[Bug c++/62227] Templated move not elided

[james.dennett at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62227

James Dennett james.dennett at gmail dot com changed:

   What|Removed |Added

 CC||james.dennett at gmail dot com

--- Comment #2 from James Dennett james.dennett at gmail dot com ---
 A template cannot be a move constructor

I'm with you so far.

 and so it is not legal to elide it.

I don't think that's the case (or at least, I'm not aware of any wording saying
so).  C++98 effectively said so, but C++03/11/14 don't, AFAICS.  Elision
applies to copying/moving, not just to copy/move constructors, and a template
can be used for copying or moving.

The standard isn't very clear on this though.

[Bug c++/58162] [C++11] bogus error: use of deleted function 'constexpr A::A(const A)'

[james.dennett at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58162

--- Comment #4 from James Dennett james.dennett at gmail dot com ---
Thanks, Jason.

[Bug c++/58162] [C++11] bogus error: use of deleted function 'constexpr A::A(const A)'

[james.dennett at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58162

James Dennett james.dennett at gmail dot com changed:

   What|Removed |Added

 CC||james.dennett at gmail dot com

--- Comment #1 from James Dennett james.dennett at gmail dot com ---
Jason pointed out to me that cv_unqualified() is more appropriate than
TYPE_MAIN_VARIANT(), and I see that we might want to do the

 if (!CLASS_TYPE_P (type)  cv_qualified_p (type))
type = cv_unqualified (type);

dance to avoid needless work for other types.

[Bug c++/43663] Can't take a const-ref to a bit field

[james.dennett at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43663

James Dennett james.dennett at gmail dot com changed:

   What|Removed |Added

 CC||james.dennett at gmail dot
   ||com

--- Comment #6 from James Dennett james.dennett at gmail dot com 2011-05-15 
11:52:01 UTC ---
Here's a quick hack that causes a temporary to be generated when binding a
bit-field to a reference-to-const.

$ svn diff
Index: call.c
===
--- call.c(revision 173769)
+++ call.c(working copy)
@@ -8594,7 +8594,7 @@
 expr = error_mark_node;
   else
 {
-  if (!lvalue_or_rvalue_with_address_p (expr))
+  if (is_bitfield_expr_with_lowered_type (expr) ||
!lvalue_or_rvalue_with_address_p (expr))
 {
   tree init;
   var = set_up_extended_ref_temp (decl, expr, cleanup, init);

I'll try to make time to clean that up and add regression tests before running
it by someone with stronger gcc-fu.

[Bug c++/43663] Can't take a const-ref to a bit field

[james.dennett at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43663

--- Comment #7 from James Dennett james.dennett at gmail dot com 2011-05-15 
11:55:47 UTC ---
Unsurprisingly the quick hack isn't really good enough -- it'll happily bind a
non-const reference to a temporary initialized from a bitfield.  (...and I
guess that's why we have tests, and code reviews.)

[Bug c++/43663] Can't take a const-ref to a bit field

[james.dennett at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43663

--- Comment #8 from James Dennett james.dennett at gmail dot com 2011-05-15 
12:34:51 UTC ---
Interestingly this works with Apple's g++ 4.2.1, specifically
i686-apple-darwin10-g++-4.2.1 (GCC) 4.2.1 (Apple Inc. build 5666) (dot 3), but
not with their 4.0.1 release.

Tested with:

int main() {
  struct S {
S(): i(0) {}
int i : 3;
  };
  S s;
  printf(s: %p\n, (void*)s);
  int const cr(s.i);  // should compile, binding to a temporary
  printf(cr: %p\n, (void*)cr);
}

A non-const reference still correctly gives an error:
st.cc:12: error: invalid initialization of reference of type ‘int’ from
expression of type ‘signed char:3’

[Bug c++/39415] static_cast used as downcast can silently lose const

[james.dennett at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39415

James Dennett james.dennett at gmail dot com changed:

   What|Removed |Added

 CC||james.dennett at gmail dot
   ||com

--- Comment #5 from James Dennett james.dennett at gmail dot com 2010-11-12 
04:10:16 UTC ---
Jason Merrill ja...@redhat.com has given me a heavy hint leading to the
following smaller and much less ugly patch for cp/typeck.c that makes it more
symmetrical with the reference case.  This also passes my simple tests.

.../gcc/cp $ svn diff
Index: typeck.c
===
--- typeck.c(revision 164712)
+++ typeck.c(working copy)
@@ -5999,7 +5999,8 @@
   base = lookup_base (TREE_TYPE (type), TREE_TYPE (intype),
   c_cast_p ? ba_unique : ba_check,
   NULL);
-  return build_base_path (MINUS_EXPR, expr, base, /*nonnull=*/false);
+  expr = build_base_path (MINUS_EXPR, expr, base, /*nonnull=*/false);
+  return cp_fold_convert(type, expr);
 }

   if ((TYPE_PTRMEM_P (type)  TYPE_PTRMEM_P (intype))