diff mbox series

[v2] c++: FIX_TRUNC_EXPR in tsubst [PR102990]

Message ID YjykYDP3avLDvlhP@redhat.com
State New
Headers show
Series [v2] c++: FIX_TRUNC_EXPR in tsubst [PR102990] | expand

Commit Message

Marek Polacek March 24, 2022, 5:03 p.m. UTC
On Thu, Mar 24, 2022 at 09:32:19AM -0400, Jason Merrill wrote:
> On 3/23/22 19:26, Marek Polacek wrote:
> > On Wed, Mar 23, 2022 at 04:35:32PM -0400, Jason Merrill wrote:
> > > On 3/22/22 19:55, Marek Polacek wrote:
> > > > This is a crash where a FIX_TRUNC_EXPR gets into tsubst_copy_and_build
> > > > where it hits gcc_unreachable ().
> > > > 
> > > > The history of tsubst_copy_and_build/FIX_TRUNC_EXPR is such that it
> > > > was introduced in r181478, but it did the wrong thing, whereupon it
> > > > was turned into gcc_unreachable () in r258821 (see this thread:
> > > > <https://gcc.gnu.org/pipermail/gcc-patches/2018-March/495853.html>).
> > > > 
> > > > In a template, we should never create a FIX_TRUNC_EXPR (that's what
> > > > conv_unsafe_in_template_p is for).  But in this test we are NOT in
> > > > a template when we call digest_nsdmi_init which ends up calling
> > > > convert_like, converting 1.0e+0 to int, so convert_to_integer_1
> > > > gives us a FIX_TRUNC_EXPR.
> > > > 
> > > > But then when we get to parsing f's parameters, we are in a template
> > > > when processing decltype(Helpers{}), and since r268321, when the
> > > > compound literal isn't instantiation-dependent and the type isn't
> > > > type-dependent, finish_compound_literal falls back to the normal
> > > > processing, so it calls digest_init, which does fold_non_dependent_init
> > > > and since the FIX_TRUNC_EXPR isn't dependent, we instantiate and
> > > > therefore crash in tsubst_copy_and_build.
> > > 
> > > Hmm, we shouldn't be doing fold_non_dependent_init on the result of
> > > get_nsdmi.  Why does that happen?
> > 
> > OK, so we have decltype(Helpers{}), finish_compound_literal gets
> > Helpers{}, it's not type-dependent, so:
> > 
> > - call digest_init (type=Helpers, init={})
> >    init is BRACE_ENCLOSED_INITIALIZER_P, type is !TYPE_NON_AGGREGATE_CLASS
> >    so go down to...
> > - process_init_constructor_record (type=Helpers, init={})
> >     - we walk the fields of Helpers, there's 'inputs' of type knob_t
> >       type_build_ctor_call (knob_t) is true, we want to default-init this
> >       field
> >     - create a {} as the init for default-initialization
> >     - call massage_init_elt (type=knob_t, init={}) to adjust the init
> >       - we do so by calling digest_init_r (type=knob_t, init={})
> >         - here we again call process_init_constructor_record, walk knob_t's
> >           fields, see the field 'value', it has a DECL_INITIAL, so call
> > 	 get_nsdmi, that returns the FIX_TRUNC_EXPR we've created before
> >         - so digesting {} for knob_t produced
> >           init = {.value=(int) NON_LVALUE_EXPR <1.0e+0>}
> >       - then we call fold_non_dependent_init on this init, and die
> 
> Maybe we shouldn't call fold_non_dependent_init on a CONSTRUCTOR here, since
> presumably we already called it on elements that needed it in the recursive
> digest_init_r.

Ah, that works.  It's still a bit weird that we don't treat FLOAT_EXPR and
FIX_TRUNC_EXPR the same.

I've tried to remove the call to fold_non_dependent_init in massage_init_elt
to see what breaks...a lot.  So it has to stay at least in this form.

Bootstrapped/regtested on x86_64-pc-linux-gnu, ok for trunk?

Thanks,

-- >8 --
This is a crash where a FIX_TRUNC_EXPR gets into tsubst_copy_and_build
where it hits gcc_unreachable ().

The history of tsubst_copy_and_build/FIX_TRUNC_EXPR is such that it
was introduced in r181478, but it did the wrong thing, whereupon it
was turned into gcc_unreachable () in r258821 (see this thread:
<https://gcc.gnu.org/pipermail/gcc-patches/2018-March/495853.html>).

In a template, we should never create a FIX_TRUNC_EXPR (that's what
conv_unsafe_in_template_p is for).  But in this test we are NOT in
a template when we call digest_nsdmi_init which ends up calling
convert_like, converting 1.0e+0 to int, so convert_to_integer_1
gives us a FIX_TRUNC_EXPR.

But then when we get to parsing f's parameters, we are in a template
when processing decltype(Helpers{}), and since r268321, when the
compound literal isn't instantiation-dependent and the type isn't
type-dependent, finish_compound_literal falls back to the normal
processing, so it calls digest_init, which does fold_non_dependent_init
and since the FIX_TRUNC_EXPR isn't dependent, we instantiate and
therefore crash in tsubst_copy_and_build.

The fateful call to fold_non_dependent_init comes from massage_init_elt,
We shouldn't be calling f_n_d_i on the result of get_nsdmi.  This we can
avoid by eschewing calling f_n_d_i on CONSTRUCTORs; their elements have
already been folded.

	PR c++/102990

gcc/cp/ChangeLog:

	* typeck2.cc (massage_init_elt): Avoid folding CONSTRUCTORs.

gcc/testsuite/ChangeLog:

	* g++.dg/cpp0x/nsdmi-template22.C: New test.
	* g++.dg/cpp0x/nsdmi-template23.C: New test.
---
 gcc/cp/typeck2.cc                             | 13 +++++++++----
 gcc/testsuite/g++.dg/cpp0x/nsdmi-template22.C | 13 +++++++++++++
 gcc/testsuite/g++.dg/cpp0x/nsdmi-template23.C | 13 +++++++++++++
 3 files changed, 35 insertions(+), 4 deletions(-)
 create mode 100644 gcc/testsuite/g++.dg/cpp0x/nsdmi-template22.C
 create mode 100644 gcc/testsuite/g++.dg/cpp0x/nsdmi-template23.C


base-commit: fb488cba571539b6644e8f99f1dd997cdb4c82c1

Comments

Jason Merrill March 24, 2022, 5:11 p.m. UTC | #1
On 3/24/22 13:03, Marek Polacek wrote:
> On Thu, Mar 24, 2022 at 09:32:19AM -0400, Jason Merrill wrote:
>> On 3/23/22 19:26, Marek Polacek wrote:
>>> On Wed, Mar 23, 2022 at 04:35:32PM -0400, Jason Merrill wrote:
>>>> On 3/22/22 19:55, Marek Polacek wrote:
>>>>> This is a crash where a FIX_TRUNC_EXPR gets into tsubst_copy_and_build
>>>>> where it hits gcc_unreachable ().
>>>>>
>>>>> The history of tsubst_copy_and_build/FIX_TRUNC_EXPR is such that it
>>>>> was introduced in r181478, but it did the wrong thing, whereupon it
>>>>> was turned into gcc_unreachable () in r258821 (see this thread:
>>>>> <https://gcc.gnu.org/pipermail/gcc-patches/2018-March/495853.html>).
>>>>>
>>>>> In a template, we should never create a FIX_TRUNC_EXPR (that's what
>>>>> conv_unsafe_in_template_p is for).  But in this test we are NOT in
>>>>> a template when we call digest_nsdmi_init which ends up calling
>>>>> convert_like, converting 1.0e+0 to int, so convert_to_integer_1
>>>>> gives us a FIX_TRUNC_EXPR.
>>>>>
>>>>> But then when we get to parsing f's parameters, we are in a template
>>>>> when processing decltype(Helpers{}), and since r268321, when the
>>>>> compound literal isn't instantiation-dependent and the type isn't
>>>>> type-dependent, finish_compound_literal falls back to the normal
>>>>> processing, so it calls digest_init, which does fold_non_dependent_init
>>>>> and since the FIX_TRUNC_EXPR isn't dependent, we instantiate and
>>>>> therefore crash in tsubst_copy_and_build.
>>>>
>>>> Hmm, we shouldn't be doing fold_non_dependent_init on the result of
>>>> get_nsdmi.  Why does that happen?
>>>
>>> OK, so we have decltype(Helpers{}), finish_compound_literal gets
>>> Helpers{}, it's not type-dependent, so:
>>>
>>> - call digest_init (type=Helpers, init={})
>>>     init is BRACE_ENCLOSED_INITIALIZER_P, type is !TYPE_NON_AGGREGATE_CLASS
>>>     so go down to...
>>> - process_init_constructor_record (type=Helpers, init={})
>>>      - we walk the fields of Helpers, there's 'inputs' of type knob_t
>>>        type_build_ctor_call (knob_t) is true, we want to default-init this
>>>        field
>>>      - create a {} as the init for default-initialization
>>>      - call massage_init_elt (type=knob_t, init={}) to adjust the init
>>>        - we do so by calling digest_init_r (type=knob_t, init={})
>>>          - here we again call process_init_constructor_record, walk knob_t's
>>>            fields, see the field 'value', it has a DECL_INITIAL, so call
>>> 	 get_nsdmi, that returns the FIX_TRUNC_EXPR we've created before
>>>          - so digesting {} for knob_t produced
>>>            init = {.value=(int) NON_LVALUE_EXPR <1.0e+0>}
>>>        - then we call fold_non_dependent_init on this init, and die
>>
>> Maybe we shouldn't call fold_non_dependent_init on a CONSTRUCTOR here, since
>> presumably we already called it on elements that needed it in the recursive
>> digest_init_r.
> 
> Ah, that works.  It's still a bit weird that we don't treat FLOAT_EXPR and
> FIX_TRUNC_EXPR the same.
> 
> I've tried to remove the call to fold_non_dependent_init in massage_init_elt
> to see what breaks...a lot.  So it has to stay at least in this form.
> 
> Bootstrapped/regtested on x86_64-pc-linux-gnu, ok for trunk?

OK.

> Thanks,
> 
> -- >8 --
> This is a crash where a FIX_TRUNC_EXPR gets into tsubst_copy_and_build
> where it hits gcc_unreachable ().
> 
> The history of tsubst_copy_and_build/FIX_TRUNC_EXPR is such that it
> was introduced in r181478, but it did the wrong thing, whereupon it
> was turned into gcc_unreachable () in r258821 (see this thread:
> <https://gcc.gnu.org/pipermail/gcc-patches/2018-March/495853.html>).
> 
> In a template, we should never create a FIX_TRUNC_EXPR (that's what
> conv_unsafe_in_template_p is for).  But in this test we are NOT in
> a template when we call digest_nsdmi_init which ends up calling
> convert_like, converting 1.0e+0 to int, so convert_to_integer_1
> gives us a FIX_TRUNC_EXPR.
> 
> But then when we get to parsing f's parameters, we are in a template
> when processing decltype(Helpers{}), and since r268321, when the
> compound literal isn't instantiation-dependent and the type isn't
> type-dependent, finish_compound_literal falls back to the normal
> processing, so it calls digest_init, which does fold_non_dependent_init
> and since the FIX_TRUNC_EXPR isn't dependent, we instantiate and
> therefore crash in tsubst_copy_and_build.
> 
> The fateful call to fold_non_dependent_init comes from massage_init_elt,
> We shouldn't be calling f_n_d_i on the result of get_nsdmi.  This we can
> avoid by eschewing calling f_n_d_i on CONSTRUCTORs; their elements have
> already been folded.
> 
> 	PR c++/102990
> 
> gcc/cp/ChangeLog:
> 
> 	* typeck2.cc (massage_init_elt): Avoid folding CONSTRUCTORs.
> 
> gcc/testsuite/ChangeLog:
> 
> 	* g++.dg/cpp0x/nsdmi-template22.C: New test.
> 	* g++.dg/cpp0x/nsdmi-template23.C: New test.
> ---
>   gcc/cp/typeck2.cc                             | 13 +++++++++----
>   gcc/testsuite/g++.dg/cpp0x/nsdmi-template22.C | 13 +++++++++++++
>   gcc/testsuite/g++.dg/cpp0x/nsdmi-template23.C | 13 +++++++++++++
>   3 files changed, 35 insertions(+), 4 deletions(-)
>   create mode 100644 gcc/testsuite/g++.dg/cpp0x/nsdmi-template22.C
>   create mode 100644 gcc/testsuite/g++.dg/cpp0x/nsdmi-template23.C
> 
> diff --git a/gcc/cp/typeck2.cc b/gcc/cp/typeck2.cc
> index a4c825fc34d..cebe6acf487 100644
> --- a/gcc/cp/typeck2.cc
> +++ b/gcc/cp/typeck2.cc
> @@ -1433,10 +1433,15 @@ massage_init_elt (tree type, tree init, int nested, int flags,
>       new_flags |= LOOKUP_AGGREGATE_PAREN_INIT;
>     init = digest_init_r (type, init, nested ? 2 : 1, new_flags, complain);
>     /* When we defer constant folding within a statement, we may want to
> -     defer this folding as well.  */
> -  tree t = fold_non_dependent_init (init, complain);
> -  if (TREE_CONSTANT (t))
> -    init = t;
> +     defer this folding as well.  Don't call this on CONSTRUCTORs because
> +     their elements have already been folded, and we must avoid folding
> +     the result of get_nsdmi.  */
> +  if (TREE_CODE (init) != CONSTRUCTOR)
> +    {
> +      tree t = fold_non_dependent_init (init, complain);
> +      if (TREE_CONSTANT (t))
> +	init = t;
> +    }
>     return init;
>   }
>   
> diff --git a/gcc/testsuite/g++.dg/cpp0x/nsdmi-template22.C b/gcc/testsuite/g++.dg/cpp0x/nsdmi-template22.C
> new file mode 100644
> index 00000000000..4ed2501035c
> --- /dev/null
> +++ b/gcc/testsuite/g++.dg/cpp0x/nsdmi-template22.C
> @@ -0,0 +1,13 @@
> +// PR c++/102990
> +// { dg-do compile { target c++11 } }
> +
> +struct knob_t {
> +  /* Let's create a FIX_TRUNC_EXPR.  */
> +  int value = 1.0;
> +};
> +
> +struct Helpers {
> +  knob_t inputs;
> +};
> +
> +template<class> void f(decltype(Helpers{}));
> diff --git a/gcc/testsuite/g++.dg/cpp0x/nsdmi-template23.C b/gcc/testsuite/g++.dg/cpp0x/nsdmi-template23.C
> new file mode 100644
> index 00000000000..240cab4347a
> --- /dev/null
> +++ b/gcc/testsuite/g++.dg/cpp0x/nsdmi-template23.C
> @@ -0,0 +1,13 @@
> +// PR c++/102990
> +// { dg-do compile { target c++11 } }
> +
> +struct knob_t {
> +  /* Let's create a FLOAT_EXPR.  */
> +  double value = 1UL;
> +};
> +
> +struct Helpers {
> +  knob_t inputs;
> +};
> +
> +template<class> void f(decltype(Helpers{}));
> 
> base-commit: fb488cba571539b6644e8f99f1dd997cdb4c82c1
diff mbox series

Patch

diff --git a/gcc/cp/typeck2.cc b/gcc/cp/typeck2.cc
index a4c825fc34d..cebe6acf487 100644
--- a/gcc/cp/typeck2.cc
+++ b/gcc/cp/typeck2.cc
@@ -1433,10 +1433,15 @@  massage_init_elt (tree type, tree init, int nested, int flags,
     new_flags |= LOOKUP_AGGREGATE_PAREN_INIT;
   init = digest_init_r (type, init, nested ? 2 : 1, new_flags, complain);
   /* When we defer constant folding within a statement, we may want to
-     defer this folding as well.  */
-  tree t = fold_non_dependent_init (init, complain);
-  if (TREE_CONSTANT (t))
-    init = t;
+     defer this folding as well.  Don't call this on CONSTRUCTORs because
+     their elements have already been folded, and we must avoid folding
+     the result of get_nsdmi.  */
+  if (TREE_CODE (init) != CONSTRUCTOR)
+    {
+      tree t = fold_non_dependent_init (init, complain);
+      if (TREE_CONSTANT (t))
+	init = t;
+    }
   return init;
 }
 
diff --git a/gcc/testsuite/g++.dg/cpp0x/nsdmi-template22.C b/gcc/testsuite/g++.dg/cpp0x/nsdmi-template22.C
new file mode 100644
index 00000000000..4ed2501035c
--- /dev/null
+++ b/gcc/testsuite/g++.dg/cpp0x/nsdmi-template22.C
@@ -0,0 +1,13 @@ 
+// PR c++/102990
+// { dg-do compile { target c++11 } }
+
+struct knob_t {
+  /* Let's create a FIX_TRUNC_EXPR.  */
+  int value = 1.0;
+};
+
+struct Helpers {
+  knob_t inputs;
+};
+
+template<class> void f(decltype(Helpers{}));
diff --git a/gcc/testsuite/g++.dg/cpp0x/nsdmi-template23.C b/gcc/testsuite/g++.dg/cpp0x/nsdmi-template23.C
new file mode 100644
index 00000000000..240cab4347a
--- /dev/null
+++ b/gcc/testsuite/g++.dg/cpp0x/nsdmi-template23.C
@@ -0,0 +1,13 @@ 
+// PR c++/102990
+// { dg-do compile { target c++11 } }
+
+struct knob_t {
+  /* Let's create a FLOAT_EXPR.  */
+  double value = 1UL;
+};
+
+struct Helpers {
+  knob_t inputs;
+};
+
+template<class> void f(decltype(Helpers{}));