Re: [PATCH v2] c++: fix ICE with constexpr ARRAY_REF [PR110382]
On 7/24/23 18:37, Marek Polacek wrote: On Sat, Jul 22, 2023 at 12:28:59AM -0400, Jason Merrill wrote: On 7/21/23 18:38, Marek Polacek wrote: Bootstrapped/regtested on x86_64-pc-linux-gnu, ok for trunk/13? -- >8 -- This code in cxx_eval_array_reference has been hard to get right. In r12-2304 I added some code; in r13-5693 I removed some of it. Here the problematic line is "S s = arr[0];" which causes a crash on the assert in verify_ctor_sanity: gcc_assert (!ctx->object || !DECL_P (ctx->object) || ctx->global->get_value (ctx->object) == ctx->ctor); ctx->object is the VAR_DECL 's', which is correct here. The second line points to the problem: we replaced ctx->ctor in cxx_eval_array_reference: new_ctx.ctor = build_constructor (elem_type, NULL); // #1 ...and this code doesn't also clear(/set) new_ctx.object like everywhere else in constexpr.cc that sets new_ctx.ctor. Fixing that should make the testcase work. Right, but then we'd be back pre-r12-2304 or r13-5693... ...except it should work to always clear the object, like below. which I think we shouldn't have; the CONSTRUCTOR we created in cxx_eval_constant_expression/DECL_EXPR new_ctx.ctor = build_constructor (TREE_TYPE (r), NULL); had the right type. Indeed, and using it rather than building a new one seems like a valid optimization for trunk. Agreed, I kept it. I also notice that the DECL_EXPR code calls unshare_constructor, which should be unnecessary if init == ctx->ctor? It looks like init == ctx->ctor only happens only with this new testcase. I'm not sure it's worth it adding code for such a rare case? We still need #1 though. E.g., in constexpr-96241.C, we never set ctx.ctor/object before calling cxx_eval_array_reference, so we have to build a CONSTRUCTOR there. And in constexpr-101371-2.C we have a ctx.ctor, but it has the wrong type, so we need a new one. PR c++/110382 gcc/cp/ChangeLog: * constexpr.cc (cxx_eval_array_reference): Create a new constructor only when we don't already have a matching one. gcc/testsuite/ChangeLog: * g++.dg/cpp1y/constexpr-110382.C: New test. --- gcc/cp/constexpr.cc | 5 - gcc/testsuite/g++.dg/cpp1y/constexpr-110382.C | 17 + 2 files changed, 21 insertions(+), 1 deletion(-) create mode 100644 gcc/testsuite/g++.dg/cpp1y/constexpr-110382.C diff --git a/gcc/cp/constexpr.cc b/gcc/cp/constexpr.cc index fb94f3cefcb..518b7c7a2d5 100644 --- a/gcc/cp/constexpr.cc +++ b/gcc/cp/constexpr.cc @@ -4291,7 +4291,10 @@ cxx_eval_array_reference (const constexpr_ctx *ctx, tree t, else val = build_value_init (elem_type, tf_warning_or_error); - if (!SCALAR_TYPE_P (elem_type)) + if (!SCALAR_TYPE_P (elem_type) + /* Create a new constructor only if we don't already have one that +is suitable. */ + && !(ctx->ctor && same_type_p (elem_type, TREE_TYPE (ctx->ctor We generally use same_type_ignoring_top_level_qualifiers_p in the constexpr code. True, changed. Bootstrapped/regtested on x86_64-pc-linux-gnu, ok for trunk? For 13, I guess I should only clear the object and leave out the same_type_ bit. -- >8 -- This code in cxx_eval_array_reference has been hard to get right. In r12-2304 I added some code; in r13-5693 I removed some of it. Here the problematic line is "S s = arr[0];" which causes a crash on the assert in verify_ctor_sanity: gcc_assert (!ctx->object || !DECL_P (ctx->object) || ctx->global->get_value (ctx->object) == ctx->ctor); ctx->object is the VAR_DECL 's', which is correct here. The second line points to the problem: we replaced ctx->ctor in cxx_eval_array_reference: new_ctx.ctor = build_constructor (elem_type, NULL); // #1 which I think we shouldn't have; the CONSTRUCTOR we created in cxx_eval_constant_expression/DECL_EXPR new_ctx.ctor = build_constructor (TREE_TYPE (r), NULL); had the right type. We still need #1 though. E.g., in constexpr-96241.C, we never set ctx.ctor/object before calling cxx_eval_array_reference, so we have to build a CONSTRUCTOR there. And in constexpr-101371-2.C we have a ctx.ctor, but it has the wrong type, so we need a new one. We can fix the problem by always clearing the object, and, as an optimization, only create/free a new ctor when actually needed. PR c++/110382 gcc/cp/ChangeLog: * constexpr.cc (cxx_eval_array_reference): Create a new constructor only when we don't already have a matching one. Clear the object when the type is non-scalar. gcc/testsuite/ChangeLog: * g++.dg/cpp1y/constexpr-110382.C: New test. --- gcc/cp/constexpr.cc | 17 +++-- gcc/testsuite/g++.dg/cpp1y/constexpr-110382.C | 17 + 2 files changed, 32 insertions(+), 2 deletions(-) create mode 100644 gcc/testsuite/g++.dg/cpp1y/constexpr-110382.C diff --git a/gcc/cp/constexpr.cc
[PATCH v2] c++: fix ICE with constexpr ARRAY_REF [PR110382]
On Sat, Jul 22, 2023 at 12:28:59AM -0400, Jason Merrill wrote: > On 7/21/23 18:38, Marek Polacek wrote: > > Bootstrapped/regtested on x86_64-pc-linux-gnu, ok for trunk/13? > > > > -- >8 -- > > > > This code in cxx_eval_array_reference has been hard to get right. > > In r12-2304 I added some code; in r13-5693 I removed some of it. > > > > Here the problematic line is "S s = arr[0];" which causes a crash > > on the assert in verify_ctor_sanity: > > > >gcc_assert (!ctx->object || !DECL_P (ctx->object) > >|| ctx->global->get_value (ctx->object) == ctx->ctor); > > > > ctx->object is the VAR_DECL 's', which is correct here. The second > > line points to the problem: we replaced ctx->ctor in > > cxx_eval_array_reference: > > > >new_ctx.ctor = build_constructor (elem_type, NULL); // #1 > > ...and this code doesn't also clear(/set) new_ctx.object like everywhere > else in constexpr.cc that sets new_ctx.ctor. Fixing that should make the > testcase work. Right, but then we'd be back pre-r12-2304 or r13-5693... ...except it should work to always clear the object, like below. > > which I think we shouldn't have; the CONSTRUCTOR we created in > > cxx_eval_constant_expression/DECL_EXPR > > > >new_ctx.ctor = build_constructor (TREE_TYPE (r), NULL); > > > > had the right type. > > Indeed, and using it rather than building a new one seems like a valid > optimization for trunk. Agreed, I kept it. > I also notice that the DECL_EXPR code calls unshare_constructor, which > should be unnecessary if init == ctx->ctor? It looks like init == ctx->ctor only happens only with this new testcase. I'm not sure it's worth it adding code for such a rare case? > > We still need #1 though. E.g., in constexpr-96241.C, we never > > set ctx.ctor/object before calling cxx_eval_array_reference, so > > we have to build a CONSTRUCTOR there. And in constexpr-101371-2.C > > we have a ctx.ctor, but it has the wrong type, so we need a new one. > > > > PR c++/110382 > > > > gcc/cp/ChangeLog: > > > > * constexpr.cc (cxx_eval_array_reference): Create a new constructor > > only when we don't already have a matching one. > > > > gcc/testsuite/ChangeLog: > > > > * g++.dg/cpp1y/constexpr-110382.C: New test. > > --- > > gcc/cp/constexpr.cc | 5 - > > gcc/testsuite/g++.dg/cpp1y/constexpr-110382.C | 17 + > > 2 files changed, 21 insertions(+), 1 deletion(-) > > create mode 100644 gcc/testsuite/g++.dg/cpp1y/constexpr-110382.C > > > > diff --git a/gcc/cp/constexpr.cc b/gcc/cp/constexpr.cc > > index fb94f3cefcb..518b7c7a2d5 100644 > > --- a/gcc/cp/constexpr.cc > > +++ b/gcc/cp/constexpr.cc > > @@ -4291,7 +4291,10 @@ cxx_eval_array_reference (const constexpr_ctx *ctx, > > tree t, > > else > > val = build_value_init (elem_type, tf_warning_or_error); > > - if (!SCALAR_TYPE_P (elem_type)) > > + if (!SCALAR_TYPE_P (elem_type) > > + /* Create a new constructor only if we don't already have one that > > +is suitable. */ > > + && !(ctx->ctor && same_type_p (elem_type, TREE_TYPE (ctx->ctor > > We generally use same_type_ignoring_top_level_qualifiers_p in the constexpr > code. True, changed. Bootstrapped/regtested on x86_64-pc-linux-gnu, ok for trunk? For 13, I guess I should only clear the object and leave out the same_type_ bit. -- >8 -- This code in cxx_eval_array_reference has been hard to get right. In r12-2304 I added some code; in r13-5693 I removed some of it. Here the problematic line is "S s = arr[0];" which causes a crash on the assert in verify_ctor_sanity: gcc_assert (!ctx->object || !DECL_P (ctx->object) || ctx->global->get_value (ctx->object) == ctx->ctor); ctx->object is the VAR_DECL 's', which is correct here. The second line points to the problem: we replaced ctx->ctor in cxx_eval_array_reference: new_ctx.ctor = build_constructor (elem_type, NULL); // #1 which I think we shouldn't have; the CONSTRUCTOR we created in cxx_eval_constant_expression/DECL_EXPR new_ctx.ctor = build_constructor (TREE_TYPE (r), NULL); had the right type. We still need #1 though. E.g., in constexpr-96241.C, we never set ctx.ctor/object before calling cxx_eval_array_reference, so we have to build a CONSTRUCTOR there. And in constexpr-101371-2.C we have a ctx.ctor, but it has the wrong type, so we need a new one. We can fix the problem by always clearing the object, and, as an optimization, only create/free a new ctor when actually needed. PR c++/110382 gcc/cp/ChangeLog: * constexpr.cc (cxx_eval_array_reference): Create a new constructor only when we don't already have a matching one. Clear the object when the type is non-scalar. gcc/testsuite/ChangeLog: * g++.dg/cpp1y/constexpr-110382.C: New test. --- gcc/cp/constexpr.cc | 17 +++-- gcc/testsuite/g++.dg/cpp1y/constexpr-110382.C | 17 + 2
