Hi Jason, Thanks. Attached is v3 with the requested changes: the commit message is wrapped at 76 columns, the test is renamed, and the type check now uses INDIRECT_TYPE_P with the additional parentheses.
I kept the type guard because this adjustment is specifically for covariant pointer or reference returns. I also rebased onto current master. Tested on x86_64-pc-linux-gnu with no unexpected constexpr or consteval test failures. Thanks, Odysseas ________________________________ From: Jason Merrill <[email protected]> Sent: 14 August 2026 18:57 To: Odysseas Georgoudis <[email protected]>; [email protected] <[email protected]> Subject: Re: [PATCH v2] c++: Adjust constexpr covariant call results [PR126324] On 7/26/26 5:15 PM, Odysseas Georgoudis wrote: > Hi Jason, > > Thanks for catching the non-zero-offset case. > In v2, the result-thunk path now preserves null pointers, > evaluates the call only once, and adjusts the result to > the thunk’s static return type. > > The test covers both zero- and non-zero-offset bases. Looks good, just a few tweaks. > pointers, evaluate the call only once, and give the result the thunk's static > * constexpr.cc (cxx_eval_thunk_call): Preserve null pointer results and These lines are too long; the commit message should wrap at column 76 to accommodate 'git log' adding 4 spaces on the left. > * g++.dg/cpp2a/pr126324.C: New test. Let's add "constexpr-virtual" to the filename, i.e. constexpr-virtual-pr126324.C > + && scalarish_type_p (TREE_TYPE (t)) Let's use INDIRECT_TYPE_P since a covariant return type will be pointer or reference. Or drop this line entirely; is it needed? > + && !same_type_ignoring_top_level_qualifiers_p > + (TREE_TYPE (result), TREE_TYPE (t))) This needs another set of parens starting either before or after the ! to prevent emacs from moving the args to line up with the &&. Jason
PR126324-v3.patch
Description: PR126324-v3.patch
