On 7/26/26 5:46 PM, Léo Hardt wrote:
On 2026-07-26 10:39, Jason Merrill wrote:
I'd also check seen_error () to make sure we've actually printed an
error; we
can also see ERROR_MARK in SFINAE cases where we haven't actually
given an
error.
Thanks!
Below is an updated version of the patch
Finally applied, this, thanks!
Best regards,
Léo
---
On some input files, we might try to mangle names and
run into an ERROR_MARK. It would be redundantto print
a "sorry, unimplemented: mangling error mark" message,
since an error has already been printed to the user.
See the following code as an example:
template<int... i> void f() {
(..., []() requires(i) { i; });
}
int main() { f<0,1,2>(); }
gcc/cp/ChangeLog:
* mangle.cc (write_expression): don't call
sorry when mangling error caused by previously
seen error.
Signed-off-by: Léo Hardt <[email protected]>
---
gcc/cp/mangle.cc | 4 ++++
1 file changed, 4 insertions(+)
diff --git a/gcc/cp/mangle.cc b/gcc/cp/mangle.cc
index 4ec57a54c34..e73714318e4 100644
--- a/gcc/cp/mangle.cc
+++ b/gcc/cp/mangle.cc
@@ -4063,6 +4063,10 @@ write_expression (tree expr)
"use library traits instead", expr);
break;
+ case ERROR_MARK:
+ if (seen_error ())
+ break;
+ /* FALLTHROUGH */
default:
sorry ("mangling %C", code);
break;
