VitaNuo added inline comments.
================
Comment at: clang-tools-extra/include-cleaner/lib/Record.cpp:377
continue;
- // FIXME: Filter out certain Obj-C and template-related decls.
+ if (const NamedDecl *ND = dyn_cast<NamedDecl>(D))
+ if (isImplicitTemplateInstantiation(ND))
----------------
hokein wrote:
> as we already do a dyn_cast inside the `isImplicitTemplateInstantiation`, we
> can get rid of this NamedDecl cast by changing the
> `isImplicitTemplateInstantiation` to accept a `Decl*`.
Oh right, I didn't notice. Thanks.
================
Comment at: clang-tools-extra/include-cleaner/lib/Record.cpp:387
+ private:
+ bool isImplicitTemplateInstantiation(const NamedDecl *D) {
+ return isImplicitTemplateSpecialization<FunctionDecl>(D) ||
----------------
hokein wrote:
> we can move it to the above anonymous namespace as well (it is fine to have
> two functions with same name because of C++ overloads). In this case, I'd
> probably inline it in `HandleTopLevelDecl`.
I am not sure I fully understand this comment, since I cannot both move this
method to the anonymous namespace above and inline it at the same time. I have
inlined it now, hopefully it looks fine to you.
================
Comment at: clang-tools-extra/include-cleaner/unittests/RecordTest.cpp:105
+// Decl from template instantiation is filtered out from roots.
+TEST_F(RecordASTTest, TemplateInstantiations) {
+ Inputs.ExtraFiles["dispatch.h"] = R"cpp(
----------------
hokein wrote:
> maybe DropImplicitTemplateTopDecl?
Not sure, since the rest of tests above only use nouns, e.g., "Macros",
"Inclusion", "Namespace", etc. I've called it "ImplicitTemplates" now,
hopefully it's fine.
================
Comment at: clang-tools-extra/include-cleaner/unittests/RecordTest.cpp:120
+ };
+ int v = dispatch<MyGetter>();
+ )cpp";
----------------
hokein wrote:
> this example is a nice example to show the original bug (where we inserted
> unexpected additional headers) in include-cleaner.
>
> The scope of this patch is to filter out the implicit template instantiation,
> a simple testcase like below could work. (the current testcase is fine as
> well, up to you)
>
> ```
> template<typename T>
> int func(T) { return 0; }
> auto s = func<int>(1);
> ```
You're right, but IMHO I would rather keep the current example. The current
example, although simplified, still highly resembles the pattern we've found it
real code. The snippet above does not anymore. Since it's a bug fix for a real
pattern, I'd rather keep it as is. Let me know if you disagree.
================
Comment at: clang-tools-extra/include-cleaner/unittests/RecordTest.cpp:123
+ auto AST = build();
+ EXPECT_THAT(Recorded.Roots,
testing::Not(testing::ElementsAre(named("get"))));
+}
----------------
hokein wrote:
> nit: instead doing a negative verification, I'd check the captured root decl,
> which is (a `MyGetter` CXXRecordDecl, and a `v` var decl), and with a comment
> saying the implicitly-instantiated method decl is filtered out.
Sure, I have no preference.
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D141271/new/
https://reviews.llvm.org/D141271
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
