| Message ID | 20210225063712.3725111-1-ppalka@redhat.com |
|---|---|
| State | New |
| Headers | show |
| Series | c++: Fix pretty printing of context of local class [PR99213] | expand |
On 2/25/21 1:37 AM, Patrick Palka wrote: > My r10-7705 patch for PR94521 made us set TFF_NO_FUNCTION_ARGUMENTS when > pretty printing the function scope of a local class type in order to > eliminate infinite recursion with a function signature that contains > decltype([]{}). But due to the way dump_function_decl works, this > change regressed our pretty printing of local class types whose context > contains a class template specialization, as in the testcase below, in > which we wrongly pretty print the two local types as 'A<T>::f<char>::S1' > and 'B<T>::f<int>::S2'. > > This patch makes dump_scope pass TFF_NO_TEMPLATE_BINDINGS instead of > TFF_NO_FUNCTION_ARGUMENTS when pretty printing a function scope. It > appears this is the strictly better flag to use: it avoids the infinite > recursion issue, it restores pretty printing of the function parameter > list, and it stops dump_function_decl from trying to print a function > template specialization in its own weird way. > > Summary of pretty printing differences for the below testcase: > > r10-7704: A<T>::f() [with U = char; T = int]::S1 > B<T>::f() [with T = int]::S2 > > r10-7705: A<T>::f<char>::S1 > B<T>::f<int>::S2 > > this patch: A<int>::f<char>()::S1 > B<int>::f()::S2 > > Bootstrapped and regtested on x86_64-pc-linux-gnu, does this look OK for > trunk? OK. > gcc/cp/ChangeLog: > > PR c++/99213 > PR c++/94521 > * error.c (dump_scope): Pass TFF_NO_TEMPLATE_BINDINGS instead of > TFF_NO_FUNCTION_ARGUMENTS when dumping a function scope. > > gcc/testsuite/ChangeLog: > > PR c++/99213 > PR c++/94521 > * g++.dg/diagnostic/local1.C: New test. > --- > gcc/cp/error.c | 4 +--- > gcc/testsuite/g++.dg/diagnostic/local1.C | 25 ++++++++++++++++++++++++ > 2 files changed, 26 insertions(+), 3 deletions(-) > create mode 100644 gcc/testsuite/g++.dg/diagnostic/local1.C > > diff --git a/gcc/cp/error.c b/gcc/cp/error.c > index 5213a8030ca..ff4ae6f4b23 100644 > --- a/gcc/cp/error.c > +++ b/gcc/cp/error.c > @@ -243,9 +243,7 @@ dump_scope (cxx_pretty_printer *pp, tree scope, int flags) > } > else if ((flags & TFF_SCOPE) && TREE_CODE (scope) == FUNCTION_DECL) > { > - if (DECL_USE_TEMPLATE (scope)) > - f |= TFF_NO_FUNCTION_ARGUMENTS; > - dump_function_decl (pp, scope, f); > + dump_function_decl (pp, scope, f | TFF_NO_TEMPLATE_BINDINGS); > pp_cxx_colon_colon (pp); > } > } > diff --git a/gcc/testsuite/g++.dg/diagnostic/local1.C b/gcc/testsuite/g++.dg/diagnostic/local1.C > new file mode 100644 > index 00000000000..5905b571a44 > --- /dev/null > +++ b/gcc/testsuite/g++.dg/diagnostic/local1.C > @@ -0,0 +1,25 @@ > +// PR c++/99213 > +// { dg-do compile { target c++14 } } > + > +template <class T> > +struct A { > + template <class U> > + static auto f() { > + struct S1{}; > + return S1{}; > + } > +}; > + > +using type = void; > +using type = decltype(A<int>::f<char>()); // { dg-error "A<int>::f<char>\\(\\)::S1" } > + > +template <class T> > +struct B { > + static auto f() { > + struct S2{}; > + return S2{}; > + } > +}; > + > +using type = void; > +using type = decltype(B<int>::f()); // { dg-error "B<int>::f\\(\\)::S2" } >
diff --git a/gcc/cp/error.c b/gcc/cp/error.c index 5213a8030ca..ff4ae6f4b23 100644 --- a/gcc/cp/error.c +++ b/gcc/cp/error.c @@ -243,9 +243,7 @@ dump_scope (cxx_pretty_printer *pp, tree scope, int flags) } else if ((flags & TFF_SCOPE) && TREE_CODE (scope) == FUNCTION_DECL) { - if (DECL_USE_TEMPLATE (scope)) - f |= TFF_NO_FUNCTION_ARGUMENTS; - dump_function_decl (pp, scope, f); + dump_function_decl (pp, scope, f | TFF_NO_TEMPLATE_BINDINGS); pp_cxx_colon_colon (pp); } } diff --git a/gcc/testsuite/g++.dg/diagnostic/local1.C b/gcc/testsuite/g++.dg/diagnostic/local1.C new file mode 100644 index 00000000000..5905b571a44 --- /dev/null +++ b/gcc/testsuite/g++.dg/diagnostic/local1.C @@ -0,0 +1,25 @@ +// PR c++/99213 +// { dg-do compile { target c++14 } } + +template <class T> +struct A { + template <class U> + static auto f() { + struct S1{}; + return S1{}; + } +}; + +using type = void; +using type = decltype(A<int>::f<char>()); // { dg-error "A<int>::f<char>\\(\\)::S1" } + +template <class T> +struct B { + static auto f() { + struct S2{}; + return S2{}; + } +}; + +using type = void; +using type = decltype(B<int>::f()); // { dg-error "B<int>::f\\(\\)::S2" }
My r10-7705 patch for PR94521 made us set TFF_NO_FUNCTION_ARGUMENTS when pretty printing the function scope of a local class type in order to eliminate infinite recursion with a function signature that contains decltype([]{}). But due to the way dump_function_decl works, this change regressed our pretty printing of local class types whose context contains a class template specialization, as in the testcase below, in which we wrongly pretty print the two local types as 'A<T>::f<char>::S1' and 'B<T>::f<int>::S2'. This patch makes dump_scope pass TFF_NO_TEMPLATE_BINDINGS instead of TFF_NO_FUNCTION_ARGUMENTS when pretty printing a function scope. It appears this is the strictly better flag to use: it avoids the infinite recursion issue, it restores pretty printing of the function parameter list, and it stops dump_function_decl from trying to print a function template specialization in its own weird way. Summary of pretty printing differences for the below testcase: r10-7704: A<T>::f() [with U = char; T = int]::S1 B<T>::f() [with T = int]::S2 r10-7705: A<T>::f<char>::S1 B<T>::f<int>::S2 this patch: A<int>::f<char>()::S1 B<int>::f()::S2 Bootstrapped and regtested on x86_64-pc-linux-gnu, does this look OK for trunk? gcc/cp/ChangeLog: PR c++/99213 PR c++/94521 * error.c (dump_scope): Pass TFF_NO_TEMPLATE_BINDINGS instead of TFF_NO_FUNCTION_ARGUMENTS when dumping a function scope. gcc/testsuite/ChangeLog: PR c++/99213 PR c++/94521 * g++.dg/diagnostic/local1.C: New test. --- gcc/cp/error.c | 4 +--- gcc/testsuite/g++.dg/diagnostic/local1.C | 25 ++++++++++++++++++++++++ 2 files changed, 26 insertions(+), 3 deletions(-) create mode 100644 gcc/testsuite/g++.dg/diagnostic/local1.C