re PR fortran/87566 (ICE with class(*) and select)

2018-10-15  Paul Thomas  <pault@gcc.gnu.org>
    Tobias Burnus  <burnus@gcc.gnu.org>

PR fortran/87566
* resolve.c (resolve_assoc_var): Add missing array spec for
class associate names.
(resolve_select_type): Handle case where last typed component
of the selector has a different type to the expression.
* trans-expr.c (gfc_find_and_cut_at_last_class_ref): Replace
call to gfc_expr_to_initialize with call to gfc_copy_expr.
(gfc_conv_class_to_class): Guard assignment to 'len' field
against case where zero constant is supplied.

2018-10-15  Paul Thomas  <pault@gcc.gnu.org>
    Tobias Burnus  <burnus@gcc.gnu.org>

PR fortran/87566
* gfortran.dg/select_type_44.f90: New test.
* gfortran.dg/associate_42.f90: New test.

Co-Authored-By: Tobias Burnus <burnus@gcc.gnu.org>
From-SVN: r265171

re PR target/87572 (ICE in emit_move_insn, at expr.c:3722)

PR target/87572
* common/config/i386/i386-common.c (OPTION_MASK_ISA_AVX512F_UNSET):
Remove OPTION_MASK_ISA_AVX5124FMAPS_UNSET and
OPTION_MASK_ISA_AVX5124VNNIW_UNSET.

From-SVN: r265169

Adjust test to pass with latest glibc

Glibc changed the it_IT locales to use thousands separators,
invalidating this test. Use nl_NL instead, as Dutch only uses grouping
for money not numbers.

* testsuite/22_locale/numpunct/members/char/3.cc: Adjust test to
account for change to glibc it_IT localedata (glibc bz#10797).

From-SVN: r265165

PR libstdc++/87587 prevent -Wabi warnings

The warnings about changes to empty struct parameter passing can be
ignored because the callers are all internal to the library, and so
compiled with the same -fabi-version as the function definitions.

It would be preferable to use #pragma GCC diagnostic warning "-Wabi=12"
to get warnings about any other ABI changes in future versions, but
until PR c++/87611 is fixed the warnings must be completely disabled
with #pragma GCC diagnostic ignroed "-Wabi".

PR libstdc++/87587
* src/c++11/cxx11-shim_facets.cc: Suppress -Wabi warnings.

From-SVN: r265163

re PR tree-optimization/87022 (miscompilation with -ftree-loop-distribution)

PR tree-optimization/87022
* tree-loop-distribution.c (pg_add_dependence_edges): Check all
bits in dist vector rather than the first one.

gcc/testsuite
PR tree-optimization/87022
* gcc.dg/tree-ssa/pr87022.c: New test.

From-SVN: r265161

re PR middle-end/87610 (wrong-code with restrict)

2018-10-15  Richard Biener  <rguenther@suse.de>

PR middle-end/87610
* tree-ssa-structalias.c (struct vls_data): Add escaped_p member.
(visit_loadstore): When a used restrict tag escaped verify that
the points-to solution of "other" pointers do not include
escaped.
(compute_dependence_clique): If a used restrict tag escaped
communicated that down to visit_loadstore.

* gcc.dg/torture/restrict-6.c: New testcase.

From-SVN: r265160

S/390: Fix problem with vec_init expander

gcc/ChangeLog:

2018-10-15  Andreas Krebbel  <krebbel@linux.ibm.com>

* config/s390/s390.c (s390_expand_vec_init): Force vector element
into reg if it isn't a general operand.

gcc/testsuite/ChangeLog:

2018-10-15  Andreas Krebbel  <krebbel@linux.ibm.com>

* g++.dg/vec-init-1.C: New test.

From-SVN: r265158

vector (vector<>::cbegin()): Use C++11 direct initialization.

2018-10-15  François Dumont  <fdumont@gcc.gnu.org>

* include/debug/vector (vector<>::cbegin()): Use C++11 direct
initialization.
(vector<>::cend()): Likewise.
(vector<>::emplace(const_iterator, _Args&&...)): Likewise and use
consistent iterator comparison.
(vector<>::insert(const_iterator, size_type, const _Tp&)): Likewise.
(vector<>::insert(const_iterator, _InputIterator, _InputIterator)):
Likewise.
(vector<>::erase(const_iterator)): Likewise.
(vector<>::erase(const_iterator, const_iterator)): Likewise.

From-SVN: r265156

Daily bump.

From-SVN: r265155

i386: Add register source to movddup

Add register source to movddup so that IRA will allow register source
for *vec_dupv2di when SSE3 is enabled.

gcc/

PR target/87599
* config/i386/sse.md (*vec_dupv2di): Add register source to
movddup.

gcc/testsuite/

PR target/87599
* gcc.target/i386/pr87599.c: New test.

From-SVN: r265151

i386: Also disable AVX512IFMA/AVX5124FMAPS/AVX5124VNNIW

Also disable AVX512IFMA, AVX5124FMAPS and AVX5124VNNIW when disabling
AVX512F.

gcc/

PR target/87572
* common/config/i386/i386-common.c (OPTION_MASK_ISA_AVX512F_UNSET):
Add OPTION_MASK_ISA_AVX512IFMA_UNSET,
OPTION_MASK_ISA_AVX5124FMAPS_UNSET and
OPTION_MASK_ISA_AVX5124VNNIW_UNSET.

gcc/testsuite/

PR target/87572
* gcc.target/i386/pr87572.c: New test.

From-SVN: r265150

* MAINTAINERS: Update my email address.

From-SVN: r265149

re PR fortran/83146 (ICE on SELECT CASE statement with associate name)

2018-10-14  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/83146
* gfortran.dg/associate_43.f90: New test.

From-SVN: r265148

This patch adjusts one of the c-torture tests to account for the possible lack of divide-by-zero exceptions on certain moxie targets.

From-SVN: r265147

Daily bump.

From-SVN: r265146

dwarf2cfi.c (struct dw_trace_info): Add args_size_defined_for_eh.

* dwarf2cfi.c (struct dw_trace_info): Add args_size_defined_for_eh.
(notice_args_size): Set it in the current trace if no insn that can
throw internally has been seen yet.
(connect_traces): When connecting args_size between traces, allow the
incoming values not to match if there is an insn setting it before the
first insn that can throw internally; in that case, force the creation
of a CFI note on this latter insn.

From-SVN: r265142

Fix compilation failure with C++98 compilers

* opt-problem.h (opt_wrapper): Use template-argument-list when naming
the base class, because using the injected-class-name was not clearly
specified until DR 176.

From-SVN: r265140

Daily bump.

From-SVN: r265139

tighten the toplevel guard on ibm-ldouble.c

2018-10-12  Olivier Hainque  <hainque@adacore.com>

        * config/rs6000/ibm-ldouble.c: Augment the toplevel guard with
        defined (__FLOAT128_TYPE__) || defined (__LONG_DOUBLE_128__).

From-SVN: r265135

pdp11.md (doloop_end): New expander.

    * config/pdp11/pdp11.md (doloop_end): New expander.
    (doloop_end_insn): renamed from "doloop_end".
    (addqi3): New pattern.
    (subqi3): New pattern.
    * config/pdp11/predicates.md (incdec_operand): New predicate.

From-SVN: r265132

Add pattern to remove useless float casts in comparisons.

PR middle-end/81376

gcc/
* real.c (format_helper::can_represent_integral_type_p): New function
* real.h (format_helper::can_represent_integral_type_p): Ditto.
* match.pd: New pattern.

gcc/testsuite/
* c-c++-common/pr81376.c: New test.
* gcc.target/i386/387-ficom-2.c: Update test.
* gcc.target/i386/387-ficom-2.c: Ditto.

From-SVN: r265131

introduce --enable-large-address-aware

Add a configure knob for mingw32 and 64 toolchains to default passing
--large-address-aware to the linker, when creating 32-bit binaries.
-Wl,--disable-large-address-aware can still reverse its effects.

for  gcc/ChangeLog

* configure.ac: Introduce --enable-large-address-aware
to define MINGW_DEFAULT_LARGE_ADDR_AWARE.
* doc/install.texi: Document it.
* configure, config.in: Rebuilt.
* config/i386/mingw32.h (LINK_SPEC_LARGE_ADDR_AWARE): Define,
based on MINGW_DEFAULT_LARGE_ADDR_AWARE.
(LINK_SPEC): Insert it.
* config/i386/mingw-264.h: Likewise.

From-SVN: r265130

define HAVE_GAS_ALIGNED_COMM

HAVE_GAS_ALIGNED_COMM is referenced in an initializer in cygming.opt,
but it's not guaranteed to be defined by configure, so define it to
zero in a cygming-specific header if it's not defined.

for  gcc/ChangeLog

* cygming.h (HAVE_GAS_ALIGNED_COMM): Fallback-define.

From-SVN: r265129

[C++ PATCH] more TU parsing refactoring

https://gcc.gnu.org/ml/gcc-patches/2018-10/msg00788.html
* parser.h (struct cp_parser): Drop implicit_extern_c.
* parser.c (cp_debug_parser): Drop implicit_extern_c.
(cp_parser_new): Likewise.
(cp_parser_translation_unit): Handle implicit extern c here.  Call
cp_parser_toplevel_declaration.
(cp_parser_toplevel_declaration): New, broken out of ...
(cp_parser_declaration_seq_opt): ... here.  Call it.  Drop
implicit extern C handling.

From-SVN: r265127

Fix off-by-one issue with inline matmul

        PR fortran/87597
        * expr.c (gfc_simplify_expr): Avoid simplifying
        the 'array' argument to lbound/ubound/lcobound/
        ucobound.

        PR fortran/87597
        * gfortran.dg/inline_matmul_24.f90: New.

From-SVN: r265126

Fix error-recovery ICE in check_proc_interface

        PR fortran/58787
        * decl.c (get_proc_name): Return with error before
        creating sym_tree.

        PR fortran/58787
        * gfortran.dg/goacc/pr77765.f90: Modify dg-error.
        * gfortran.dg/interface_42.f90: Ditto.
        * gfortran.dg/internal_references_1.f90: Ditto.
        * gfortran.dg/invalid_procedure_name.f90: Ditto.
        * gfortran.dg/pr65453.f90: Ditto.
        * gfortran.dg/pr77414.f90: Ditto.
        * gfortran.dg/pr78741.f90: Ditto.
        * gfortran.dg/same_name_2.f90: Ditto.

From-SVN: r265125

re PR rtl-optimization/87600 (Fix for PRs 86939 and 87479 causes build issues for several targets)

PR rtl-optimization/87600
* ira-lives (non_conflicting_reg_copy_p): Disable for non LRA targets.

From-SVN: r265113

md.texi (doloop_end): Document that the pattern code may need to check operand mode.

* doc/md.texi (doloop_end): Document that the pattern code may
need to check operand mode.

From-SVN: r265102

Initial commit of Networking TS implementation

* include/Makefile.am: Add new headers.
* include/Makefile.in: Regenerate.
* include/experimental/bits/net.h: New header for common
implementation details of Networking TS.
* include/experimental/buffer: New header.
* include/experimental/executor: New header.
* include/experimental/internet: New header.
* include/experimental/io_context: New header.
* include/experimental/net: New header.
* include/experimental/netfwd: New header.
* include/experimental/socket: New header.
* include/experimental/timer: New header.
* testsuite/experimental/net/buffer/arithmetic.cc: New test.
* testsuite/experimental/net/buffer/const.cc: New test.
* testsuite/experimental/net/buffer/creation.cc: New test.
* testsuite/experimental/net/buffer/mutable.cc: New test.
* testsuite/experimental/net/buffer/size.cc: New test.
* testsuite/experimental/net/buffer/traits.cc: New test.
* testsuite/experimental/net/execution_context/use_service.cc: New
test.
* testsuite/experimental/net/headers.cc: New test.
* testsuite/experimental/net/internet/address/v4/comparisons.cc: New
test.
* testsuite/experimental/net/internet/address/v4/cons.cc: New test.
* testsuite/experimental/net/internet/address/v4/creation.cc: New
test.
* testsuite/experimental/net/internet/address/v4/members.cc: New
test.
* testsuite/experimental/net/internet/resolver/base.cc: New test.
* testsuite/experimental/net/internet/resolver/ops/lookup.cc: New
test.
* testsuite/experimental/net/internet/resolver/ops/reverse.cc: New
test.
* testsuite/experimental/net/timer/waitable/cons.cc: New test.
* testsuite/experimental/net/timer/waitable/dest.cc: New test.
* testsuite/experimental/net/timer/waitable/ops.cc: New test.

From-SVN: r265080

[AArch64] Support zero-extended move to FP register

The popcount expansion uses SIMD instructions acting on 64-bit values.
As a result a popcount of a 32-bit integer requires zero-extension before
moving the zero-extended value into an FP register.  This patch adds
support for zero-extended int->FP moves to avoid the redundant uxtw.
Similarly, add support for 32-bit zero-extending load->FP register
and 32-bit zero-extending FP->FP and FP->int moves.
Add a missing 'fp' arch attribute to the related 8/16-bit pattern and
fix an incorrect type attribute.

To complete zero-extended load support, add a new alternative to
load_pair_zero_extendsidi2_aarch64 to support LDP into FP registers too.

int f (int a)
{
  return __builtin_popcount (a);
}

Before:
uxtw x0, w0
fmov d0, x0
cnt v0.8b, v0.8b
addv b0, v0.8b
fmov w0, s0
ret

After:
fmov s0, w0
cnt v0.8b, v0.8b
addv b0, v0.8b
fmov w0, s0
ret

Passes regress & bootstrap on AArch64.

    gcc/
* config/aarch64/aarch64.md (zero_extendsidi2_aarch64): Add alternatives
to zero-extend between int and floating-point registers.
(load_pair_zero_extendsidi2_aarch64): Add alternative for zero-extended
ldp into floating-point registers.  Add type and arch attributes.
(zero_extend<SHORT:mode><GPI:mode>2_aarch64): Add arch attribute.
Use f_loads for type attribute.

    testsuite/
* gcc.target/aarch64/popcnt.c: Test zero-extended popcount.
* gcc.target/aarch64/vec_zeroextend.c: Test zero-extended vectors.

From-SVN: r265079

Fix typo in previous commit.

2018-10-12  Christophe Lyon  <christophe.lyon@linaro.org>

PR c++/87364
* c-pretty-print.c (c_pretty_printer::constant): Fix typo.

From-SVN: r265078

re PR c++/87364 (Pretty print of enumerator never prints the id, always falls back to C-style cast output)

PR c++/87364
* c-pretty-print.h (pp_c_type_cast): Prototype.
(pp_c_integer_constant): Likewise.
* c-pretty-print.c (pp_c_type_cast): No longer static.
(pp_c_integer_constant): Likewise.
(pp_c_enumeration_constant): Fix loop termination when finding
name of constant.  No longer returns a value.  Call
pp_c_integer_constant.
(c_pretty_printer::constant): Update for changes to
pp_c_enumeration_constant.

PR c++/87364
* cxx-pretty-print.c (pp_cxx_enumeration_constant): New function.
(cxx_pretty_printer::constant): Use it.

From-SVN: r265077

extend.texi (attribute packed): Correct typos.

gcc/ChangeLog:

* doc/extend.texi (attribute packed): Correct typos.

From-SVN: r265076

extend.texi (attribute flatten): Mention interaction with noinline.

gcc/ChangeLog:

* doc/extend.texi (attribute flatten): Mention interaction with
noinline.

From-SVN: r265075

re PR target/87156 (ICE building libstdc++ for mips64)

PR target/87156
* cgraphclones.c (cgraph_node::create_version_clone_with_body):
Set new_decl virtual flag to zero.

From-SVN: r265074

PR middle-end/87593 - conflicting format_arg attributes on a declaration accepted

gcc/ChangeLog:

* doc/extend.texi (attribute format_arg): Discuss using multiple
attributes on a single function.

From-SVN: r265073

Daily bump.

From-SVN: r265072

PR libstdc++/77691 increase allocation size to at least alignment

It's not safe to assume that malloc(n) returns memory aligned to more
than n, so when relying on the guaranteed alignment of malloc ensure
that the number of bytes allocated is at least as large as the
alignment.

PR libstdc++/77691
* include/experimental/memory_resource (__resource_adaptor_imp): Do
not allocate sizes smaller than alignment when relying on guaranteed
alignment.
* testsuite/experimental/memory_resource/new_delete_resource.cc:
Adjust expected number of bytes allocated for alignof(max_align_t).

From-SVN: r265068

fold-vec-insert-char-p8.c: New.

[testsuite]

2018-10-10  Will Schmidt  <will_schmidt@vnet.ibm.com>

* gcc.target/powerpc/fold-vec-insert-char-p8.c: New.
* gcc.target/powerpc/fold-vec-insert-char-p9.c: New.
* gcc.target/powerpc/fold-vec-insert-double.c: New.
* gcc.target/powerpc/fold-vec-insert-float-p8.c: New.
* gcc.target/powerpc/fold-vec-insert-float-p9.c: New.
* gcc.target/powerpc/fold-vec-insert-int-p8.c: New.
* gcc.target/powerpc/fold-vec-insert-int-p9.c: New.
* gcc.target/powerpc/fold-vec-insert-longlong.c: New.
* gcc.target/powerpc/fold-vec-insert-short-p8.c: New.
* gcc.target/powerpc/fold-vec-insert-short-p9.c: New.

From-SVN: r265067

fold-vec-extract-char.p7.c: New.

[testsuite]
* gcc.target/powerpc/fold-vec-extract-char.p7.c: New.
* gcc.target/powerpc/fold-vec-extract-char.p8.c: New.
* gcc.target/powerpc/fold-vec-extract-char.p9.c: New.
* gcc.target/powerpc/fold-vec-extract-double.p7.c: New.
* gcc.target/powerpc/fold-vec-extract-double.p8.c: New.
* gcc.target/powerpc/fold-vec-extract-double.p9.c: New.
* gcc.target/powerpc/fold-vec-extract-float.p7.c: New.
* gcc.target/powerpc/fold-vec-extract-float.p8.c: New.
* gcc.target/powerpc/fold-vec-extract-float.p9.c: New.
* gcc.target/powerpc/fold-vec-extract-int.p7.c: New.
* gcc.target/powerpc/fold-vec-extract-int.p8.c: New.
* gcc.target/powerpc/fold-vec-extract-int.p9.c: New.
* gcc.target/powerpc/fold-vec-extract-longlong.p7.c: New.
* gcc.target/powerpc/fold-vec-extract-longlong.p8.c: New.
* gcc.target/powerpc/fold-vec-extract-longlong.p9.c: New.
* gcc.target/powerpc/fold-vec-extract-short.p7.c: New.
* gcc.target/powerpc/fold-vec-extract-short.p8.c: New.
* gcc.target/powerpc/fold-vec-extract-short.p9.c: New.

From-SVN: r265066

fold-vec-select-char.c: New.

[testsuite]

2018-10-11  Will Schmidt  <will_schmidt@vnet.ibm.com>

* gcc.target/powerpc/fold-vec-select-char.c: New.
* gcc.target/powerpc/fold-vec-select-double.c: New.
* gcc.target/powerpc/fold-vec-select-float.c: New.
* gcc.target/powerpc/fold-vec-select-int.c: New.
* gcc.target/powerpc/fold-vec-select-longlong.c: New.
* gcc.target/powerpc/fold-vec-select-short.c: New.

From-SVN: r265065

re PR tree-optimization/86829 (Missing sin(atan(x)) and cos(atan(x)) optimizations)

PR tree-optimization/86829
* match.pd (sin (atan (x))): New simplification rules.
(cos (atan (x))): Likewise.
* real.c (build_sinatan_real): New function.
* real.h (build_sinatan_real): Prototype.

PR tree-optimization/86829
* gcc.dg/sinatan-1.c: New test.
* gcc.dg/sinatan-2.c: New test.
* gcc.dg/sinatan-3.c: New test.

From-SVN: r265064

rs6000.c (map_to_integral_tree_type): New helper function.

[gcc]

2018-10-09  Will Schmidt <will_schmidt@vnet.ibm.com>

* config/rs6000/rs6000.c (map_to_integral_tree_type): New helper
function.
(fold_mergeeo_helper): New helper function.
(rs6000_gimple_fold_builtin): Add hooks for vec_mergee and vec_mergeo
intrinsics.  Correct some whitespace indentation issues.

From-SVN: r265063

fold-vec-mergeeo-floatdouble.c: New.

[testsuite]

2018-10-09  Will Schmidt  <will_schmidt@vnet.ibm.com>

* gcc.target/powerpc/fold-vec-mergeeo-floatdouble.c: New.
* gcc.target/powerpc/fold-vec-mergeeo-int.c: New.
* gcc.target/powerpc/fold-vec-mergeeo-longlong.c: New.

From-SVN: r265062

2018-10-11  François Dumont  <fdumont@gcc.gnu.org>

* include/debug/forward_list
(forward_list<>::before_begin()): Use C++11 direct initialization.
(forward_list<>::begin()): Likewise.
(forward_list<>::end()): Likewise.
(forward_list<>::cbefore_begin()): Likewise.
(forward_list<>::cbegin()): Likewise.
(forward_list<>::cend()): Likewise.
(forward_list<>::emplace_after<>(const_iterator, _Args&&...)): Likewise.
(forward_list<>::insert_after(const_iterator, const _Tp&)): Likewise.
(forward_list<>::insert_after(const_iterator, _Tp&&)): Likewise.
(forward_list<>::insert_after(const_iterator, size_type, const _Tp&)):
Likewise.
(forward_list<>::insert_after(const_iterator, initializer_list<>)):
Likewise.
(forward_list<>::erase_after(const_iterator)): Likewise.
(forward_list<>::erase_after(const_iterator, const_iterator)): Likewise
and ensure consistent iterator comparison.

From-SVN: r265061

2018-10-11  François Dumont  <fdumont@gcc.gnu.org>

* include/bits/forward_list.h
(_Fwd_list_iterator<>::operator==): Replace member function with inline
friend.
(_Fwd_list_iterator<>::operator!=): Likewise.
(_Fwd_list_const_iterator<>::operator==): Likewise.
(_Fwd_list_const_iterator<>::operator!=): Likewise.
(operator==(const _Fwd_list_iterator<>&,
const _Fwd_list_const_iterator<>&)): Remove.
(operator!=(const _Fwd_list_iterator<>&,
const _Fwd_list_const_iterator<>&)): Remove.
(forward_list<>::_Node): Take typedef from base type.
(forward_list<>::iterator): Likewise.
(forward_list<>::const_iterator): Likewise.

From-SVN: r265060

revert: re PR fortran/83522 (ICE on allocatable string reference, string(:)(:))

2018-10-11  Tobias Burnus <burnus@net-b.de>

        Revert:
        2018-10-09  Tobias Burnus <burnus@net-b.de>

        PR fortran/83522
        * resolve.c (resolve_ref): Reject nonscalar
        substring references.

2018-10-11  Tobias Burnus <burnus@net-b.de>

        Revert:
        2018-10-09  Tobias Burnus <burnus@net-b.de>

        PR fortran/83522
        * gfortran.dg/actual_array_substr_1.f90: Add dg-error,
        change to dg-do compile.
        * gfortran.dg/actual_array_substr_2.f90: Ditto.
        * gfortran.dg/array_initializer_1.f90: Use array
        element not size-one section.
        * gfortran.dg/array_substring.f90: New.

From-SVN: r265059

[AArch64] Fix PR87511

As mentioned in PR87511, the shift used in aarch64_mask_and_shift_for_ubfiz_p
should be evaluated as a HOST_WIDE_INT rather than int.

Passes bootstrap & regress.

    gcc/
PR target/87511
* config/aarch64/aarch64.c (aarch64_mask_and_shift_for_ubfiz_p):
Use HOST_WIDE_INT_1U for shift.

    testsuite/
PR target/87511
* gcc.target/aarch64/pr87511.c: Add new test.

From-SVN: r265058

C++: suggestions for misspelled private members (PR c++/84993)

PR c++/84993 identifies a problem with our suggestions for
misspelled member names in the C++ FE for the case where the
member is private.

For example, given:

class foo
{
public:
  double get_ratio() const { return m_ratio; }

private:
  double m_ratio;
};

void test(foo *ptr)
{
  if (ptr->ratio >= 0.5)
    ;// etc
}

...we currently emit this suggestion:

<source>: In function 'void test(foo*)':
<source>:12:12: error: 'class foo' has no member named 'ratio'; did you mean 'm_ratio'?
   if (ptr->ratio >= 0.5)
            ^~~~~
            m_ratio

...but if the user follows this suggestion, they get:

<source>: In function 'void test(foo*)':
<source>:12:12: error: 'double foo::m_ratio' is private within this context
   if (ptr->m_ratio >= 0.5)
            ^~~~~~~
<source>:7:10: note: declared private here
   double m_ratio;
          ^~~~~~~
<source>:12:12: note: field 'double foo::m_ratio' can be accessed via 'double foo::get_ratio() const'
   if (ptr->m_ratio >= 0.5)
            ^~~~~~~
            get_ratio()

It feels wrong to be emitting a fix-it hint that doesn't compile, so this
patch adds the accessor fix-it hint logic to this case, so that we directly
offer a valid suggestion:

<source>: In function 'void test(foo*)':
<source>:12:12: error: 'class foo' has no member named 'ratio'; did you mean
'double foo::m_ratio'? (accessible via 'double foo::get_ratio() const')
   if (ptr->ratio >= 0.5)
            ^~~~~
            get_ratio()

gcc/cp/ChangeLog:
PR c++/84993
* call.c (enforce_access): Move diagnostics to...
(complain_about_access): ...this new function.
* cp-tree.h (class access_failure_info): Rename split out field
"m_field_decl" into "m_decl" and "m_diag_decl".
(access_failure_info::record_access_failure): Add tree param.
(access_failure_info::was_inaccessible_p): New accessor.
(access_failure_info::get_decl): New accessor.
(access_failure_info::get_diag_decl): New accessor.
(access_failure_info::get_any_accessor): New member function.
(access_failure_info::add_fixit_hint): New static member function.
(complain_about_access): New decl.
* typeck.c (access_failure_info::record_access_failure): Update
for change to fields.
(access_failure_info::maybe_suggest_accessor): Split out into...
(access_failure_info::get_any_accessor): ...this new function...
(access_failure_info::add_fixit_hint): ...and this new function.
(finish_class_member_access_expr): Split out "has no member named"
error-handling into...
(complain_about_unrecognized_member): ...this new function, and
check that the guessed name is accessible along the access path.
Only provide a spell-correction fix-it hint if it is; otherwise,
attempt to issue an accessor fix-it hint.

gcc/testsuite/ChangeLog:
PR c++/84993
* g++.dg/torture/accessor-fixits-9.C: New test.

From-SVN: r265056

[C++ PATCH] parser simplification

https://gcc.gnu.org/ml/gcc-patches/2018-10/msg00689.html
cp/
* parser.c (cp_parser_translation_unit): Return void.  Don't fail
at first extra }, simplify logic.
(c_parse_file): Call finish_translation_unit here.

testsuite/
* g++.dg/parse/close-brace.C: New.
* g++.dg/cpp0x/noexcept16.C: Avoid warning.
* g++.old-deja/g++.other/crash32.C: Add another error

From-SVN: r265055

leverage linker relaxation on ppc vxworks RTPs

2018-10-11  Doug Rupp  <rupp@adacore.com>
            Olivier Hainque  <hainque@adacore.com>

        * config/rs6000/vxworks.h (VXWORKS_RELAX_LINK_SPEC): New macro.
        Pass --relax to the linker for RTPs.
        (LINK_SPEC): Append VXWORKS_RELAX_LINK_SPEC.

Co-Authored-By: Olivier Hainque <hainque@adacore.com>
From-SVN: r265054

re PR c++/85070 (ICE on C++ code: in lazily_declare_fn, at cp/method.c:2409)

PR c++/85070
* g++.dg/cpp0x/pr85070.C: Change effective target for diagnostics from
c++14_only to c++14.

From-SVN: r265052

multiline.exp: complain about mismatched dg-{begin|end}-multiline-output

Mismatched dg-{begin|end}-multiline-output directives are currently
silently ignored, leading to difficult-to-diagnose test failures
involving excess output.

This patch makes multiline.exp complain about them.

gcc/testsuite/ChangeLog:
* lib/multiline.exp (dg-begin-multiline-output): Issue an error if
there hasn't been a dg-end-multiline-output since the last
dg-begin-multiline-output.
(dg-end-multiline-output): Issue an error if there hasn't been a
dg-begin-multiline-output.  Reset _multiline_last_beginning_line
as soon possible.  Rename "line" to "last_line".

From-SVN: r265046

PR libstdc++/80538 Only call sleep for non-zero values

Avoid a system call when no sleep is required. Sleep in a loop (actually
two loops) to handle interruption by signals.

PR libstdc++/80538
* src/c++11/thread.cc (this_thread::__sleep_for)
[_GLIBCXX_HAVE_SLEEP]: Only call sleep for non-zero values.
Loop while sleep call is interrupted and until steady_clock
shows requested duration has elapsed.
(!_GLIBCXX_HAVE_USLEEP]: Use the _GLIBCXX_HAVE_SLEEP code path, but
avoiding the usleep call.
* testsuite/30_threads/this_thread/60421.cc: Test repeated
signal interruptions.

From-SVN: r265044

Elide repeated RTL elements.

GCN's 64-lane vectors tend to make RTL dumps very long.  This patch makes them
far more bearable by eliding long sequences of the same element into "repeated"
messages.

This also takes care of reading repeated sequences in the RTL front-end.

There are self tests for both reading and writing.

2018-10-11  Andrew Stubbs  <ams@codesourcery.com>
    Jan Hubicka  <jh@suse.cz>
    Martin Jambor  <mjambor@suse.cz>

gcc/
* print-rtl.c (print_rtx_operand_codes_E_and_V): Print how many times
the same elements are repeated rather than printing all of them.
* read-rtl.c (rtx_reader::read_rtx_operand): Recognize and expand
"repeated" elements.
* read-rtl-function.c (test_loading_repeat): New function.
(read_rtl_function_c_tests): Call test_loading_repeat.
* rtl-tests.c (test_dumping_repeat): New function.
(rtl_tests_c_tests): Call test_dumping_repeat.

gcc/testsuite/
* selftests/repeat.rtl: New file.

Co-Authored-By: Jan Hubicka <jh@suse.cz>
Co-Authored-By: Martin Jambor <mjambor@suse.cz>
From-SVN: r265042

re PR c++/87582 (Returning a reference to a data member via structured bindings dangles)

PR c++/87582
* typeck.c (maybe_warn_about_returning_address_of_local): If
whats_returned is a structured binding identifier and the structured
binding is a reference, recurse on its initializer.

* g++.dg/cpp1z/decomp48.C: New test.

From-SVN: r265041

libcpp: show macro definition when used with wrong argument count

Consider:

demo.c: In function 'test':
demo.c:5:40: error: macro "LOG_2" requires 3 arguments, but only 2 given
5 |   LOG_2 ("loading file: %s\n", filename);
  |                                        ^

This patch adds a note showing the definition of the macro in
question, giving:

demo.c: In function 'test':
demo.c:5:40: error: macro "LOG_2" requires 3 arguments, but only 2 given
5 |   LOG_2 ("loading file: %s\n", filename);
  |                                        ^
In file included from demo.c:1:
logging.h:1: note: macro "LOG_2" defined here
1 | #define LOG_2(FMT, ARG0, ARG1) do { fprintf (stderr, (FMT), (ARG0), (ARG1)); }
  |

gcc/testsuite/ChangeLog:
* g++.dg/diagnostic/macro-arg-count.C: Move to...
* c-c++-common/cpp/macro-arg-count-1.c: ...here, generalizing
output for C vs C++.  Expect notes showing the definitions of the
macros.
* c-c++-common/cpp/macro-arg-count-2.c: New test, adapted from the
above.

libcpp/ChangeLog:
* macro.c (_cpp_arguments_ok): If the argument count is wrong, add
a note showing the definition of the macro.

From-SVN: r265040

x86-tune-costs.h (bdver?_memcpy, [...]): Unify to ...

2018-10-11  Richard Biener  <rguenther@suse.de>

* config/i386/x86-tune-costs.h (bdver?_memcpy, bdver?_memset,
bdver?_cost): Unify to ...
(bdver_memcpy, bdver_memset, bdver_cost): ... this.
* config/i386/i386.c (processor_cost_table): Adjust.

From-SVN: r265038

[PATCH] A couple of line map fixes

https://gcc.gnu.org/ml/gcc-patches/2018-10/msg00623.html
* include/line-map.h (LINEMAPS_MACRO_LOWEST_LOCATION): Fix
off-by-one error.
* line-map.c (linemap_enter_macro): Use RAII.  Clear all of the
macro_locations.

From-SVN: r265037

Make ext allocators support heterogeneous equality comparison

The Allocator requirements include the ability to compare different
specializations of the same allocator class template. This did not work
for __gnu_cxx::new_allocator and other extension allocators.  This patch
replaces the equality operators for those allocators with inline friends
that support heterogeneous comparisons.  (I'm not changing all ext
allocators because some are bit-rotted already).

Additionally, the equality operators for comparing two std::allocator
objects of the same type are now defined as inline friends. Those
overloads don't need to be declared at namespace scope, because they
aren't specified in the standard (but they're needed in this
implementation to avoid ambiguities caused by the extra overloads
defined for the base allocator type).

* include/bits/allocator.h
(operator==(const allocator<_Tp>&, const allocator<_Tp>))
(operator!=(const allocator<_Tp>&, const allocator<_Tp>)): Replace
with inline friends.
* include/ext/debug_allocator.h (operator==, operator!=): Replace
with inline friend functions that compare to rebound allocators.
* include/ext/malloc_allocator.h (operator==, operator!=): Likewise.
* include/ext/new_allocator.h (operator==, operator!=): Likewise.
* testsuite/ext/debug_allocator/eq.cc: New test.
* testsuite/ext/ext_pointer/alloc_eq.cc: New test.
* testsuite/ext/malloc_allocator/eq.cc: New test.
* testsuite/ext/new_allocator/eq.cc: New test.

From-SVN: r265036

[testsuite] Further fixes to warn-abs-1.c

2018-10-11  Martin Jambor  <mjambor@suse.cz>

testsuite/
* gcc.dg/warn-abs-1.c: Guard tests assuming size of long double is
greater that the size of double by target large_long double.

From-SVN: r265035

c-lex.c (c_common_has_attribute): Return 201803 instead of 20180312 for no_unique_address.

* c-lex.c (c_common_has_attribute): Return 201803 instead of 20180312
for no_unique_address.

* g++.dg/cpp2a/feat-cxx2a.C: New test.

From-SVN: r265034

re PR c++/87547 (G++ reports bad type names for bit-field members)

PR c++/87547
* rtti.c (get_tinfo_decl_dynamic): Use unlowered_expr_type instead
of TREE_TYPE.

* g++.dg/rtti/typeid12.C: New test.

From-SVN: r265033

Daily bump.

From-SVN: r265032

re PR middle-end/87574 (ICE in add_data_member_location_attribute at gcc/gcc/dwarf2out.c:19226 since r264943)

PR middle-end/87574
* cgraphunit.c (cgraph_node::expand_thunk): Force DECL_IGNORED_P on
the thunk when expanding to GIMPLE.

From-SVN: r265028

PR c++/87567 - constexpr rejects call to non-constexpr function.

* constexpr.c (potential_constant_expression_1) <case FOR_STMT>: Return
true if the condition is always false.
<case WHILE_STMT>: Likewise.

* g++.dg/cpp1y/constexpr-loop7.C: New test.

From-SVN: r265027

Fat-fingered my recent patch adding the SSE3 testcases for powerpc,

most likely by twice applying the patch which added the testcases.

This patch removes the duplicated code.

[gcc/testsuite]

2018-10-10  Paul A. Clarke  <pc@us.ibm.com>

PR target/87579
* gcc.target/powerpc/sse3-check.h: Remove duplicated code.
* gcc.target/powerpc/sse3-addsubps.c: Likewise.
* gcc.target/powerpc/sse3-addsubpd.c: Likewise.
* gcc.target/powerpc/sse3-haddps.c: Likewise.
* gcc.target/powerpc/sse3-hsubps.c: Likewise.
* gcc.target/powerpc/sse3-haddpd.c: Likewise.
* gcc.target/powerpc/sse3-hsubpd.c: Likewise.
* gcc.target/powerpc/sse3-lddqu.c: Likewise.
* gcc.target/powerpc/sse3-movsldup.c: Likewise.
* gcc.target/powerpc/sse3-movshdup.c: Likewise.
* gcc.target/powerpc/sse3-movddup.c: Likewise.
* gcc.target/powerpc/pr37191.c: Likewise.

From-SVN: r265026

varasm.c (mergeable_string_section): Don't try to move zero-length strings to the merge section.

2018-10-10  Bernd Edlinger  <bernd.edlinger@hotmail.de>

        * varasm.c (mergeable_string_section): Don't try to move zero-length
        strings to the merge section.

From-SVN: r265025

PR c/54391 - transparent_union typedef'ing inconsistent

gcc/testsuite/ChangeLog:
* gcc.dg/transparent-union-6.c: New.

From-SVN: r265024

PR libstdc++/87544 limit max_size() to PTRDIFF_MAX / sizeof(T)

The C++17 standard requires the default implementation for
allocator_traits::max_size to return SIZE_MAX / sizeof(value_type).
That causes GCC to warn because the value could be larger than can
sensibly be passed to malloc. This patch changes the new_allocator and
malloc_allocator max_size() members to use PTRDIFF_MAX instead of
SIZE_MAX (and because they define it, the allocator_traits default isn't
used). This also changes vector::max_size to impose a sensible limit
using PTRDIFF_MAX for cases where the value from the allocator or
allocator_traits is not sensible.

PR libstdc++/87544
* include/bits/stl_vector.h (vector::_S_max_size): Limit size to
PTRDIFF_MAX / sizeof(value_type).
* include/ext/malloc_allocator.h (malloc_allocator::max_size):
Likewise.
* include/ext/new_allocator.h (new_allocator::max_size): Likewise.
* testsuite/23_containers/vector/allocator/minimal.cc: Adjust
expected value for max_size().
* testsuite/23_containers/vector/capacity/87544.cc: New test.

From-SVN: r265021

memchr-1.c: Avoid assuming 4-byte wchar_t.

gcc/testsuite/ChangeLog:

* gcc.c-torture/execute/memchr-1.c: Avoid assuming 4-byte wchar_t.
Add a test for 2-byte wchar_t.
* gcc.dg/builtin-memchr.c: New test.

From-SVN: r265020

re PR target/87573 (error: could not split insn since r264877)

PR target/87573
* config/i386/mmx.md (const_vector 0 -> mem splitter): New splitter.

testsuite/ChangeLog:

PR target/87573
* gcc.target/i386/pr87573.c: New test.

From-SVN: r265019

re PR target/87550 (Intrinsics for rdpmc (__rdpmc, __builtin_ia32_rdpmc) are interpreted as pure functions)

PR target/87550
* config/i386/i386-builtin.def (IX86_BUILTIN_RDPMC): Move from args set
to special_args set.

* gcc.target/i386/pr87550.c: New test.

From-SVN: r265007

re PR c/87286 (ICE on vectors of enums)

PR c/87286
* gcc.dg/pr87286.c: Add -Wno-psabi to dg-options.

From-SVN: r265006

sse.md (reduc_plus_scal_v8df, [...]): Merge into pattern reducing to half width and recursing and pattern terminating...

2018-10-10  Richard Biener  <rguenther@suse.de>

* config/i386/sse.md (reduc_plus_scal_v8df, reduc_plus_scal_v4df,
reduc_plus_scal_v2df, reduc_plus_scal_v16sf, reduc_plus_scal_v8sf,
reduc_plus_scal_v4sf): Merge into pattern reducing to half width
and recursing and pattern terminating the recursion on SSE
vector width using ix86_expand_reduc.
(reduc_sminmax_scal_<mode>): Split into part reducing to half
width and recursing and SSE2 vector variant doing the final
reduction with ix86_expand_reduc.
(reduc_uminmax_scal_<mode>): Likewise for the AVX512 variants
with terminating the recursion at AVX level, splitting that
to SSE there.

From-SVN: r265004

Daily bump.

From-SVN: r265003

Cleanup of libcpp diagnostic callbacks

This patch renames the "error" callback within libcpp
to "diagnostic", and uses the pair of enums in cpplib.h, rather
than passing two different kinds of "int" around.

gcc/c-family/ChangeLog:
* c-common.c (c_option_controlling_cpp_error): Rename to...
(c_option_controlling_cpp_diagnostic): ...this, and convert
"reason" from int to enum.
(c_cpp_error): Rename to...
(c_cpp_diagnostic): ...this, converting level and reason to enums.
* c-common.h (c_cpp_error): Rename to...
(c_cpp_diagnostic): ...this, converting level and reason to enums.
* c-opts.c (c_common_init_options): Update for renaming.

gcc/fortran/ChangeLog:
* cpp.c (gfc_cpp_init_0): Update for renamings.
(cb_cpp_error): Rename to...
(cb_cpp_diagnostic): ...this, converting level and reason to
enums.

gcc/ChangeLog:
* genmatch.c (error_cb): Rename to...
(diagnostic_cb): ...this, converting int params to enums.
(fatal_at): Update for renaming.
(warning_at): Likewise.
(main): Likewise.
* input.c (selftest::ebcdic_execution_charset::apply):
Update for renaming of...
(selftest::ebcdic_execution_charset::on_error): ...this, renaming
to...
(selftest::ebcdic_execution_charset::on_diagnostic): ...this,
converting level and reason to enums.
(class selftest::lexer_error_sink): Rename to...
(class selftest::lexer_test_options): ...this, renaming field
"m_errors" to "m_diagnostics".
(selftest::lexer_test_options::apply): Update for renaming of...
(selftest::lexer_test_options::on_error): ...this, renaming to...
(selftest::lexer_test_options::on_diagnostic): ...this
converting level and reason to enums.
(selftest::test_lexer_string_locations_raw_string_unterminated):
Update for renamings.
* opth-gen.awk (struct cpp_reason_option_codes_t): Use enum for
"reason".

libcpp/ChangeLog:
* charset.c (noop_error_cb): Rename to...
(noop_diagnostic_cb): ...this, converting params to enums.
(cpp_interpret_string_ranges): Update for renaming and enums.
* directives.c (check_eol_1): Convert reason to enum.
(do_diagnostic): Convert code and reason to enum.
(do_error): Use CPP_W_NONE rather than 0.
(do_pragma_dependency): Likewise.
* errors.c (cpp_diagnostic_at): Convert level and reason to enums.
Update for renaming.
(cpp_diagnostic): Convert level and reason to enums.
(cpp_error): Convert level to enum.
(cpp_warning): Convert reason to enums.
(cpp_pedwarning): Likewise.
(cpp_warning_syshdr): Likewise.
(cpp_diagnostic_with_line): Convert level and reason to enums.
Update for renaming.
(cpp_error_with_line): Convert level to enum.
(cpp_warning_with_line): Convert reason to enums.
(cpp_pedwarning_with_line): Likewise.
(cpp_warning_with_line_syshdr): Likewise.
(cpp_error_at): Convert level to enum.
(cpp_errno): Likewise.
(cpp_errno_filename): Likewise.
* include/cpplib.h (enum cpp_diagnostic_level): Name this enum,
and move to before struct cpp_callbacks.
(enum cpp_warning_reason): Likewise.
(cpp_callbacks::diagnostic): Convert params from int to enums.
(cpp_error): Convert int param to enum cpp_diagnostic_level.
(cpp_warning): Convert int param to enum cpp_warning_reason.
(cpp_pedwarning): Likewise.
(cpp_warning_syshdr): Likewise.
(cpp_errno): Convert int param to enum cpp_diagnostic_level.
(cpp_errno_filename): Likewise.
(cpp_error_with_line): Likewise.
(cpp_warning_with_line): Convert int param to enum
cpp_warning_reason.
(cpp_pedwarning_with_line): Likewise.
(cpp_warning_with_line_syshdr): Likewise.
(cpp_error_at): Convert int param to enum cpp_diagnostic_level.
* macro.c (create_iso_definition): Convert int to enum.
(_cpp_create_definition): Likewise.

From-SVN: r264999

re PR c++/84423 ([concepts] ICE with invalid using declaration)

/cp
2018-10-09  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/84423
* pt.c (convert_template_argument): Immediately return error_mark_node
if the second argument is erroneous.
* parser.c (cp_parser_type_id): Add location_t * parameter.
(cp_parser_type_id_1): Likewise.
(cp_parser_alias_declaration): Adjust cp_parser_type_id call,
obtain the location of the type and save it.
(cp_parser_template_type_arg): Adjust.
(cp_parser_trailing_type_id): Likewise.
* decl.c (grokdeclarator): Improve error message for 'auto' in
alias declaration.

/testsuite
2018-10-09  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/84423
* g++.dg/concepts/pr84423-1.C: New.
* g++.dg/concepts/pr84423-2.C: Likewise.
* g++.dg/cpp0x/auto39.C: Test location too.
* g++.dg/cpp0x/auto9.C: Likewise.
* g++.dg/cpp1y/pr60384.C: Likewise.

From-SVN: r264996

2018-10-09  François Dumont  <fdumont@gcc.gnu.org>

* include/bits/stl_list.h
(_List_operator<>::operator==): Replace member function with inline
friend.
(_List_operator<>::operator!=): Likewise.
(_List_const_operator<>::operator==): Likewise.
(_List_const_operator<>::operator!=): Likewise.
(operator==(const _List_iterator<>&, const _List_const_iterator<>&)):
Remove.
(operator!=(const _List_iterator<>&, const _List_const_iterator<>&)):
Remove.

From-SVN: r264993

This is part 2/2 for contributing PPC64LE support for X86 SSE3 instrisics.

This is part 2/2 for contributing PPC64LE support for X86 SSE3
instrisics. This patch includes testsuite/gcc.target tests for the
intrinsics defined in pmmintrin.h.

Tested on POWER8 ppc64le and ppc64 (-m64 and -m32, the latter only reporting
10 new unsupported tests.)

[gcc/testsuite]

2018-10-09  Paul A. Clarke  <pc@us.ibm.com>

* gcc.target/powerpc/sse3-check.h: New file.
* gcc.target/powerpc/sse3-addsubps.c: New file.
* gcc.target/powerpc/sse3-addsubpd.c: New file.
* gcc.target/powerpc/sse3-haddps.c: New file.
* gcc.target/powerpc/sse3-hsubps.c: New file.
* gcc.target/powerpc/sse3-haddpd.c: New file.
* gcc.target/powerpc/sse3-hsubpd.c: New file.
* gcc.target/powerpc/sse3-lddqu.c: New file.
* gcc.target/powerpc/sse3-movsldup.c: New file.
* gcc.target/powerpc/sse3-movshdup.c: New file.
* gcc.target/powerpc/sse3-movddup.c: New file.
* gcc.target/powerpc/pr37191.c: New file.

From-SVN: r264992

This is a follow-on to earlier commits for adding compatibility implementations of x86 intrinsics for PPC64LE.

This is a follow-on to earlier commits for adding compatibility
implementations of x86 intrinsics for PPC64LE.  This is the first of
two patches.  This patch adds 11 of the 13 x86 intrinsics from
<pmmintrin.h> ("SSE3").  (Patch 2/2 adds tests for these intrinsics,
and briefly describes the tests performed.)

Implementations are relatively straightforward, with occasional
extra effort for vector element ordering.

Not implemented are _mm_wait and _mm_monitor, as there are no
direct or trivial analogs in the POWER ISA.

./gcc/ChangeLog:
2018-10-05  Paul A. Clarke  <pc@us.ibm.com>

* config.gcc (powerpc*-*-*): Add pmmintrin.h to extra_headers.
* config/rs6000/pmmintrin.h: New file.

From-SVN: r264991

re PR fortran/83522 (ICE on allocatable string reference, string(:)(:))

2018-10-09  Tobias Burnus <burnus@net-b.de>

PR fortran/83522
* resolve.c (resolve_ref): Reject nonscalar
substring references.

        PR fortran/83522
* gfortran.dg/actual_array_substr_1.f90: Add dg-error,
change to dg-do compile.
* gfortran.dg/actual_array_substr_2.f90: Ditto.
* gfortran.dg/array_initializer_1.f90: Use array
element not size-one section.
* gfortran.dg/array_substring.f90: New.

From-SVN: r264990

re PR tree-optimization/86659 (gnat.dg/sso/q[23].adb FAIL)

PR tree-optimization/86659
* gimple-match.h (gimple_match_op constructors): Initialize reverse.

From-SVN: r264986

runtime: skip testSetPanicOnFault for gollvm

    LLVM doesn't support non-call exception. This test was passing
    more or less by luck: if the faulting instruction is between two
    calls with the same landing pad (in instruction layout order,
    not the program's logic order), it generates a merged PC range
    that covers the faulting instruction. If the instruction layout
    order changes, or it uses two different (but may be degenerate)
    landing pads, this doesn't work.

    Reviewed-on: https://go-review.googlesource.com/c/140517

From-SVN: r264985

[Ada] Fix spurious -Wuninitialized warnings for small records

This change is aimed at getting rid of spurious -Wuninitialized warnings
issued for small records passed by copy and containing default values
for some of their components.

The source of the problem is that the _Init parameter of the
initialization routine is declared as an in/out parameter, so the
uninitialized object is passed by copy to it and this can be flagged by
-Wuninitialized.

That's why the mode of the parameter is changed to out, except for the
cases where information really needs to be passed in: unconstrained
array types, protected and task types.

For the following record type Rec!

 type Rec is record
    B : Boolean := True;
  end record;

the initialization routine must now be:

      procedure r__recIP (_init : out r__rec1) is
      begin
         _init.b := true;
         return;
      end r__recIP;

2018-10-09  Eric Botcazou  <ebotcazou@adacore.com>

gcc/ada/

* exp_ch3.adb (Is_Null_Statement_List): New predicate.
(Build_Array_Init_Proc): Use it to find out whether the
initialization procedure Is_Null_Init_Proc; if so, set
Warnings_Off on the parameter.
(Build_Init_Procedure): Likewise.
(Init_Formals): Use an in/out first parameter only for
unconstrained arrays and for records either containing or built
for proteced types or task types; use an out parameter in all
the other cases.
* fe.h (Is_Init_Proc): Declare.
* gcc-interface/decl.c (type_requires_init_of_formal): Do not
return true for a discriminant in an unchecked union.
(gnat_to_gnu_param): Do not create a PARM_DECL for the Out
parameter of an initialization procedure.

From-SVN: r264984

[Ada] Fix incorrect computation of Component_Size in ASIS mode

2018-10-09  Eric Botcazou  <ebotcazou@adacore.com>

gcc/ada/

* gcc-interface/decl.c (gnat_to_gnu_entity) <E_Constant>: If
this is not a definition, retrieve the expression in all cases
even if we are just annotating types.
* gcc-interface/trans.c (gnat_to_gnu): Also translate Size
attribute for constrained types in this mode.

From-SVN: r264983

[Ada] Minor tweaks to suppress warnings

2018-10-09  Eric Botcazou  <ebotcazou@adacore.com>

gcc/ada/

* adaint.c (__gnat_get_executable_suffix_ptr): Add new line.
(__gnat_locate_exec): Check that HOST_EXECUTABLE_SUFFIX is not
empty before doing a string search for it.
* socket.c (ATTRIBUTE_UNUSED): Define.
(__gnat_disable_sigpipe): Add ATTRIBUTE_UNUSED on parameter.
* terminals.c (ATTRIBUTE_UNUSED): Move around.
(__gnat_setup_child_communication): Add ATTRIBUTE_UNUSED on
parameter.
(__gnat_send_header): Add ATTRIBUTE_UNUSED on parameters.

From-SVN: r264982

[Ada] Ignore pragmas Compile_Time_Error/Warning in GNATprove mode

GNATprove does not have sometimes the precise information of the
compiler about size of types and objects, so that it cannot evaluate the
expressions in pragma Compile_Time_Error/Warning the same way the
compiler does.  Thus, these pragmas should be ignored in GNATprove mode,
as it can neither verify them nor assume them (if the expression cannot
be evaluated at compile time, then the semantics for GNAT is to ignore
them).

2018-10-09  Yannick Moy  <moy@adacore.com>

gcc/ada/

* sem_prag.adb (Process_Compile_Time_Warning_Or_Error): Rewrite
pragmas as null statements in GNATprove mode.

From-SVN: r264981

[Ada] Fix spurious error on derived record passed as Out parameter

This fixlet gets rid of a spurious error issued in the specific case of
a call to a subprogram taking an Out parameter of a discriminated record
type without default discriminants, if the actual parameter is the
result of the conversion to the record type of a variable whose type is
derived from the record and has a representation clause.

The compiler was failing to initialize the temporary made around the
call because of the representation clause, but this is required for a
type with discriminants because discriminants may be read by the called
subprogram.

2018-10-09  Eric Botcazou  <ebotcazou@adacore.com>

gcc/ada/

* exp_ch6.adb (Add_Call_By_Copy_Code): Initialize the temporary
made for an Out parameter if the formal type has discriminants.

gcc/testsuite/

* gnat.dg/derived_type5.adb, gnat.dg/derived_type5_pkg.ads: New
testcase.

From-SVN: r264980

[Ada] SPARK: fix bug related to non access object permissions

2018-10-09  Maroua Maalej  <maalej@adacore.com>

gcc/ada/

* sem_spark.adb (Check_Declaration): fix bug related to non
access object permissions.

From-SVN: r264979

[Ada] Minor copyright header fix

2018-10-09  Doug Rupp  <rupp@adacore.com>

gcc/ada/

* libgnat/a-ncelfu.ads: Fix name in header to match package.

From-SVN: r264978

[Ada] Fix strange warning when using Ada.Iterator_Interface

The back-end was recently changed to issue more -Wuninitialized warnings
on Out parameters and this has caught a case related to
Ada.Iterator_Interface.: This patchlet simply kills this uninteresting
warning.

2018-10-09  Eric Botcazou  <ebotcazou@adacore.com>

gcc/ada/

* exp_disp.adb (Make_Disp_Asynchronous_Select_Spec): Set
Warnings_Off on the B out parameter.

gcc/testsuite/

* gnat.dg/warn17.adb: New testcase.

From-SVN: r264977

[Ada] Einfo: minor comment fix

2018-10-09  Eric Botcazou  <ebotcazou@adacore.com>

gcc/ada/

* einfo.ads: Small comment fix.

From-SVN: r264976

[Ada] Internal error on inlined renaming of subprogram instance

This fixes a recent regression introduced in the compiler for the
inlined renaming of a subprogram instantiated in a package body.  It was
wrongly clearing the Is_Public flag on the entity associated with the
body.

2018-10-09  Eric Botcazou  <ebotcazou@adacore.com>

gcc/ada/

* sem_ch7.adb (Has_Referencer): Add comment for the
N_Freeze_Entity case.  Do not rely on
Has_Referencer_Of_Non_Subprograms to clear the Is_Public flag on
subprogram entities.

gcc/testsuite/

* gnat.dg/inline14.adb, gnat.dg/inline14_pkg.adb,
gnat.dg/inline14_pkg.ads: New testcase.

From-SVN: r264975

[Ada] Unnesting: fix handling of private types that are synchronized

2018-10-09  Ed Schonberg  <schonberg@adacore.com>

gcc/ada/

* exp_unst.adb (In_Synchronized_Call): Handle properly private
types whose full views are synchronized types, in order to
recognize generated subprograms whose bodies must be considered
reachable even if no direct calls to them are visible in the
source.

From-SVN: r264974

[Ada] Fix expansion of operations on nonbinary modular types

2018-10-09  Ed Schonberg  <schonberg@adacore.com>

gcc/ada/

* exp_ch4.adb (Expand_Modular_Op): When expanding an operation
on nonbinary modular types, convert the opersnds to an integer
type that is large enough to hold the modulus of the type, which
may be larger than Integer'Last.

From-SVN: r264973

[Ada] Unnesting: fix handling of uplevel refs to unconstrained formals

2018-10-09  Ed Schonberg  <schonberg@adacore.com>

gcc/ada/

* exp_unst.adb (Unnest_Subprogram):  When an uplevel reference
is to an unconstrained formal, the 'Access reference that is
created to initialize the corresponding component of the
activation record must be wrapped in an unchecked conversion to
the generated type of the component. Otherwise, spurious suvtype
conformance errors will be generated when the code is within an
instantiation and the type of the formal is a formal type of the
enclosing generic. Note that during unnesting there is no simple
way to determine that the code appears within an instance
because ther is no scope stack.

From-SVN: r264972

[Ada] Remove spurious -Wuninitialized warning for small array at -O2

2018-10-09  Eric Botcazou  <ebotcazou@adacore.com>

gcc/ada/

* gcc-interface/decl.c (type_requires_init_of_formal): New
predicate.
(gnat_to_gnu_param): Use it to determine whether a PARM_DECL
must be created for an Out parameter passed by copy.

From-SVN: r264971

[Ada] Bounded_Holders: fix a typo

2018-10-09  Arnaud Charlet  <charlet@adacore.com>

gcc/ada/

* libgnat/a-coboho.ads (Storage_Element): Fix a typo.

From-SVN: r264970

[Ada] Functional_Vectors: remove default value for max size

2018-10-09  Claire Dross  <dross@adacore.com>

gcc/ada/

* libgnat/a-cfinve.ads: Remove default value for
Max_Size_In_Storage_Elements as it was not supported in SPARK.

From-SVN: r264969