[clang][Sema] Add -Wswitch-default warning option

[dongjianqiang2]

Adds a warning, issued by the clang semantic analysis. The patch warns on switch which don't have the default branch.

This is a counterpart of gcc's Wswitch-default.

[llvmbot]

@llvm/pr-subscribers-clang

Author: dong jianqiang (dongjianqiang2)

Changes

Adds a warning, issued by the clang semantic analysis. The patch warns on switch which don't have the default branch.

This is a counterpart of gcc's Wswitch-default.

---

Full diff: https://github.com/llvm/llvm-project/pull/73077.diff

4 Files Affected:

• (modified) clang/include/clang/Basic/DiagnosticGroups.td (+1-1)
• (modified) clang/include/clang/Basic/DiagnosticSemaKinds.td (+2)
• (modified) clang/lib/Sema/SemaStmt.cpp (+3)
• (added) clang/test/Sema/switch-default.c (+17)

diff --git a/clang/include/clang/Basic/DiagnosticGroups.td b/clang/include/clang/Basic/DiagnosticGroups.td
index ff028bbbf74261e..12b11527b30571a 100644
--- a/clang/include/clang/Basic/DiagnosticGroups.td
+++ b/clang/include/clang/Basic/DiagnosticGroups.td
@@ -632,7 +632,7 @@ def ShadowAll : DiagGroup<"shadow-all", [Shadow, ShadowFieldInConstructor,
 def Shorten64To32 : DiagGroup<"shorten-64-to-32">;
 def : DiagGroup<"sign-promo">;
 def SignCompare : DiagGroup<"sign-compare">;
-def : DiagGroup<"switch-default">;
+def SwitchDefault  : DiagGroup<"switch-default">;
 def : DiagGroup<"synth">;
 def SizeofArrayArgument : DiagGroup<"sizeof-array-argument">;
 def SizeofArrayDecay : DiagGroup<"sizeof-array-decay">;
diff --git a/clang/include/clang/Basic/DiagnosticSemaKinds.td b/clang/include/clang/Basic/DiagnosticSemaKinds.td
index 990692c06d7d3a8..17c9627910bb6ce 100644
--- a/clang/include/clang/Basic/DiagnosticSemaKinds.td
+++ b/clang/include/clang/Basic/DiagnosticSemaKinds.td
@@ -10044,6 +10044,8 @@ def warn_missing_case : Warning<"%plural{"
   "3:enumeration values %1, %2, and %3 not handled in switch|"
   ":%0 enumeration values not handled in switch: %1, %2, %3...}0">,
   InGroup<Switch>;
+def warn_switch_default : Warning<"switch missing default case">,
+  InGroup<SwitchDefault>, DefaultIgnore;
 
 def warn_unannotated_fallthrough : Warning<
   "unannotated fall-through between switch labels">,
diff --git a/clang/lib/Sema/SemaStmt.cpp b/clang/lib/Sema/SemaStmt.cpp
index 2b45aa5dff7be7c..63348d27a8c94a1 100644
--- a/clang/lib/Sema/SemaStmt.cpp
+++ b/clang/lib/Sema/SemaStmt.cpp
@@ -1327,6 +1327,9 @@ Sema::ActOnFinishSwitchStmt(SourceLocation SwitchLoc, Stmt *Switch,
     }
   }
 
+  if (!TheDefaultStmt)
+    Diag(SwitchLoc, diag::warn_switch_default);
+
   if (!HasDependentValue) {
     // If we don't have a default statement, check whether the
     // condition is constant.
diff --git a/clang/test/Sema/switch-default.c b/clang/test/Sema/switch-default.c
new file mode 100644
index 000000000000000..3f2e21693303378
--- /dev/null
+++ b/clang/test/Sema/switch-default.c
@@ -0,0 +1,17 @@
+// RUN: %clang_cc1 -fsyntax-only -verify -Wswitch-default %s
+
+int f1(int a) {
+  switch (a) {                // expected-warning {{switch missing default case}}
+    case 1: a++; break;
+    case 2: a += 2; break;
+  }
+  return a;
+}
+
+int f2(int a) {
+  switch (a) {                // no-warning
+    default:
+      ;
+  }
+  return a;
+}

[xgupta]

There is one clang-tidy check (bugprone-switch-missing-default-case) also for this feature.

[dongjianqiang2]

There is one clang-tidy check (bugprone-switch-missing-default-case) also for this feature.

Thank you for your reply. It may be a more convenient and straightforward way if can be identified during compile time.
On the other hand, it it more compatibile with GCC's hehavior. : )

[cor3ntin]

I'd like other to weight in on that but even though GCC has that flag (which clang recognizes and ignores), I'm, a bit reluctant to add a warning for it.
We have a policy to avoid off-by-default warnings most of the time... but more importantly, whether switch statements
should have a default is not something the industry seem to agree on, so it would be a very opinionated warning.

For example, Folks at google advise against using default when switching over an enum https://abseil.io/tips/147.
Misra requires a default statement, but makes an exception for enums. (6.4.6 in misra 2008)

LLVM development heavily relies on -Wswitch-enum, for example.

Maybe the existing tidy check is the right tool for this.

[hstk30-hw]

For example, Folks at google advise against using default when switching over an enum https://abseil.io/tips/147.
Misra requires a default statement, but makes an exception for enums. (6.4.6 in misra 2008)

I see it's easy to implement in LLVM like clang tidy (not warning for enum and constant expr).

The clang tidy may seperate the switch warning ( I have to use tidy to see switch-default, but I can see switch-enum switch-bool in compiler).

IMAO, if we can implement it together in compiler, why not?

[cor3ntin]

@AaronBallman @erichkeane @shafik WDYT?

[aaronpuchert]

There is one clang-tidy check (bugprone-switch-missing-default-case) also for this feature.

This excludes enumeration types though, while GCC -Wswitch-default warns on them as well. (Even if all enumeration values are covered.)

whether switch statements should have a default is not something the industry seem to agree on, so it would be a very opinionated warning.

Yes. Though I have to say, even though I'm squarely in the -Wcovered-switch-default camp (i.e. no default label in switches that cover all enumeration values), I think it's still valuable to have this warning because some style guides/policies demand a default, and it is very little effort on our side to maintain this.

I understand the policy against off-by-default warnings to avoid crowding the core compiler, but this is essentially just two lines of code, and if GCC has it we're not going out on a limb here. And since we already seem to support the flag for GCC compatibility, we might as well make it work if it's that easy.

LLVM development heavily relies on -Wswitch-enum, for example.

I think we're using -Wswitch, as -Wswitch-enum requires to cover all enumeration values even if there is a default label. Our policy is: cover all enumeration values or have a default. This is "enforced" by -Wswitch in combination with -Wcovered-switch-default.

[aaronpuchert]

In the end it probably boils down to how you view an enumeration type. Is it a type with the declared enumerators as inhabitants, or is it an integer type of some length with associated symbolic constants of that type? In other words, is an enumeration a "strong" type or a "weak" type?

The standard itself seems more in the "weak" camp. Yes, it does not allow integer values larger than what the largest enumeration value requires (if the underlying integer type isn't declared), but otherwise integer values that don't correspond to enumeration values can be cast to the enumeration type just fine. In [dcl.enum]p8:

[For an enumeration whose underlying type is not fixed,] the values of the enumeration are the values representable by a hypothetical integer type with minimal width M such that all enumerators can be represented.

However, nothing prevents a developer from assuming a stricter model, where only the declared enumeration values are allowed and everything else is undefined behavior. (If not in terms of the language, then in terms of the program.) That is what we have done in the LLVM code base, and it's also the model that I personally prefer.

But it is probably legitimate to follow the weaker model implied by the standard and assume that enumerations can have non-enumerator values, which requires a default label, enforced by this warning.

[erichkeane]

I think there is value to adding this if only for the GCC compat story here. I realize we don't like non-on-by-default warnings, but it IS something that folks use reasonably often, and we've even acquiesced to making it an ignored flag IIRC. Aaron is the real decision maker here, but I think I'd have no problem holding my nose and allowing the non-on-by-default here, if only to match GCC.

BUT, this should make sure to match behavior of GCC WRT enums/etc. But if folks want a warning for a fully-covered enum, I think there is value in giving it to them.

Patch itself need some additional testing (particularly around templates/dependence/etc), a release note, additional documentation in our commandline reference/etc.

[aaronpuchert]

Aaron is the real decision maker here

Specifically @AaronBallman, not me.

[AaronBallman]

Suggested change

	def warn_switch_default : Warning<"switch missing default case">,
	def warn_switch_default : Warning<"'switch' missing 'default' label">,

[dongjianqiang2]

Thanks for the correction!

[AaronBallman]

I think there is value to adding this if only for the GCC compat story here. I realize we don't like non-on-by-default warnings, but it IS something that folks use reasonably often, and we've even acquiesced to making it an ignored flag IIRC. Aaron is the real decision maker here, but I think I'd have no problem holding my nose and allowing the non-on-by-default here, if only to match GCC.

I think GCC compatibility is sufficient reason to support this despite being off-by-default and basically no hope of ever enabling it by default. There's evidence that people enable this in the wild, so I think it meets the bar: https://sourcegraph.com/search?q=context%3Aglobal+-Wswitch-default+-file%3A.*clang.*+-file%3A.*gcc.*&patternType=standard&sm=1&groupBy=repo

[AaronBallman]

LGTM!

[gendalph]

enum class test
{
  err1,
  err2,
  ok
};

int check_err (test v)
{
  switch (v)
    {
      case test::err1:
        return 1;
      case test::err2:
        return 2;
      case test::ok:
        break;
    }
  return 0;
}

report:
main.cxx:40:3: error: 'switch' missing 'default' label [-Werror,-Wswitch-default]

how to solve for enum class?
try to build libfmt project with
fmt/include/fmt/format.h:3878:3: error: 'switch' missing 'default' label [-Werror,-Wswitch-default]
and other errors

[dongjianqiang2]

enum class test
{
  err1,
  err2,
  ok
};

int check_err (test v)
{
  switch (v)
    {
      case test::err1:
        return 1;
      case test::err2:
        return 2;
      case test::ok:
        break;
    }
  return 0;
}

report: main.cxx:40:3: error: 'switch' missing 'default' label [-Werror,-Wswitch-default]

how to solve for enum class? try to build libfmt project with fmt/include/fmt/format.h:3878:3: error: 'switch' missing 'default' label [-Werror,-Wswitch-default] and other errors

Hi, -Wswitch-default is a warning option on switch which don't have the default branch, the option is disabled by default, You can check whether -Wswitch-default is added in the build script.

[shafik]

Might be worth adding a test case that shows that this warns even for completely covered switches e.g. https://godbolt.org/z/zbqr6P5xc

[dongjianqiang2]

See PR(#75900 )

[gendalph]

In real projects mix enum class with old int switch. Try to update defaults and have noise from enum class case.
My be mix options -Wswitch -Wswitch-default can disable warning in enum class case without default if full covered?

https://godbolt.org/z/xrdafn6WK

[gendalph]

Using an enum class without default is a good opportunity to get a warning in all places of use in the switch construct when adding new elements to the enum

[aaronpuchert]

@gendalph, this warning is meant to always warn if a default label is missing, even if all enumeration values are covered, like GCC does it. If you don't want a warning on enumerations, use the previously mentioned clang-tidy check bugprone-switch-missing-default-case. We can probably discuss adding this as warning, though I doubt we'll get consensus on that. This warning was added despite never going to be always-on because GCC has it (under the same name) and some code styles ask for it.