[clang][coverage] fixing "if constexpr" and "if consteval" coverage report

[hanickadot]

It was a while since I noticed coverage report is broken for "if constexpr" and "if consteval" (as shown on first picture).

Main problem was replacement of non-taken "if constexpr" branch with NullStmt but such object doesn't have begin/end for source location properly. So I introduced a new constructor for empty CompoundStmt and used it.

With "if consteval" I'm no longer introducing new branch counter for non-taken "branch". But in future it would be useful to mark whole gap there as skipped instead. If there is interest I would do it in another PR.

[llvmbot]

@llvm/pr-subscribers-clang-codegen

@llvm/pr-subscribers-clang

Author: Hana Dusíková (hanickadot)

Changes

It was a while since I noticed coverage report is broken for "if constexpr" and "if consteval" (as shown on first picture).
<img width="453" alt="Screenshot 2024-01-07 at 00 29 17" src="https://github.com/llvm/llvm-project/assets/6557263/dbdbc8a6-ad16-44da-882d-8e229ee69093">
(notice wrong coverage "if constexpr" for positive one, and completely missing for negative one, also notice "if consteval" marking always the same branch as uncovered)

Report after this change:
<img width="453" alt="Screenshot 2024-01-07 at 00 25 32" src="https://github.com/llvm/llvm-project/assets/6557263/d7ca85b0-34c7-40b5-9cc7-4efd8c18649b">

Main problem was replacement of non-taken "if constexpr" branch with NullStmt but such object doesn't have begin/end for source location properly. So I introduce a new constructor for empty CompoundStmt and used it.

With "if consteval" I'm no longer introducing new branch counter for non-taken "branch". But in future it would be useful to mark whole gap there as skipped instead. If there is interest I would do it in another PR.

---

Full diff: https://github.com/llvm/llvm-project/pull/77214.diff

3 Files Affected:

• (modified) clang/include/clang/AST/Stmt.h (+4-2)
• (modified) clang/lib/CodeGen/CoverageMappingGen.cpp (+11-2)
• (modified) clang/lib/Sema/TreeTransform.h (+6-1)

diff --git a/clang/include/clang/AST/Stmt.h b/clang/include/clang/AST/Stmt.h
index da7b37ce0e1211..fb50212083316e 100644
--- a/clang/include/clang/AST/Stmt.h
+++ b/clang/include/clang/AST/Stmt.h
@@ -1631,8 +1631,10 @@ class CompoundStmt final
                               SourceLocation RB);
 
   // Build an empty compound statement with a location.
-  explicit CompoundStmt(SourceLocation Loc)
-      : Stmt(CompoundStmtClass), LBraceLoc(Loc), RBraceLoc(Loc) {
+  explicit CompoundStmt(SourceLocation Loc) : CompoundStmt(Loc, Loc) {}
+
+  explicit CompoundStmt(SourceLocation Loc, SourceLocation EndLoc)
+      : Stmt(CompoundStmtClass), LBraceLoc(Loc), RBraceLoc(EndLoc) {
     CompoundStmtBits.NumStmts = 0;
     CompoundStmtBits.HasFPFeatures = 0;
   }
diff --git a/clang/lib/CodeGen/CoverageMappingGen.cpp b/clang/lib/CodeGen/CoverageMappingGen.cpp
index bf227386a71b78..b245abd16c3f4a 100644
--- a/clang/lib/CodeGen/CoverageMappingGen.cpp
+++ b/clang/lib/CodeGen/CoverageMappingGen.cpp
@@ -1712,7 +1712,11 @@ struct CounterCoverageMappingBuilder
       extendRegion(S->getCond());
 
     Counter ParentCount = getRegion().getCounter();
-    Counter ThenCount = getRegionCounter(S);
+
+    // If this is "if !consteval" the then-branch will never be taken, we don't
+    // need to change counter
+    Counter ThenCount =
+        S->isNegatedConsteval() ? ParentCount : getRegionCounter(S);
 
     if (!S->isConsteval()) {
       // Emitting a counter for the condition makes it easier to interpret the
@@ -1729,7 +1733,12 @@ struct CounterCoverageMappingBuilder
     extendRegion(S->getThen());
     Counter OutCount = propagateCounts(ThenCount, S->getThen());
 
-    Counter ElseCount = subtractCounters(ParentCount, ThenCount);
+    // If this is "if consteval" the else-branch will never be taken, we don't
+    // need to change counter
+    Counter ElseCount = S->isNonNegatedConsteval()
+                            ? ParentCount
+                            : subtractCounters(ParentCount, ThenCount);
+
     if (const Stmt *Else = S->getElse()) {
       bool ThenHasTerminateStmt = HasTerminateStmt;
       HasTerminateStmt = false;
diff --git a/clang/lib/Sema/TreeTransform.h b/clang/lib/Sema/TreeTransform.h
index c8c5a51bf9f94e..0033c851b618a1 100644
--- a/clang/lib/Sema/TreeTransform.h
+++ b/clang/lib/Sema/TreeTransform.h
@@ -7732,7 +7732,8 @@ TreeTransform<Derived>::TransformIfStmt(IfStmt *S) {
     if (Then.isInvalid())
       return StmtError();
   } else {
-    Then = new (getSema().Context) NullStmt(S->getThen()->getBeginLoc());
+    Then = new (getSema().Context)
+        CompoundStmt(S->getThen()->getBeginLoc(), S->getThen()->getEndLoc());
   }
 
   // Transform the "else" branch.
@@ -7741,6 +7742,10 @@ TreeTransform<Derived>::TransformIfStmt(IfStmt *S) {
     Else = getDerived().TransformStmt(S->getElse());
     if (Else.isInvalid())
       return StmtError();
+  } else if (S->getElse() && ConstexprConditionValue &&
+             *ConstexprConditionValue) {
+    Else = new (getSema().Context)
+        CompoundStmt(S->getElse()->getBeginLoc(), S->getElse()->getEndLoc());
   }
 
   if (!getDerived().AlwaysRebuild() &&

[hanickadot]

(notice wrong coverage "if constexpr" for positive one, and completely missing for negative one, also notice "if consteval" marking always the same branch as uncovered)

[cor3ntin]

Thanks for the patch!

I think this might need a release note (referencing the couple open issues about this bug), and probably some tests.

[hanickadot]

Fixes #54419.

[cor3ntin]

Adding a few folks to the review (We do not seem to update the coverage mapping very often)

[chfast]

But in future it would be useful to mark whole gap there as skipped instead. If there is interest I would do it in another PR.

This would be very useful not to treat compile-time unreachable code as uncovered. Can the skipped code be marked separately in the reports?

[hanickadot]

Yes, and it will be a bit bigger change. This is currently my biggest change yet :) But I want to do it as next. This fix's intention is to make the source location of regions properly done.

[hanickadot]

Added some tests. And also pinging @ornata for review.

[ornata]

Added some comments, mostly nits on the test.

[ornata]

LGTM

[cor3ntin]

LGTM
@hanickadot do you need me to merge that for you?

[hanickadot]

yes please, I don't have merge rights

[cor3ntin]

Oups, can you rebase?