[C++ Patch] PR 84618 ("[8 Regression] ICE in build_capture_proxy, at cp/lambda.c:460")

[C++ Patch] PR 84618 ("[8 Regression] ICE in build_capture_proxy, at cp/lambda.c:460")

[Paolo Carlini]

[-- Attachment #1: Type: text/plain, Size: 542 bytes --]

Hi,

a rather simple ice on invalid (not sure why only P4 given that no 
meaningful diagnostic is emitted before ICEing). What happens is that 
the ill-formed capture naming b, an OVERLOAD, escapes the early check in 
cp_parser_lambda_introducer to eventually trigger a recently added 
gcc_assert in build_capture_proxy. Adjusting the check means we also 
provide slightly clearer (IMO, matching clang, anyway) error messages 
for two testcases we already have.

Tested x86_64-linux.

Thanks, Paolo.

///////////////////////////////////////


[-- Attachment #2: CL_84618 --]
[-- Type: text/plain, Size: 398 bytes --]

/cp
2018-03-05  Paolo Carlini  <paolo.carlini@oracle.com>

	PR c++/84618
	* parser.c (cp_parser_lambda_introducer): Reject any capture not
	involving a VAR_DECL or a PARM_DECL.

/testsuite
2018-03-05  Paolo Carlini  <paolo.carlini@oracle.com>

	PR c++/84618
	* g++.dg/cpp0x/lambda/lambda-ice29.C: New.
	* g++.dg/cpp0x/lambda/lambda-ice17.C: Adjust.
	* g++.dg/cpp0x/lambda/lambda-ice23.C: Likewise.

[-- Attachment #3: patch_84618 --]
[-- Type: text/plain, Size: 2347 bytes --]

Index: cp/parser.c
===================================================================
--- cp/parser.c	(revision 258235)
+++ cp/parser.c	(working copy)
@@ -10377,15 +10377,15 @@ cp_parser_lambda_introducer (cp_parser* parser, tr
 	      unqualified_name_lookup_error (capture_id);
 	      continue;
 	    }
-	  else if (DECL_P (capture_init_expr)
-		   && (!VAR_P (capture_init_expr)
-		       && TREE_CODE (capture_init_expr) != PARM_DECL))
+	  else if (!VAR_P (capture_init_expr)
+		   && TREE_CODE (capture_init_expr) != PARM_DECL)
 	    {
 	      error_at (capture_token->location,
-			"capture of non-variable %qD ",
+			"capture of non-variable %qE ",
 			capture_init_expr);
-	      inform (DECL_SOURCE_LOCATION (capture_init_expr),
-		      "%q#D declared here", capture_init_expr);
+	      if (DECL_P (capture_init_expr))
+		inform (DECL_SOURCE_LOCATION (capture_init_expr),
+			"%q#D declared here", capture_init_expr);
 	      continue;
 	    }
 	  if (VAR_P (capture_init_expr)
Index: testsuite/g++.dg/cpp0x/lambda/lambda-ice17.C
===================================================================
--- testsuite/g++.dg/cpp0x/lambda/lambda-ice17.C	(revision 258235)
+++ testsuite/g++.dg/cpp0x/lambda/lambda-ice17.C	(working copy)
@@ -5,7 +5,7 @@ void foo (int);
 
 void foo (void)
 {
-  [&foo] // { dg-error "cannot capture" }
+  [&foo] // { dg-error "5:capture of non-variable" }
   {
     foo (0);
   };
Index: testsuite/g++.dg/cpp0x/lambda/lambda-ice23.C
===================================================================
--- testsuite/g++.dg/cpp0x/lambda/lambda-ice23.C	(revision 258235)
+++ testsuite/g++.dg/cpp0x/lambda/lambda-ice23.C	(working copy)
@@ -3,7 +3,7 @@
 
 template <typename T>
 constexpr int r(T x) {
-  auto f = [r,x]() { return r(x); }; // { dg-error "incomplete type" }
+  auto f = [r,x]() { return r(x); }; // { dg-error "13:capture of non-variable" }
   return 0;
 }
 
Index: testsuite/g++.dg/cpp0x/lambda/lambda-ice29.C
===================================================================
--- testsuite/g++.dg/cpp0x/lambda/lambda-ice29.C	(nonexistent)
+++ testsuite/g++.dg/cpp0x/lambda/lambda-ice29.C	(working copy)
@@ -0,0 +1,8 @@
+// PR c++/84618
+// { dg-do compile { target c++11 } }
+
+template <int>
+struct S {
+  void b() const;
+  void b() { [b] {}; } // { dg-error "15:capture of non-variable" }
+};

Re: [C++ Patch] PR 84618 ("[8 Regression] ICE in build_capture_proxy, at cp/lambda.c:460")

[Jason Merrill]

OK.

On Sun, Mar 4, 2018 at 8:55 PM, Paolo Carlini <paolo.carlini@oracle.com> wrote:
> Hi,
>
> a rather simple ice on invalid (not sure why only P4 given that no
> meaningful diagnostic is emitted before ICEing). What happens is that the
> ill-formed capture naming b, an OVERLOAD, escapes the early check in
> cp_parser_lambda_introducer to eventually trigger a recently added
> gcc_assert in build_capture_proxy. Adjusting the check means we also provide
> slightly clearer (IMO, matching clang, anyway) error messages for two
> testcases we already have.
>
> Tested x86_64-linux.
>
> Thanks, Paolo.
>
> ///////////////////////////////////////
>