Warn on enum assignment to bitfields that can't fit all values

[rnk]

Summary

This adds -Wbitfield-enum-conversion, which warns on implicit
conversions that happen on bitfield assignment that change the value of
some enumerators.

Values of enum type typically take on a very small range of values, so
they are frequently stored in bitfields. Unfortunately, there is no
convenient way to calculate the minimum number of bits necessary to
store all possible values at compile time, so users usually hard code a
bitwidth that works today and widen it as necessary to pass basic
testing and validation. This is very error-prone, and leads to stale
widths as enums grow. This warning aims to catch such bugs.

This would have found two real bugs in clang and two instances of
questionable code. See r297680 and r297654 for the full description of
the issues.

This warning is currently disabled by default while we investigate its
usefulness outside of LLVM.

The major cause of false positives with this warning is this kind of
enum:

enum E { W, X, Y, Z, SENTINEL_LAST };

The last enumerator is an invalid value used to validate inputs or size
an array. Depending on the prevalance of this style of enum across a
codebase, this warning may be more or less feasible to deploy. It also
has trouble on sentinel values such as ~0U.

[thakis]

test/Sema/warn-bitfield-enum-conversion.c

3

can you add an enum with an explicit underlying type? will this warning always fire for those?

[rnk]

• Make test C++, add fixed type enum

[rnk]

test/Sema/warn-bitfield-enum-conversion.c

3

I was worried for a moment, but it looks like EnumDecl::getNumPositiveBits does what I want and ignores the fixed underlying type.

[rnk]

• Actually make this warning off by default

[thakis]

Looks like a cool warning. Two suggestions for more exhaustive testing, but I think this looks good.

test/SemaCXX/warn-bitfield-enum-conversion.cpp

1 ↗

(On Diff #91653)

Consider also running this test with a triple where enums don't default to signed.

15 ↗

(On Diff #91653)

Also add a : 4 version and check that that doesn't warn?

[thakis]

Maybe it should have some "to suppress the warning, do X" fixit?

[hfinkel]

In D30923#700620, @thakis wrote:

Maybe it should have some "to suppress the warning, do X" fixit?

I think we should at least include how many bits the field should have to fix the problem (pointing to the relevant field definition certainly seems helpful).

[rnk]

In D30923#700626, @hfinkel wrote:

In D30923#700620, @thakis wrote:

Maybe it should have some "to suppress the warning, do X" fixit?

I think we should at least include how many bits the field should have to fix the problem (pointing to the relevant field definition certainly seems helpful).

Agreed, I was just hacking that together. :)

Do you think it's worth indicating that the error can be suppressed with an explicit cast, or is that wasted space?

[rnk]

• Beef up test as Nico suggests
• Add two notes, one to change the bitfield signedness, one to indicate the required bitwidth

[hfinkel]

In D30923#700696, @rnk wrote:

In D30923#700626, @hfinkel wrote:

In D30923#700620, @thakis wrote:

Maybe it should have some "to suppress the warning, do X" fixit?

I think we should at least include how many bits the field should have to fix the problem (pointing to the relevant field definition certainly seems helpful).

Agreed, I was just hacking that together. :)

Do you think it's worth indicating that the error can be suppressed with an explicit cast, or is that wasted space?

What might this look like? Also, I don't see a regression test for this.

[rnk]

• Test that explicit casts suppress the warning

[rnk]

In D30923#700708, @hfinkel wrote:

In D30923#700696, @rnk wrote:

Do you think it's worth indicating that the error can be suppressed with an explicit cast, or is that wasted space?

What might this look like? Also, I don't see a regression test for this.

The warning only looks through implicit casts and paren exprs, so it could look like this: f.two_bits = (unsigned)three_bits; I added a test for it.

[hfinkel]

In D30923#700780, @rnk wrote:

In D30923#700708, @hfinkel wrote:

In D30923#700696, @rnk wrote:

Do you think it's worth indicating that the error can be suppressed with an explicit cast, or is that wasted space?

What might this look like? Also, I don't see a regression test for this.

The warning only looks through implicit casts and paren exprs, so it could look like this: f.two_bits = (unsigned)three_bits; I added a test for it.

Great, thanks!

lib/Sema/SemaChecking.cpp

8765

Line is too long.

[rnk]

Cool, thanks for the reviews, this is definitely in better shape now. This is off by default, so I'm going to commit and experiment with it on a few codebases.

I'd still like to hear if either of the Richards have diagnostic wording suggestions, but we can do that in a follow-up.

include/clang/Basic/DiagnosticGroups.td

 def ArrayBoundsPointerArithmetic : DiagGroup<"array-bounds-pointer-arithmetic">;
 def Availability : DiagGroup<"availability">;
 def Section : DiagGroup<"section">;
 def AutoImport : DiagGroup<"auto-import">;
 def CXX14BinaryLiteral : DiagGroup<"c++14-binary-literal">;
 def GNUBinaryLiteral : DiagGroup<"gnu-binary-literal">;
 def GNUCompoundLiteralInitializer : DiagGroup<"gnu-compound-literal-initializer">;
 def BitFieldConstantConversion : DiagGroup<"bitfield-constant-conversion">;
+def BitFieldEnumConversion : DiagGroup<"bitfield-enum-conversion">;
 def BitFieldWidth : DiagGroup<"bitfield-width">;
 def ConstantConversion :
   DiagGroup<"constant-conversion", [ BitFieldConstantConversion ] >;
 def LiteralConversion : DiagGroup<"literal-conversion">;
 def StringConversion : DiagGroup<"string-conversion">;
 def SignConversion : DiagGroup<"sign-conversion">;
 def PointerBoolConversion : DiagGroup<"pointer-bool-conversion">;
 def UndefinedBoolConversion : DiagGroup<"undefined-bool-conversion">;

 //   - conversion warnings with constant sources are on by default
 //   - conversion warnings for literals are on by default
 //   - bool-to-pointer conversion warnings are on by default
 //   - __null-to-integer conversion warnings are on by default
 def Conversion : DiagGroup<"conversion",
                            [BoolConversion,
                             ConstantConversion,
                             EnumConversion,
+                            BitFieldEnumConversion,
                             FloatConversion,
                             Shorten64To32,
                             IntConversion,
                             LiteralConversion,
                             NonLiteralNullConversion, // (1-1)->pointer (etc)
                             NullConversion, // NULL->non-pointer
                             ObjCLiteralConversion,
                             SignConversion,

include/clang/Basic/DiagnosticSemaKinds.td

 
 // Used by C++ which allows bit-fields that are wider than the type.
 def warn_bitfield_width_exceeds_type_width: Warning<
   "width of bit-field %0 (%1 bits) exceeds the width of its type; value will "
   "be truncated to %2 bit%s2">, InGroup<BitFieldWidth>;
 def warn_anon_bitfield_width_exceeds_type_width : Warning<
   "width of anonymous bit-field (%0 bits) exceeds width of its type; value "
   "will be truncated to %1 bit%s1">, InGroup<BitFieldWidth>;
+def warn_bitfield_too_small_for_enum: Warning<
+  "bit-field %0 is not wide enough to store all enumerators of enum type %1">,
+  InGroup<BitFieldEnumConversion>;
+def warn_unsigned_bitfield_assigned_signed_enum: Warning<
+  "assigning value of signed enum type %1 to unsigned bit-field %0; "
+  "negative enumerators of enum %1 will be converted to positive values">,
+  InGroup<BitFieldEnumConversion>;
+def warn_signed_bitfield_enum_conversion: Warning<
+  "signed bit-field %0 needs an extra bit to represent the largest positive "
+  "values of enum type %1">,
+  InGroup<BitFieldEnumConversion>;
 
 def warn_missing_braces : Warning<
   "suggest braces around initialization of subobject">,
   InGroup<MissingBraces>, DefaultIgnore;
 
 def err_redefinition_of_label : Error<"redefinition of label %0">;
 def err_undeclared_label_use : Error<"use of undeclared label %0">;
 def err_goto_ms_asm_label : Error<

lib/Sema/SemaChecking.cpp

bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init,
   // Ignore value- or type-dependent expressions.
   if (Bitfield->getBitWidth()->isValueDependent() ||
       Bitfield->getBitWidth()->isTypeDependent() ||
       Init->isValueDependent() ||
       Init->isTypeDependent())
     return false;
 
   Expr *OriginalInit = Init->IgnoreParenImpCasts();
+  unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context);
 
   llvm::APSInt Value;
-  if (!OriginalInit->EvaluateAsInt(Value, S.Context, Expr::SE_AllowSideEffects))
+  if (!OriginalInit->EvaluateAsInt(Value, S.Context,
+                                   Expr::SE_AllowSideEffects)) {
+    // The RHS is not constant.  If the RHS has an enum type, make sure the
+    // bitfield is wide enough to hold all the values of the enum without
+    // truncation.
+    if (const auto *EnumTy = OriginalInit->getType()->getAs<EnumType>()) {
+      EnumDecl *ED = EnumTy->getDecl();
+      bool SignedBitfield = BitfieldType->isSignedIntegerType();
+
+      // Check if this might store negative values into an unsigned bitfield.
+      if (!SignedBitfield && ED->getNumNegativeBits() > 0) {
+        S.Diag(InitLoc, diag::warn_unsigned_bitfield_assigned_signed_enum)
+            << Bitfield << ED;
+      }
+
+      // Check for sufficient width.
+      if (ED->getNumPositiveBits() > FieldWidth ||
+          ED->getNumNegativeBits() > FieldWidth) {
+        S.Diag(InitLoc, diag::warn_bitfield_too_small_for_enum)
+            << Bitfield << ED;
+      } else if (SignedBitfield && ED->getNumPositiveBits() == FieldWidth) {
+        // If the bitfield type is signed and the positive field widths are
+        // equal, values with the high bit set will become negative, which is
+        // usually unintended.
+        S.Diag(InitLoc, diag::warn_signed_bitfield_enum_conversion)
+            << Bitfield << ED;
+      }
+    }
+
     return false;
+  }
 
   unsigned OriginalWidth = Value.getBitWidth();
-  unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context);
 
   if (!Value.isSigned() || Value.isNegative())
     if (UnaryOperator *UO = dyn_cast<UnaryOperator>(OriginalInit))
       if (UO->getOpcode() == UO_Minus || UO->getOpcode() == UO_Not)
         OriginalWidth = Value.getMinSignedBits();
 
   if (OriginalWidth <= FieldWidth)
     return false;

test/Sema/warn-bitfield-enum-conversion.c

+// RUN: %clang_cc1 -triple x86_64-windows-msvc -verify %s -Wbitfield-enum-conversion
+
+enum TwoBits { Hi1 = 3 } two_bits;
+enum TwoBitsSigned { Lo2 = -2, Hi2 = 1 } two_bits_signed;
+enum ThreeBits { Hi3 = 7 } three_bits;
+enum ThreeBitsSigned { Lo4 = -4, Hi4 = 3 } three_bits_signed;
+
+struct Foo {
+  unsigned two_bits : 2;
+  int two_bits_signed : 2;
+  unsigned three_bits : 3;
+  int three_bits_signed : 3;
+
+  enum ThreeBits three_bits_enum : 3;
+};
+
+void f() {
+  struct Foo f;
+
+  f.two_bits = two_bits;
+  f.two_bits = two_bits_signed;            // expected-warning {{negative enumerators}}
+  f.two_bits = three_bits;                 // expected-warning {{not wide enough}}
+  f.two_bits = three_bits_signed;          // expected-warning {{negative enumerators}} expected-warning {{not wide enough}}
+
+  f.two_bits_signed = two_bits;            // expected-warning {{needs an extra bit}}
+  f.two_bits_signed = two_bits_signed;
+  f.two_bits_signed = three_bits;          // expected-warning {{not wide enough}}
+  f.two_bits_signed = three_bits_signed;   // expected-warning {{not wide enough}}
+
+  f.three_bits = two_bits;
+  f.three_bits = two_bits_signed;          // expected-warning {{negative enumerators}}
+  f.three_bits = three_bits;
+  f.three_bits = three_bits_signed;        // expected-warning {{negative enumerators}}
+
+  f.three_bits_signed = two_bits;
+  f.three_bits_signed = two_bits_signed;
+  f.three_bits_signed = three_bits;        // expected-warning {{needs an extra bit}}
+  f.three_bits_signed = three_bits_signed;
+
+  f.three_bits_enum = three_bits;          // expected-warning {{needs an extra bit}}
+}