269 //
270 // Rule cases have an additional, implicit, component: the parameters. These are
271 // portions of the pattern which are left unspecified, yet bound in the pattern
272 // so that we can reference them in the edits.
273 //
274 // The \c Transformer class can be used to apply the rewrite rule and obtain the
275 // corresponding replacements.
276 struct RewriteRule {
277 struct Case {
278 ast_matchers::internal::DynTypedMatcher Matcher;
279 EditGenerator Edits;
280 TextGenerator Explanation;
281 };
282 // We expect RewriteRules will most commonly include only one case.
283 SmallVector<Case, 1> Cases;
284
285 /// DEPRECATED: use `::clang::transformer::RootID` instead.
286 static const llvm::StringRef RootID;
287 };
288
289 /// Constructs a simple \c RewriteRule.
290 RewriteRule makeRule(ast_matchers::internal::DynTypedMatcher M,
291 EditGenerator Edits, TextGenerator Explanation = nullptr);
292
293 /// Constructs a \c RewriteRule from multiple `ASTEdit`s.
294 inline RewriteRule makeRule(ast_matchers::internal::DynTypedMatcher M,
295 llvm::SmallVector<ASTEdit, 1> Edits,
296 TextGenerator Explanation = nullptr) {
297 return makeRule(std::move(M), editList(std::move(Edits)),
298 std::move(Explanation));
299 }
300
301 /// Overload of \c makeRule for common case of only one edit.
302 inline RewriteRule makeRule(ast_matchers::internal::DynTypedMatcher M,
303 ASTEdit Edit,
304 TextGenerator Explanation = nullptr) {
305 return makeRule(std::move(M), edit(std::move(Edit)), std::move(Explanation));
306 }
307
308 /// For every case in Rule, adds an include directive for the given header. The
309 /// common use is assumed to be a rule with only one case. For example, to
310 /// replace a function call and add headers corresponding to the new code, one
311 /// could write:
312 /// \code
313 /// auto R = makeRule(callExpr(callee(functionDecl(hasName("foo")))),
314 /// changeTo(cat("bar()")));
315 /// AddInclude(R, "path/to/bar_header.h");
316 /// AddInclude(R, "vector", IncludeFormat::Angled);
317 /// \endcode
318 void addInclude(RewriteRule &Rule, llvm::StringRef Header,
319 IncludeFormat Format = IncludeFormat::Quoted);
320
321 /// Applies the first rule whose pattern matches; other rules are ignored. If
322 /// the matchers are independent then order doesn't matter. In that case,
323 /// `applyFirst` is simply joining the set of rules into one.
324 //
325 // `applyFirst` is like an `anyOf` matcher with an edit action attached to each
326 // of its cases. Anywhere you'd use `anyOf(m1.bind("id1"), m2.bind("id2"))` and
327 // then dispatch on those ids in your code for control flow, `applyFirst` lifts
328 // that behavior to the rule level. So, you can write `applyFirst({makeRule(m1,
329 // action1), makeRule(m2, action2), ...});`
330 //
331 // For example, consider a type `T` with a deterministic serialization function,
332 // `serialize()`. For performance reasons, we would like to make it
333 // non-deterministic. Therefore, we want to drop the expectation that
334 // `a.serialize() = b.serialize() iff a = b` (although we'll maintain
335 // `deserialize(a.serialize()) = a`).
336 //
337 // We have three cases to consider (for some equality function, `eq`):
338 // ```
339 // eq(a.serialize(), b.serialize()) --> eq(a,b)
340 // eq(a, b.serialize()) --> eq(deserialize(a), b)
341 // eq(a.serialize(), b) --> eq(a, deserialize(b))
342 // ```
343 //
344 // `applyFirst` allows us to specify each independently:
345 // ```
346 // auto eq_fun = functionDecl(...);
347 // auto method_call = cxxMemberCallExpr(...);
348 //
349 // auto two_calls = callExpr(callee(eq_fun), hasArgument(0, method_call),
350 // hasArgument(1, method_call));
351 // auto left_call =
352 // callExpr(callee(eq_fun), callExpr(hasArgument(0, method_call)));
353 // auto right_call =
354 // callExpr(callee(eq_fun), callExpr(hasArgument(1, method_call)));
355 //
356 // RewriteRule R = applyFirst({makeRule(two_calls, two_calls_action),
357 // makeRule(left_call, left_call_action),
358 // makeRule(right_call, right_call_action)});
359 // ```
360 RewriteRule applyFirst(ArrayRef<RewriteRule> Rules);
361
362 /// Applies `Rule` to all descendants of the node bound to `NodeId`. `Rule` can
363 /// refer to nodes bound by the calling rule. `Rule` is not applied to the node
364 /// itself.
365 ///
366 /// For example,
367 /// ```
368 /// auto InlineX =
369 /// makeRule(declRefExpr(to(varDecl(hasName("x")))), changeTo(cat("3")));
370 /// makeRule(functionDecl(hasName("f"), hasBody(stmt().bind("body"))).bind("f"),
371 /// flatten(
372 /// changeTo(name("f"), cat("newName")),
373 /// rewriteDescendants("body", InlineX)));
374 /// ```
375 /// Here, we find the function `f`, change its name to `newName` and change all
376 /// appearances of `x` in its body to `3`.
377 EditGenerator rewriteDescendants(std::string NodeId, RewriteRule Rule);
378
379 /// The following three functions are a low-level part of the RewriteRule
380 /// API. We expose them for use in implementing the fixtures that interpret
381 /// RewriteRule, like Transformer and TransfomerTidy, or for more advanced
382 /// users.
383 //
384 // FIXME: These functions are really public, if advanced, elements of the
385 // RewriteRule API. Recast them as such. Or, just declare these functions
386 // public and well-supported and move them out of `detail`.
387 namespace detail {
388 /// The following overload set is a version of `rewriteDescendants` that
389 /// operates directly on the AST, rather than generating a Transformer
390 /// combinator. It applies `Rule` to all descendants of `Node`, although not
391 /// `Node` itself. `Rule` can refer to nodes bound in `Result`.
392 ///
393 /// For example, assuming that "body" is bound to a function body in MatchResult
394 /// `Results`, this will produce edits to change all appearances of `x` in that
395 /// body to `3`.
396 /// ```
397 /// auto InlineX =
398 /// makeRule(declRefExpr(to(varDecl(hasName("x")))), changeTo(cat("3")));
399 /// const auto *Node = Results.Nodes.getNodeAs<Stmt>("body");
400 /// auto Edits = rewriteDescendants(*Node, InlineX, Results);
401 /// ```
402 /// @{
403 llvm::Expected<SmallVector<Edit, 1>>
404 rewriteDescendants(const Decl &Node, RewriteRule Rule,
405 const ast_matchers::MatchFinder::MatchResult &Result);
406
407 llvm::Expected<SmallVector<Edit, 1>>
408 rewriteDescendants(const Stmt &Node, RewriteRule Rule,
409 const ast_matchers::MatchFinder::MatchResult &Result);
410
411 llvm::Expected<SmallVector<Edit, 1>>
412 rewriteDescendants(const TypeLoc &Node, RewriteRule Rule,
413 const ast_matchers::MatchFinder::MatchResult &Result);
414
415 llvm::Expected<SmallVector<Edit, 1>>
416 rewriteDescendants(const DynTypedNode &Node, RewriteRule Rule,
417 const ast_matchers::MatchFinder::MatchResult &Result);
418 /// @}
419
420 /// Builds a single matcher for the rule, covering all of the rule's cases.
421 /// Only supports Rules whose cases' matchers share the same base "kind"
422 /// (`Stmt`, `Decl`, etc.) Deprecated: use `buildMatchers` instead, which
423 /// supports mixing matchers of different kinds.
424 ast_matchers::internal::DynTypedMatcher buildMatcher(const RewriteRule &Rule);
425
426 /// Builds a set of matchers that cover the rule.
427 ///
428 /// One matcher is built for each distinct node matcher base kind: Stmt, Decl,
429 /// etc. Node-matchers for `QualType` and `Type` are not permitted, since such
430 /// nodes carry no source location information and are therefore not relevant
431 /// for rewriting. If any such matchers are included, will return an empty
432 /// vector.
433 std::vector<ast_matchers::internal::DynTypedMatcher>
434 buildMatchers(const RewriteRule &Rule);
435
436 /// Gets the beginning location of the source matched by a rewrite rule. If the
437 /// match occurs within a macro expansion, returns the beginning of the
438 /// expansion point. `Result` must come from the matching of a rewrite rule.
439 SourceLocation
440 getRuleMatchLoc(const ast_matchers::MatchFinder::MatchResult &Result);
441
442 /// Returns the \c Case of \c Rule that was selected in the match result.
443 /// Assumes a matcher built with \c buildMatcher.
444 const RewriteRule::Case &
445 findSelectedCase(const ast_matchers::MatchFinder::MatchResult &Result,
446 const RewriteRule &Rule);
447 } // namespace detail
448 } // namespace transformer
449 } // namespace clang
450
451 #endif // LLVM_CLANG_TOOLING_TRANSFORMER_REWRITE_RULE_H_
452