[PATCH] D57086: Ignore trailing NullStmts in StmtExprs for GCC compatibility

[Aaron Ballman via Phabricator via cfe-commits]

aaron.ballman added inline comments.


================
Comment at: clang/include/clang/AST/Stmt.h:1300
+  void setStmtExpr(Stmt *S) {
+    assert(!body_empty() && "setStmtExpr");
+    unsigned ExprResult = getLastNonNullStmt();
----------------
domdom wrote:
> getLastNonNullStmt asserts anyway, should I remove this?
If you did switch to using an optional, then you can assert the optional has a value here. The effect should be the same as your current approach, but with slightly more tolerance.


================
Comment at: clang/include/clang/AST/Stmt.h:1259
+  // the index of the last one.
+  unsigned getIndexOfLastNonNullStmt() const {
+    assert(!body_empty() && "getIndexOfLastNonNullStmt");
----------------
What do you think about having this function return an `optional<unsigned>` rather than asserting? This will keep the API safe when called on an empty `CompoundStmt`.


================
Comment at: clang/include/clang/AST/Stmt.h:1309
+  Stmt *getStmtExprResult() const {
+    unsigned ExprResult = getIndexOfLastNonNullStmt();
+    return body_begin()[ExprResult];
----------------
And if `getLastIndexOfNonNullStmt()` returns an empty value here, you can return `nullptr` to signify that there is no statement expression result possible.


================
Comment at: clang/lib/CodeGen/CGStmt.cpp:390
+      E = S.body_end(); I != E; ++I) {
+    if (GetLast && ExprResult == *I) {
+      // We have to special case labels here.  They are statements, but when put
----------------
domdom wrote:
> aaron.ballman wrote:
> > What happens if `ExprResult` is `nullptr`?
> Then ExprResult == *I should not be evaluated. (Since GetLast would be false)
I think I was worried about a different situation that isn't present in this patch. However, if you switch to using optional, then `getStmtExprResult()` could potentially return null as well (when `GetLast` is true), so we'd have to handle that (probably with an assertion).


================
Comment at: clang/test/CodeGen/exprs.c:202
+int f19() {
+  return ({ 3;;4;; });
+}
----------------
domdom wrote:
> aaron.ballman wrote:
> > Does this test need the extra null statement between the `3;` and `4;`?
> Not strictly speaking, no. Just added it to ensure it has no effect.
That sounds good to me!


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D57086/new/

https://reviews.llvm.org/D57086