================
@@ -2446,6 +2446,14 @@ static bool
interp__builtin_is_within_lifetime(InterpState &S, CodePtr OpPC,
return false;
}
+ if (!Ptr.isBlockPointer()) {
+ S.FFDiag(S.Current->getSource(OpPC),
+ diag::note_constexpr_access_unreadable_object)
+ << AK_Read << Ptr.toDiagnosticString(S.getASTContext());
+ noteValueLocation(S, Ptr);
----------------
ojhunt wrote:
Today in terrible wording: I mean "this changes the diagnostics produced" not
"the diagnostic message content changed".
The changes to noteValueLocation introduced a direct string type check which
does not seem to obviously exist in the original versions of these functions -
It's possible the temporaries path in `noteValueLocation(InterpState &S, const
Block *B)` previously captured this, but simply looking at the diff it looks
like the `note_constexpr_temporary_here` is not gated on the string being a
temporary.
So it _looks_ like the diagnostics change, beyond just the added one - I could
be wrong, if `noteValueLocation(InterpState &S, const Pointer &Ptr)` is only
ever called on a temporary then adding an assertion would make that clearer.
But the other point is that the check string for the diagnostic is "whose value
is not known" which is shared with `note_constexpr_access_unknown_variable` and
`note_constexpr_access_unreadable_object` so is insufficient to distinguish.
https://github.com/llvm/llvm-project/pull/228337
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
