[clang] Fix unexpected warnings after a01307a

[Fznamznon]

a01307a broke silencing of -Wmissing-field-initializers warnings in C for nested designators. This fixes the issue.

[llvmbot]

@llvm/pr-subscribers-clang

Author: Mariya Podchishchaeva (Fznamznon)

Changes

a01307a broke silencing of -Wmissing-field-initializers warnings in C for nested designators. This fixes the issue.

---

Full diff: https://github.com/llvm/llvm-project/pull/75591.diff

2 Files Affected:

• (modified) clang/lib/Sema/SemaInit.cpp (+8)
• (modified) clang/test/Sema/missing-field-initializers.c (+23)

diff --git a/clang/lib/Sema/SemaInit.cpp b/clang/lib/Sema/SemaInit.cpp
index de0d92edb550dd..1cd2198503abaf 100644
--- a/clang/lib/Sema/SemaInit.cpp
+++ b/clang/lib/Sema/SemaInit.cpp
@@ -864,6 +864,14 @@ InitListChecker::FillInEmptyInitializations(const InitializedEntity &Entity,
       WarnIfMissingField &=
           SemaRef.getLangOpts().CPlusPlus || !hasAnyDesignatedInits(SForm);
 
+      if (OuterILE) {
+        InitListExpr *OuterSForm = OuterILE->isSyntacticForm()
+                                       ? OuterILE
+                                       : OuterILE->getSyntacticForm();
+        WarnIfMissingField &= SemaRef.getLangOpts().CPlusPlus ||
+                              !hasAnyDesignatedInits(OuterSForm);
+      }
+
       unsigned NumElems = numStructUnionElements(ILE->getType());
       if (!RDecl->isUnion() && RDecl->hasFlexibleArrayMember())
         ++NumElems;
diff --git a/clang/test/Sema/missing-field-initializers.c b/clang/test/Sema/missing-field-initializers.c
index 8653591ff1187a..8dc8288ad92e6c 100644
--- a/clang/test/Sema/missing-field-initializers.c
+++ b/clang/test/Sema/missing-field-initializers.c
@@ -61,3 +61,26 @@ struct S {
 // f1, now we no longer issue that warning (note, this code is still unsafe
 // because of the buffer overrun).
 struct S s = {1, {1, 2}};
+
+struct S1 {
+  long int l;
+  struct  { int a, b; } d1;
+};
+
+struct S1 s01 = { 1, {1} }; // expected-warning {{missing field 'b' initializer}}
+struct S1 s02 = { .d1.a = 1 }; // designator avoids MFI warning
+
+union U1 {
+  long int l;
+  struct  { int a, b; } d1;
+};
+
+union U1 u01 = { 1 };
+union U1 u02 = { .d1.a = 1 }; // designator avoids MFI warning
+
+struct S2 {
+  long int l;
+  struct { int a, b; struct {int c; } d2; } d1;
+};
+
+struct S2 s22 = { .d1.d2.c = 1 }; // designator avoids MFI warning

[mikaelholmen]

I really don't know this, and I have no easy way to test that this really fixes all the new warnings we see with a01307a, but at least it seems to fix the added cases in the missing-field-initializers.c test so it looks like a step in the right direction.

[mikaelholmen]

Ok, ideally this should be reviewed by someone who knows the code but it LGTM.

[Fznamznon]

Ok, ideally this should be reviewed by someone who knows the code but it LGTM.

I know, but a lot of folks are now on vacation already. The change is not complicated, so I'll probably wait till green pre-commit, then leave it for post-commit review to resolve the issue.

[mikaelholmen]

Ok, ideally this should be reviewed by someone who knows the code but it LGTM.

I know, but a lot of folks are now on vacation already. The change is not complicated, so I'll probably wait till green pre-commit, then leave it for post-commit review to resolve the issue.

Yeah, sounds good. Thanks!

[Fznamznon]

Ok, adding more reviewers. Please feel free to do post-commit review.