[PATCH] c++ modules: ICE with bitfield member in class template

[PATCH] c++ modules: ICE with bitfield member in class template

[Patrick Palka]

According to grokbitfield, DECL_BITFIELD_REPRESENTATIVE may "temporarily"
contain the width of the bitfield until we layout the class type (after
which it'll contain a FIELD_DECL).  But for a class template, it'll always
be the width since we don't/can't layout dependent types.

Tested on x86_64-pc-linux-gnu, does this look OK for trunk?

gcc/cp/ChangeLog:

	* module.cc (trees_out::mark_class_def): Guard against
	DECL_BITFIELD_REPRESENTATIVE not being a decl.

gcc/testsuite/ChangeLog:

	* g++.dg/modules/bfield-3.H: New test.
---
 gcc/cp/module.cc                        | 3 ++-
 gcc/testsuite/g++.dg/modules/bfield-3.H | 8 ++++++++
 2 files changed, 10 insertions(+), 1 deletion(-)
 create mode 100644 gcc/testsuite/g++.dg/modules/bfield-3.H

diff --git a/gcc/cp/module.cc b/gcc/cp/module.cc
index cb1929bc5d5..172a72e92b9 100644
--- a/gcc/cp/module.cc
+++ b/gcc/cp/module.cc
@@ -11919,7 +11919,8 @@ trees_out::mark_class_def (tree defn)
 	mark_class_member (member);
 	if (TREE_CODE (member) == FIELD_DECL)
 	  if (tree repr = DECL_BIT_FIELD_REPRESENTATIVE (member))
-	    mark_declaration (repr, false);
+	    if (DECL_P (repr))
+	      mark_declaration (repr, false);
       }
 
   /* Mark the binfo hierarchy.  */
diff --git a/gcc/testsuite/g++.dg/modules/bfield-3.H b/gcc/testsuite/g++.dg/modules/bfield-3.H
new file mode 100644
index 00000000000..4fd4db7116a
--- /dev/null
+++ b/gcc/testsuite/g++.dg/modules/bfield-3.H
@@ -0,0 +1,8 @@
+// { dg-additional-options -fmodule-header }
+// { dg-module-cmi {} }
+
+template<int N>
+struct A {
+  int x : 1;
+  int y : N;
+};
-- 
2.38.0.rc2

Re: [PATCH] c++ modules: ICE with bitfield member in class template

[Nathan Sidwell]

On 10/7/22 11:09, Patrick Palka wrote:
> According to grokbitfield, DECL_BITFIELD_REPRESENTATIVE may "temporarily"
> contain the width of the bitfield until we layout the class type (after
> which it'll contain a FIELD_DECL).  But for a class template, it'll always
> be the width since we don't/can't layout dependent types.
> 
> Tested on x86_64-pc-linux-gnu, does this look OK for trunk?

ok, but could you add a comment on why it might not be a decl?


> 
> gcc/cp/ChangeLog:
> 
> 	* module.cc (trees_out::mark_class_def): Guard against
> 	DECL_BITFIELD_REPRESENTATIVE not being a decl.
> 
> gcc/testsuite/ChangeLog:
> 
> 	* g++.dg/modules/bfield-3.H: New test.
> ---
>   gcc/cp/module.cc                        | 3 ++-
>   gcc/testsuite/g++.dg/modules/bfield-3.H | 8 ++++++++
>   2 files changed, 10 insertions(+), 1 deletion(-)
>   create mode 100644 gcc/testsuite/g++.dg/modules/bfield-3.H
> 
> diff --git a/gcc/cp/module.cc b/gcc/cp/module.cc
> index cb1929bc5d5..172a72e92b9 100644
> --- a/gcc/cp/module.cc
> +++ b/gcc/cp/module.cc
> @@ -11919,7 +11919,8 @@ trees_out::mark_class_def (tree defn)
>   	mark_class_member (member);
>   	if (TREE_CODE (member) == FIELD_DECL)
>   	  if (tree repr = DECL_BIT_FIELD_REPRESENTATIVE (member))
> -	    mark_declaration (repr, false);
> +	    if (DECL_P (repr))
> +	      mark_declaration (repr, false);
>         }
>   
>     /* Mark the binfo hierarchy.  */
> diff --git a/gcc/testsuite/g++.dg/modules/bfield-3.H b/gcc/testsuite/g++.dg/modules/bfield-3.H
> new file mode 100644
> index 00000000000..4fd4db7116a
> --- /dev/null
> +++ b/gcc/testsuite/g++.dg/modules/bfield-3.H
> @@ -0,0 +1,8 @@
> +// { dg-additional-options -fmodule-header }
> +// { dg-module-cmi {} }
> +
> +template<int N>
> +struct A {
> +  int x : 1;
> +  int y : N;
> +};

-- 
Nathan Sidwell

Re: [PATCH] c++ modules: ICE with bitfield member in class template

[Patrick Palka]

On Fri, 7 Oct 2022, Nathan Sidwell wrote:

> On 10/7/22 11:09, Patrick Palka wrote:
> > According to grokbitfield, DECL_BITFIELD_REPRESENTATIVE may "temporarily"
> > contain the width of the bitfield until we layout the class type (after
> > which it'll contain a FIELD_DECL).  But for a class template, it'll always
> > be the width since we don't/can't layout dependent types.
> > 
> > Tested on x86_64-pc-linux-gnu, does this look OK for trunk?
> 
> ok, but could you add a comment on why it might not be a decl?

Thanks a lot, I added a comment and committed the following:

-- >8 --

Subject: [PATCH] c++ modules: ICE with bitfield in class template

According to grokbitfield, DECL_BIT_FIELD_REPRESENTATIVE may temporarily
contain the width of the bitfield until we layout the class type (after
which it'll contain a decl).  Thus for a bitfield in a class template
it'll always be the width, and this patch makes us avoid ICEing from
mark_class_def in this case.

gcc/cp/ChangeLog:

	* module.cc (trees_out::mark_class_def): Guard against
	DECL_BIT_FIELD_REPRESENTATIVE not being a decl.

gcc/testsuite/ChangeLog:

	* g++.dg/modules/bfield-3.H: New test.
---
 gcc/cp/module.cc                        | 6 +++++-
 gcc/testsuite/g++.dg/modules/bfield-3.H | 8 ++++++++
 2 files changed, 13 insertions(+), 1 deletion(-)
 create mode 100644 gcc/testsuite/g++.dg/modules/bfield-3.H

diff --git a/gcc/cp/module.cc b/gcc/cp/module.cc
index cb1929bc5d5..a7c199c00a4 100644
--- a/gcc/cp/module.cc
+++ b/gcc/cp/module.cc
@@ -11919,7 +11919,11 @@ trees_out::mark_class_def (tree defn)
 	mark_class_member (member);
 	if (TREE_CODE (member) == FIELD_DECL)
 	  if (tree repr = DECL_BIT_FIELD_REPRESENTATIVE (member))
-	    mark_declaration (repr, false);
+	    /* If we're marking a class template definition, then
+	       DECL_BIT_FIELD_REPRESENTATIVE will contain the width
+	       instead of a decl (as set by grokbitfield).  */
+	    if (DECL_P (repr))
+	      mark_declaration (repr, false);
       }
 
   /* Mark the binfo hierarchy.  */
diff --git a/gcc/testsuite/g++.dg/modules/bfield-3.H b/gcc/testsuite/g++.dg/modules/bfield-3.H
new file mode 100644
index 00000000000..4fd4db7116a
--- /dev/null
+++ b/gcc/testsuite/g++.dg/modules/bfield-3.H
@@ -0,0 +1,8 @@
+// { dg-additional-options -fmodule-header }
+// { dg-module-cmi {} }
+
+template<int N>
+struct A {
+  int x : 1;
+  int y : N;
+};
-- 
2.38.0.rc2