]> git.ipfire.org Git - thirdparty/gcc.git/log
thirdparty/gcc.git
18 months agoc: Fix ICE for nested enum redefinitions with/without fixed underlying type [PR112571]
Joseph Myers [Wed, 31 Jan 2024 21:39:53 +0000 (21:39 +0000)] 
c: Fix ICE for nested enum redefinitions with/without fixed underlying type [PR112571]

Bug 112571 reports an ICE-on-invalid for cases where an enum is
defined, without a fixed underlying type, inside the enum type
specifier for a definition of that same enum with a fixed underlying
type.

The ultimate cause is attempting to access ENUM_UNDERLYING_TYPE in a
case where it is NULL.  Avoid this by clearing
ENUM_FIXED_UNDERLYING_TYPE_P in thie case of inconsistent definitions.

Bootstrapped wth no regressions for x86_64-pc-linux-gnu.

PR c/112571

gcc/c/
* c-decl.cc (start_enum): Clear ENUM_FIXED_UNDERLYING_TYPE_P when
defining without a fixed underlying type an enumeration previously
declared with a fixed underlying type.

gcc/testsuite/
* gcc.dg/c23-enum-9.c, gcc.dg/c23-enum-10.c: New tests.

18 months agomatch: Fix vcond into conditional op folding [PR113607].
Robin Dapp [Tue, 30 Jan 2024 17:39:08 +0000 (18:39 +0100)] 
match: Fix vcond into conditional op folding [PR113607].

In PR113607 we see an invalid fold of

  _429 = .COND_SHL (mask_patt_205.47_276, vect_cst__262, vect_cst__262, { 0, ... });
  vect_prephitmp_129.51_282 = _429;
  vect_iftmp.55_287 = VEC_COND_EXPR <mask_patt_209.54_286, vect_prephitmp_129.51_282, vect_cst__262>;

to

  Applying pattern match.pd:9607, gimple-match-10.cc:3817
  gimple_simplified to vect_iftmp.55_287 = .COND_SHL (mask_patt_205.47_276, vect_cst__262, vect_cst__262, { 0, ... });

where we essentially use COND_SHL's else instead of VEC_COND_EXPR's.

This patch adjusts the corresponding match.pd pattern and makes it only
match when the else values are the same.

That, however, causes the exact test case for which this pattern was
introduced for to fail.  Therefore XFAIL it for now.

gcc/ChangeLog:

PR middle-end/113607

* match.pd: Make sure else values match when folding a
vec_cond into a conditional operation.

gcc/testsuite/ChangeLog:

* gcc.target/aarch64/sve/pre_cond_share_1.c: XFAIL.
* gcc.target/riscv/rvv/autovec/pr113607-run.c: New test.
* gcc.target/riscv/rvv/autovec/pr113607.c: New test.

18 months agoc++: add deprecation notice for -fconcepts-ts
Marek Polacek [Tue, 30 Jan 2024 22:11:34 +0000 (17:11 -0500)] 
c++: add deprecation notice for -fconcepts-ts

We plan to remove -fconcepts-ts in GCC 15 and thus remove the flag_concepts_ts
code.  This note is an admonishing reminder to convert the Concepts TS
code to C++20 Concepts.

gcc/c-family/ChangeLog:

* c-opts.cc (c_common_post_options): Add an inform saying that
-fconcepts-ts is deprecated and will be removed in GCC 15.

gcc/ChangeLog:

* doc/invoke.texi: Mention that -fconcepts-ts was deprecated in GCC 14.

18 months agoFix ICE with -g and -std=c23 when forming composite types [PR113438]
Martin Uecker [Tue, 23 Jan 2024 12:33:34 +0000 (13:33 +0100)] 
Fix ICE with -g and -std=c23 when forming composite types [PR113438]

Set TYPE_STUB_DECL to an artificial decl when creating a new structure
as a composite type.

PR c/113438

gcc/c/
* c-typeck.cc (composite_type_internal): Set TYPE_STUB_DECL.

gcc/testsuite/
* gcc.dg/pr113438.c: New test.

18 months agouninit-pr108968-register.c: use __UINTPTR_TYPE__ for LLP64
Jonathan Yong [Wed, 31 Jan 2024 13:31:30 +0000 (13:31 +0000)] 
uninit-pr108968-register.c: use __UINTPTR_TYPE__ for LLP64

Ensure sp variable is long enough by using __UINTPTR_TYPE__ for
rsp.

gcc/testsuite/ChangeLog:

* c-c++-common/analyzer/uninit-pr108968-register.c:
Use __UINTPTR_TYPE__ instead of unsigned long for LLP64.

18 months agomodula2: tidyup patch
Gaius Mulley [Wed, 31 Jan 2024 15:51:49 +0000 (15:51 +0000)] 
modula2: tidyup patch

This patch improves a comment and also adds the location tokenno to
possibly exported idents as they are encountered.

gcc/m2/ChangeLog:

* gm2-compiler/M2Comp.mod (Pass0CheckMod): Tidy up comment.
* gm2-compiler/P1Build.bnf (PossiblyExportIdent): Replace
PushTF with PushTFtok.

Signed-off-by: Gaius Mulley <gaiusmod2@gmail.com>
18 months agoPR modula2/111627 Excess test fails with a case-preserving-case-insensitive source...
Gaius Mulley [Wed, 31 Jan 2024 15:44:32 +0000 (15:44 +0000)] 
PR modula2/111627 Excess test fails with a case-preserving-case-insensitive source tree

This patch renames gm2 testsuite modules whose names conflict with library
modules.  The conflict is not seen on case preserving case sensitive file
systems.

gcc/testsuite/ChangeLog:

PR modula2/111627
* gm2/pim/pass/stdio.mod: Moved to...
* gm2/pim/pass/teststdio.mod: ...here.
* gm2/pim/run/pass/builtins.mod: Moved to...
* gm2/pim/run/pass/testbuiltins.mod: ...here.
* gm2/pim/run/pass/math.mod: Moved to...
* gm2/pim/run/pass/testmath.mod: ...here.
* gm2/pim/run/pass/math2.mod: Moved to...
* gm2/pim/run/pass/testmath2.mod: ...here.

Signed-off-by: Gaius Mulley <gaiusmod2@gmail.com>
18 months agoAArch64: relax cbranch tests to accepted inverted branches [PR113502]
Tamar Christina [Wed, 31 Jan 2024 14:52:59 +0000 (14:52 +0000)] 
AArch64: relax cbranch tests to accepted inverted branches [PR113502]

Recently something in the midend had started inverting the branches by inverting
the condition and the branches.

While this is fine, it makes it hard to actually test.  In RTL I disable
scheduling and BB reordering to prevent this.  But in GIMPLE there seems to be
nothing I can do.  __builtin_expect seems to have no impact on the change since
I suspect this is happening during expand where conditions can be flipped
regardless of probability during compare_and_branch.

Since the mid-end has plenty of correctness tests, this weakens the backend
tests to just check that a correct looking sequence is emitted.

gcc/testsuite/ChangeLog:

PR testsuite/113502
* gcc.target/aarch64/sve/vect-early-break-cbranch.c: Ignore exact branch.
* gcc.target/aarch64/vect-early-break-cbranch.c: Likewise.

18 months agohwasan: Remove testsuite check for a complaint message [PR112644]
Tamar Christina [Wed, 31 Jan 2024 14:51:36 +0000 (14:51 +0000)] 
hwasan: Remove testsuite check for a complaint message [PR112644]

With recent updates to hwasan runtime libraries, the error reporting for
this particular check is has been reworked.

I would question why it has lost this message.  To me it looks strange
that num_descriptions_printed is incremented whenever we call
PrintHeapOrGlobalCandidate whether that function prints anything or not.
(See PrintAddressDescription in libsanitizer/hwasan/hwasan_report.cpp).

The message is no longer printed because we increment this
num_descriptions_printed variable indicating that we have found some
description.

I would like to question this upstream, but it doesn't look that much of
a problem and if pressed for time we should just change our testsuite.
Bootstrapped Regtested on aarch64-none-linux-gnu and no issues.

gcc/testsuite/ChangeLog:

PR sanitizer/112644
* c-c++-common/hwasan/hwasan-thread-clears-stack.c: Update testcase.

18 months agohwasan: instrument new memory and string functions [PR112644]
Tamar Christina [Wed, 31 Jan 2024 14:50:33 +0000 (14:50 +0000)] 
hwasan: instrument new memory and string functions [PR112644]

Recent libhwasan updates[1] intercept various string and memory functions.
These functions have checking in them, which means there's no need to
inline the checking.

This patch marks said functions as intercepted, and adjusts a testcase
to handle the difference.  It also looks for HWASAN in a check in
expand_builtin.  This check originally is there to avoid using expand to
inline the behaviour of builtins like memset which are intercepted by
ASAN and hence which we rely on the function call staying as a function
call.  With the new reliance on function calls in HWASAN we need to do
the same thing for HWASAN too.

HWASAN and ASAN don't seem to however instrument the same functions.

Looking into libsanitizer/sanitizer_common/sanitizer_common_interceptors_memintrinsics.inc
it looks like the common ones are memset, memmove and memcpy.

The rest of the routines for asan seem to be defined in
compiler-rt/lib/asan/asan_interceptors.h however compiler-rt/lib/hwasan/
does not have such a file but it does have
compiler-rt/lib/hwasan/hwasan_platform_interceptors.h which it looks like is
forcing off everything but memset, memmove, memcpy, memcmp and bcmp.

As such I've taken those as the final list that hwasan currently supports.
This also means that on future updates this list should be cross checked.

[1] https://discourse.llvm.org/t/hwasan-question-about-the-recent-interceptors-being-added/75351

gcc/ChangeLog:

PR sanitizer/112644
* asan.h (asan_intercepted_p): Incercept memset, memmove, memcpy and
memcmp.
* builtins.cc (expand_builtin): Include HWASAN when checking for
builtin inlining.

gcc/testsuite/ChangeLog:

PR sanitizer/112644
* c-c++-common/hwasan/builtin-special-handling.c: Update testcase.

Co-Authored-By: Matthew Malcomson <matthew.malcomson@arm.com>
18 months agolibsanitizer: Sync fixes for asan interceptors from upstream
Tamar Christina [Wed, 31 Jan 2024 14:44:35 +0000 (14:44 +0000)] 
libsanitizer: Sync fixes for asan interceptors from upstream

This cherry-picks and squashes the differences between commits

d3e5c20ab846303874a2a25e5877c72271fc798b..76e1e45922e6709392fb82aac44bebe3dbc2ea63
from LLVM upstream from compiler-rt/lib/hwasan/ to GCC on the changes relevant
for GCC.

This is required to fix the linked PR.

As mentioned in the PR the last sync brought in a bug from upstream[1] where
operations became non-recoverable and as such the tests in AArch64 started
failing.  This cherry picks the fix and there are minor updates needed to GCC
after this to fix the cases.

[1] https://github.com/llvm/llvm-project/pull/74000

PR sanitizer/112644
Cherry-pick llvm-project revision
672b71cc1003533460a82f06b7d24fbdc02ffd58,
5fcf3bbb1acfe226572474636714ede86fffcce8,
3bded112d02632209bd55fb28c6c5c234c23dec3 and
76e1e45922e6709392fb82aac44bebe3dbc2ea63.

18 months agomiddle-end/110176 - wrong zext (bool) <= (int) 4294967295u folding
Richard Biener [Wed, 31 Jan 2024 13:40:24 +0000 (14:40 +0100)] 
middle-end/110176 - wrong zext (bool) <= (int) 4294967295u folding

The following fixes a wrong pattern that didn't match the behavior
of the original fold_widened_comparison in that get_unwidened
returned a constant always in the wider type.  But here we're
using (int) 4294967295u without the conversion applied.  Fixed
by doing as earlier in the pattern - matching constants only
if the conversion was actually applied.

PR middle-end/110176
* match.pd (zext (bool) <= (int) 4294967295u): Make sure
to match INTEGER_CST only without outstanding conversion.

* gcc.dg/torture/pr110176.c: New testcase.

18 months agoaarch64: Avoid out-of-range shrink-wrapped saves [PR111677]
Alex Coplan [Tue, 30 Jan 2024 10:22:48 +0000 (10:22 +0000)] 
aarch64: Avoid out-of-range shrink-wrapped saves [PR111677]

The PR shows us ICEing due to an unrecognizable TFmode save emitted by
aarch64_process_components.  The problem is that for T{I,F,D}mode we
conservatively require mems to be in range for x-register ldp/stp.  That
is because (at least for TImode) it can be allocated to both GPRs and
FPRs, and in the GPR case that is an x-reg ldp/stp, and the FPR case is
a q-register load/store.

As Richard pointed out in the PR, aarch64_get_separate_components
already checks that the offsets are suitable for a single load, so we
just need to choose a mode in aarch64_reg_save_mode that gives the full
q-register range.  In this patch, we choose V16QImode as an alternative
16-byte "bag-of-bits" mode that doesn't have the artificial range
restrictions imposed on T{I,F,D}mode.

For T{F,D}mode in GCC 15 I think we could consider relaxing the
restriction imposed in aarch64_classify_address, as typically T{F,D}mode
should be allocated to FPRs.  But such a change seems too invasive to
consider for GCC 14 at this stage (let alone backports).

Fortunately the new flexible load/store pair patterns in GCC 14 allow
this mode change to work without further changes.  The backports are
more involved as we need to adjust the load/store pair handling to cater
for V16QImode in a few places.

Note that for the testcase we are relying on the torture options to add
-funroll-loops at -O3 which is necessary to trigger the ICE on trunk
(but not on the 13 branch).

gcc/ChangeLog:

PR target/111677
* config/aarch64/aarch64.cc (aarch64_reg_save_mode): Use
V16QImode for the full 16-byte FPR saves in the vector PCS case.

gcc/testsuite/ChangeLog:

PR target/111677
* gcc.target/aarch64/torture/pr111677.c: New test.

18 months agotestsuite: i386: Disable .eh_frame in gcc.target/i386/auto-init-5.c etc.
Rainer Orth [Wed, 31 Jan 2024 13:49:22 +0000 (14:49 +0100)] 
testsuite: i386: Disable .eh_frame in gcc.target/i386/auto-init-5.c etc.

The gcc.target/i386/auto-init-5.c and gcc.target/i386/auto-init-6.c
tests FAIL on 64-bit Solaris/x86 with the native assembler:

FAIL: gcc.target/i386/auto-init-5.c scan-assembler-times \\\\.long\\t0 14
FAIL: gcc.target/i386/auto-init-6.c scan-assembler-times long\\t0 8

/bin/as doesn't fully support the CFI directives, so the .eh_frame
sections are emitted directly and contain .long.  Since .eh_frame
doesn't matter for those tests, this patch disables its generation in
the first place.

Tested on i386-pc-solaris2.11 (as and gas) and i686-pc-linux-gnu.

2024-01-30  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>

gcc/testsuite:
* gcc.target/i386/auto-init-5.c: Add
-fno-asynchronous-unwind-tables to dg-options.
* gcc.target/i386/auto-init-6.c: Likewise.

18 months agotree-optimization/111444 - avoid insertions when skipping defs
Richard Biener [Wed, 31 Jan 2024 09:42:48 +0000 (10:42 +0100)] 
tree-optimization/111444 - avoid insertions when skipping defs

The following avoids inserting expressions for IPA CP discovered
equivalences into the VN hashtables when we are optimistically
skipping may-defs in the attempt to prove it's redundant.

PR tree-optimization/111444
* tree-ssa-sccvn.cc (vn_reference_lookup_3): Do not use
vn_reference_lookup_2 when optimistically skipping may-defs.

* gcc.dg/torture/pr111444.c: New testcase.

18 months agotestsuite: Require ucn in g++.dg/cpp0x/udlit-extended-id-1.C
Rainer Orth [Wed, 31 Jan 2024 12:34:07 +0000 (13:34 +0100)] 
testsuite: Require ucn in g++.dg/cpp0x/udlit-extended-id-1.C

g++.dg/cpp0x/udlit-extended-id-1.C FAILs on Solaris/SPARC and x86 with
the native assembler:

UNRESOLVED: g++.dg/cpp0x/udlit-extended-id-1.C  -std=c++14 compilation failed to produce executable
FAIL: g++.dg/cpp0x/udlit-extended-id-1.C  -std=c++17 (test for excess errors)
UNRESOLVED: g++.dg/cpp0x/udlit-extended-id-1.C  -std=c++17 compilation failed to produce executable
FAIL: g++.dg/cpp0x/udlit-extended-id-1.C  -std=c++20 (test for excess errors)
UNRESOLVED: g++.dg/cpp0x/udlit-extended-id-1.C  -std=c++20 compilation failed to produce executable

/bin/as doesn't support UCN identifiers:

/usr/ccs/bin/as: "/var/tmp//ccCl_9fa.s", line 4: error: invalid character (0xcf)
/usr/ccs/bin/as: "/var/tmp//ccCl_9fa.s", line 4: error: invalid character (0x80)
/usr/ccs/bin/as: "/var/tmp//ccCl_9fa.s", line 4: error: statement syntax
/usr/ccs/bin/as: "/var/tmp//ccCl_9fa.s", line 4: error: statement syntax
[...]

To avoid this, this patch requires ucn support.

Tested on i386-pc-solaris2.11 (as and gas), sparc-sun-solaris2.11 (as
and gas), and i686-pc-linux-gnu.

2024-01-30  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>

gcc/testsuite:
* g++.dg/cpp0x/udlit-extended-id-1.C: Require ucn support.

18 months agotree-optimization/113630 - invalid code hoisting
Richard Biener [Wed, 31 Jan 2024 10:28:50 +0000 (11:28 +0100)] 
tree-optimization/113630 - invalid code hoisting

The following avoids code hoisting (but also PRE insertion) of
expressions that got value-numbered to another one that are not
a valid replacement (but still compute the same value).  This time
because the access path ends in a structure with different size,
meaning we consider a related access as not trapping because of the
size of the base of the access.

PR tree-optimization/113630
* tree-ssa-pre.cc (compute_avail): Avoid registering a
reference with a representation with not matching base
access size.

* gcc.dg/torture/pr113630.c: New testcase.

18 months agosimplify-rtx: Fix up last argument to simplify_gen_unary [PR113656]
Jakub Jelinek [Wed, 31 Jan 2024 09:56:56 +0000 (10:56 +0100)] 
simplify-rtx: Fix up last argument to simplify_gen_unary [PR113656]

When simplifying e.g. (float_truncate:SF (float_truncate:DF (reg:XF))
or (float_truncate:SF (float_extend:XF (reg:DF)) etc. into
(float_truncate:SF (reg:XF)) or (float_truncate:SF (reg:DF)) we call
simplify_gen_unary with incorrect op_mode argument, it should be
the argument's mode, but we call it with the outer mode instead.
As these are all floating point operations, the argument always
has non-VOIDmode and so we can just use that mode (as done in similar
simplifications a few lines later), but neither FLOAT_TRUNCATE nor
FLOAT_EXTEND are operations that should have the same modes of operand
and result.  This bug hasn't been a problem for years because normally
op_mode is used only if the mode of op is VOIDmode, otherwise it is
redundant, but r10-2139 added an assertion in some spots that op_mode
is right even in such cases.

2024-01-31  Jakub Jelinek  <jakub@redhat.com>

PR rtl-optimization/113656
* simplify-rtx.cc (simplify_context::simplify_unary_operation_1)
<case FLOAT_TRUNCATE>: Fix up last argument to simplify_gen_unary.

* gcc.target/i386/pr113656.c: New test.

18 months agodwarf2out: Fix ICE on large _BitInt in loc_list_from_tree_1 [PR113637]
Jakub Jelinek [Wed, 31 Jan 2024 09:56:15 +0000 (10:56 +0100)] 
dwarf2out: Fix ICE on large _BitInt in loc_list_from_tree_1 [PR113637]

This spot uses SCALAR_INT_TYPE_MODE which obviously ICEs for large/huge
BITINT_TYPE types which have BLKmode.  But such large BITINT_TYPEs certainly
don't fit into DWARF2_ADDR_SIZE either, so we can just assume it would be
false if type has BLKmode.

2024-01-31  Jakub Jelinek  <jakub@redhat.com>

PR debug/113637
* dwarf2out.cc (loc_list_from_tree_1): Assume integral types
with BLKmode are larger than DWARF2_ADDR_SIZE.

* gcc.dg/bitint-80.c: New test.

18 months agolower-bitint: Fix up VIEW_CONVERT_EXPR handling in handle_operand_addr [PR113639]
Jakub Jelinek [Wed, 31 Jan 2024 09:50:33 +0000 (10:50 +0100)] 
lower-bitint: Fix up VIEW_CONVERT_EXPR handling in handle_operand_addr [PR113639]

Yet another spot where we need to treat VIEW_CONVERT_EXPR differently
from NOP_EXPR/CONVERT_EXPR.

2024-01-31  Jakub Jelinek  <jakub@redhat.com>

PR tree-optimization/113639
* gimple-lower-bitint.cc (bitint_large_huge::handle_operand_addr):
For VIEW_CONVERT_EXPR set rhs1 to its operand.

* gcc.dg/bitint-79.c: New test.

18 months agolibstdc++: Enable std::text_encoding for darwin and FreeBSD
Jonathan Wakely [Tue, 30 Jan 2024 14:28:09 +0000 (14:28 +0000)] 
libstdc++: Enable std::text_encoding for darwin and FreeBSD

The <xlocale.h> header is needed for newlocale and locale_t on these
targets.

libstdc++-v3/ChangeLog:

* acinclude.m4 (GLIBCXX_CHECK_TEXT_ENCODING): Use <xlocale.h> if
needed for newlocale.
* configure: Regenerate.
* src/c++26/text_encoding.cc: Use <xlocale.h>.

Reviewed-by: Iain Sandoe <iain@sandoe.co.uk>
18 months agolibstdc++: Add "ASCII" as an alias for std::text_encoding::id::ASCII
Jonathan Wakely [Tue, 23 Jan 2024 14:57:15 +0000 (14:57 +0000)] 
libstdc++: Add "ASCII" as an alias for std::text_encoding::id::ASCII

As noted in LWG 4043, "ASCII" is not an alias for any known registered
character encoding, so std::text_encoding("ASCII").mib() == id::other.
Add the alias "ASCII" to the implementation-defined superset of aliases
for that encoding.

libstdc++-v3/ChangeLog:

* include/bits/text_encoding-data.h: Regenerate.
* scripts/gen_text_encoding_data.py: Add extra_aliases dict
containing "ASCII".
* testsuite/std/text_encoding/cons.cc: Check "ascii" is known.

Co-authored-by: Ewan Higgs <ewan.higgs@gmail.com>
Signed-off-by: Ewan Higgs <ewan.higgs@gmail.com>
18 months agolibstdc++: Add all supported headers to lists in the manual
Jonathan Wakely [Tue, 23 Jan 2024 14:49:51 +0000 (14:49 +0000)] 
libstdc++: Add all supported headers to lists in the manual

libstdc++-v3/ChangeLog:

* doc/xml/manual/using.xml: Update tables of supported headers.
* doc/html/*: Regenerate.

18 months agolibstdc++: Fix -Wshift-count-overflow warning in std::bitset
Jonathan Wakely [Sat, 20 Jan 2024 21:09:28 +0000 (21:09 +0000)] 
libstdc++: Fix -Wshift-count-overflow warning in std::bitset

This shift only happens if the unsigned long long type is wider than
unsigned long but the compiler warns when it sees the shift, without
caring if it's reachable.

Use the preprocessor to compare the sizes and just reuse _M_to_ulong()
if sizeof(long) == sizeof(long long).

libstdc++-v3/ChangeLog:

* include/std/bitset (_Base_bitset::_M_do_to_ullong): Avoid
-Wshift-count-overflow warning.

18 months agotree-optimization/113670 - gather/scatter to/from hard registers
Richard Biener [Wed, 31 Jan 2024 08:09:50 +0000 (09:09 +0100)] 
tree-optimization/113670 - gather/scatter to/from hard registers

The following makes sure we're not taking the address of hard
registers when vectorizing appearant gathers or scatters to/from
them.

PR tree-optimization/113670
* tree-vect-data-refs.cc (vect_check_gather_scatter):
Make sure we can take the address of the reference base.

* gcc.target/i386/pr113670.c: New testcase.

18 months agoAVR: Add AVR64DU and some older devices.
Georg-Johann Lay [Wed, 31 Jan 2024 08:33:37 +0000 (09:33 +0100)] 
AVR: Add AVR64DU and some older devices.

gcc/
* config/avr/avr-mcus.def: Add AVR64DU28, AVR64DU32, ATA5787,
ATA5835, ATtiny64AUTO, ATA5700M322.
* doc/avr-mmcu.texi: Rebuild.

18 months agostrub: drop nonaliased parm from build_ref_type_for [PR113394]
Alexandre Oliva [Wed, 31 Jan 2024 03:13:27 +0000 (00:13 -0300)] 
strub: drop nonaliased parm from build_ref_type_for [PR113394]

Variant type copies can't have their own alias sets any more, and it's
not like setting them affected the pointed-to objects anyway.

for  gcc/ChangeLog

PR debug/113394
* ipa-strub.cc (build_ref_type_for): Drop nonaliased.  Adjust
caller.

for  gcc/testsuite/ChangeLog

PR debug/113394
* gcc.dg/strub-internal-pr113394.c: New.

18 months ago0From: Alexandre Oliva <oliva@adacore.com>
Alexandre Oliva [Wed, 31 Jan 2024 03:13:36 +0000 (00:13 -0300)] 
0From: Alexandre Oliva <oliva@adacore.com>

strub: introduce STACK_ADDRESS_OFFSET

Since STACK_POINTER_OFFSET is not necessarily at the boundary between
caller- and callee-owned stack, as desired by
__builtin_stack_address(), and using it as if it were or not causes
problems, introduce a new macro so that ports can define it suitably,
without modifying STACK_POINTER_OFFSET.

for  gcc/ChangeLog

PR middle-end/112917
PR middle-end/113100
* builtins.cc (expand_builtin_stack_address): Use
STACK_ADDRESS_OFFSET.
* doc/extend.texi (__builtin_stack_address): Adjust.
* config/sparc/sparc.h (STACK_ADDRESS_OFFSET): Define.
* doc/tm.texi.in (STACK_ADDRESS_OFFSET): Document.
* doc/tm.texi: Rebuilt.

18 months agoc: Fix ICEs casting expressions with integer constant operands to bool [PR111059...
Joseph Myers [Wed, 31 Jan 2024 01:24:21 +0000 (01:24 +0000)] 
c: Fix ICEs casting expressions with integer constant operands to bool [PR111059, PR111911]

C front-end bugs 111059 and 111911 both report ICEs with conversions
to boolean of expressions with integer constant operands that can
appear in an integer constant expression as long as they are not
evaluated (such as division by zero).

The issue is a nested C_MAYBE_CONST_EXPR, with the inner one generated
in build_binary_op to indicate that a subexpression has been fully
folded and should not be folded again, and the outer one in
build_c_cast to indicate that the expression has integer constant
operands.  To avoid the inner one from build_binary_op,
c_objc_common_truthvalue_conversion should be given an argument
properly marked as having integer constant operands rather than that
information having been removed by the caller - but because c_convert
would then also wrap a C_MAYBE_CONST_EXPR with a NOP_EXPR converting
to boolean, it seems most convenient to have
c_objc_common_truthvalue_conversion produce the NE_EXPR directly in
the desired type (boolean in this case), before generating any
C_MAYBE_CONST_EXPR there, rather than it always producing a comparison
in integer_type_node and doing a conversion to boolean in the caller.

The same issue as in those PRs also applies for conversion to enums
with a boolean fixed underlying type; that case is also fixed and
tests added for it.  Note that not all the tests added failed before
the patch (in particular, the issue was specific to casts and did not
apply for implicit conversions, but some tests of those are added as
well).

Bootstrapped with no regressions for x86_64-pc-linux-gnu.

PR c/111059
PR c/111911

gcc/c/
* c-tree.h (c_objc_common_truthvalue_conversion): Add third
argument.
* c-convert.cc (c_convert): For conversions to boolean, pass third
argument to c_objc_common_truthvalue_conversion rather than
converting here.
* c-typeck.cc (build_c_cast): Ensure arguments with integer
operands are marked as such for conversion to boolean.
(c_objc_common_truthvalue_conversion): Add third argument TYPE.

gcc/testsuite/
* gcc.c-torture/compile/pr111059-1.c,
gcc.c-torture/compile/pr111059-2.c,
gcc.c-torture/compile/pr111059-3.c,
gcc.c-torture/compile/pr111059-4.c,
gcc.c-torture/compile/pr111059-5.c,
gcc.c-torture/compile/pr111059-6.c,
gcc.c-torture/compile/pr111059-7.c,
gcc.c-torture/compile/pr111059-8.c,
gcc.c-torture/compile/pr111059-9.c,
gcc.c-torture/compile/pr111059-10.c,
gcc.c-torture/compile/pr111059-11.c,
gcc.c-torture/compile/pr111059-12.c,
gcc.c-torture/compile/pr111911-1.c,
gcc.c-torture/compile/pr111911-2.c: New tests.

18 months agoanalyzer: handle null "var" in state_change_event::get_desc [PR113509]
David Malcolm [Wed, 31 Jan 2024 01:06:31 +0000 (20:06 -0500)] 
analyzer: handle null "var" in state_change_event::get_desc [PR113509]

Avoid ICE with  -fanalyzer-verbose-state-changes when
region_model::get_representative_tree returns nullptr in
state_change_event::get_desc.

gcc/analyzer/ChangeLog:
PR analyzer/113509
* checker-event.cc (state_change_event::get_desc): Don't assume
"var" is non-NULL.

gcc/testsuite/ChangeLog:
PR analyzer/113509
* c-c++-common/analyzer/stdarg-pr113509.c: New test.

Signed-off-by: David Malcolm <dmalcolm@redhat.com>
18 months agoRISC-V: Fix VSETLV PASS compile-time issue
Juzhe-Zhong [Mon, 29 Jan 2024 11:32:02 +0000 (19:32 +0800)] 
RISC-V: Fix VSETLV PASS compile-time issue

The compile time issue was discovered in SPEC 2017 wrf:

Use time and -ftime-report to analyze the profile data of SPEC 2017 wrf compilation .

Before this patch (Lazy vsetvl):

scheduling                         : 121.89 ( 15%)   0.53 ( 11%) 122.72 ( 15%)    13M (  1%)
machine dep reorg                  : 424.61 ( 53%)   1.84 ( 37%) 427.44 ( 53%)  5290k (  0%)
real    13m27.074s
user    13m19.539s
sys     0m5.180s

Simple vsetvl:

machine dep reorg                  :   0.10 (  0%)   0.00 (  0%)   0.11 (  0%)  4138k (  0%)
real    6m5.780s
user    6m2.396s
sys     0m2.373s

The machine dep reorg is the compile time of VSETVL PASS (424 seconds) which counts 53% of
the compilation time, spends much more time than scheduling.

After investigation, the critical patch of VSETVL pass is compute_lcm_local_properties which
is called every iteration of phase 2 (earliest fusion) and phase 3 (global lcm).

This patch optimized the codes of compute_lcm_local_properties to reduce the compilation time.

After this patch:

scheduling                         : 117.51 ( 27%)   0.21 (  6%) 118.04 ( 27%)    13M (  1%)
machine dep reorg                  :  80.13 ( 18%)   0.91 ( 26%)  81.26 ( 18%)  5290k (  0%)
real    7m25.374s
user    7m20.116s
sys     0m3.795s

The optimization of this patch is very obvious, lazy VSETVL PASS: 424s (53%) -> 80s (18%) which
spend less time than scheduling.

Tested on both RV32 and RV64 no regression.  Ok for trunk ?

PR target/113495

gcc/ChangeLog:

* config/riscv/riscv-vsetvl.cc (extract_single_source): Remove.
(pre_vsetvl::compute_vsetvl_def_data): Fix compile time issue.
(pre_vsetvl::compute_transparent): New function.
(pre_vsetvl::compute_lcm_local_properties): Fix compile time time issue.

18 months agoDaily bump.
GCC Administrator [Wed, 31 Jan 2024 00:19:27 +0000 (00:19 +0000)] 
Daily bump.

18 months agoi386: Add "Ws" constraint for symbolic address/label reference [PR105576]
Fangrui Song [Thu, 11 Jan 2024 18:24:25 +0000 (10:24 -0800)] 
i386: Add "Ws" constraint for symbolic address/label reference [PR105576]

Printing the raw symbol is useful in inline asm (e.g. in C++ to get the
mangled name).  Similar constraints are available in other targets (e.g.
"S" for aarch64/riscv, "Cs" for m68k).

There isn't a good way for x86 yet, e.g. "i" doesn't work for
PIC/-mcmodel=large.  This patch adds "Ws".  Here are possible use cases:

```
namespace ns { extern int var; }
asm (".pushsection .xxx,\"aw\"; .dc.a %0; .popsection" :: "Ws"(&var));
asm (".reloc ., BFD_RELOC_NONE, %0" :: "Ws"(&var));
```

gcc/ChangeLog:

PR target/105576
* config/i386/constraints.md: Define constraint "Ws".
* doc/md.texi: Document it.

gcc/testsuite/ChangeLog:

PR target/105576
* gcc.target/i386/asm-raw-symbol.c: New testcase.

18 months agoc++: avoid -Wdangling-reference for std::span-like classes [PR110358]
Marek Polacek [Thu, 25 Jan 2024 17:08:14 +0000 (12:08 -0500)] 
c++: avoid -Wdangling-reference for std::span-like classes [PR110358]

Real-world experience shows that -Wdangling-reference triggers for
user-defined std::span-like classes a lot.  We can easily avoid that
by considering classes like

    template<typename T>
    struct Span {
      T* data_;
      std::size len_;
    };

to be std::span-like, and not warning for them.  Unlike the previous
patch, this one considers a non-union class template that has a pointer
data member and a trivial destructor as std::span-like.

PR c++/110358
PR c++/109640

gcc/cp/ChangeLog:

* call.cc (reference_like_class_p): Don't warn for std::span-like
classes.

gcc/ChangeLog:

* doc/invoke.texi: Update -Wdangling-reference description.

gcc/testsuite/ChangeLog:

* g++.dg/warn/Wdangling-reference18.C: New test.
* g++.dg/warn/Wdangling-reference19.C: New test.
* g++.dg/warn/Wdangling-reference20.C: New test.

18 months agoxtensa: Make full transition to LRA
Takayuki 'January June' Suwa [Mon, 8 May 2023 13:38:51 +0000 (22:38 +0900)] 
xtensa: Make full transition to LRA

gcc/ChangeLog:

* config/xtensa/constraints.md (R, T, U):
Change define_constraint to define_memory_constraint.
* config/xtensa/predicates.md (move_operand): Don't check that a
constant pool operand size is a multiple of UNITS_PER_WORD.
* config/xtensa/xtensa.cc
(xtensa_lra_p, TARGET_LRA_P): Remove.
(xtensa_emit_move_sequence): Remove "if (reload_in_progress)"
clause as it can no longer be true.
(fixup_subreg_mem): Drop function.
(xtensa_output_integer_literal_parts): Consider 16-bit wide
constants.
(xtensa_legitimate_constant_p): Add short-circuit path for
integer load instructions. Don't check that mode size is
at least UNITS_PER_WORD.
* config/xtensa/xtensa.md (movsf): Use can_create_pseudo_p()
rather reload_in_progress and reload_completed.
(doloop_end): Drop operand 2.
(movhi_internal): Add alternative loading constant from a
literal pool.
(define_split for DI register_operand): Don't limit to
!TARGET_AUTO_LITPOOLS.
* config/xtensa/xtensa.opt (mlra): Change to no effect.

18 months agoc++: add original testcase [PR67898]
Patrick Palka [Tue, 30 Jan 2024 17:30:32 +0000 (12:30 -0500)] 
c++: add original testcase [PR67898]

The original testcase from this PR (fixed by r14-8291) seems rather
different from the others, so let's add it to the testsuite.

PR c++/67898

gcc/testsuite/ChangeLog:

* g++.dg/cpp0x/temp_default8.C: New test.

18 months agotestsuite: fix anon6 mangling [PR112846]
Jason Merrill [Tue, 30 Jan 2024 17:07:21 +0000 (12:07 -0500)] 
testsuite: fix anon6 mangling [PR112846]

As with r14-6796-g2fa122cae50cd8, avoid mangling compatibility aliases in
mangling tests, and test the new mangling.

PR c++/112846

gcc/testsuite/ChangeLog:

* g++.dg/abi/anon6.C: Specify ABI v18.
* g++.dg/abi/anon6a.C: New test for ABI v19.

18 months agotestsuite: mangle-reparm1a options [PR113451]
Jason Merrill [Tue, 30 Jan 2024 16:36:53 +0000 (11:36 -0500)] 
testsuite: mangle-reparm1a options [PR113451]

When I added -fabi-compat-version=8 to avoid mangling aliases it also
suppressed the -Wabi warning.

PR c++/113451

gcc/testsuite/ChangeLog:

* g++.dg/abi/mangle-regparm1a.C: Use -Wabi=0.

18 months agoc++: duplicated side effects of xobj arg [PR113640]
Patrick Palka [Tue, 30 Jan 2024 15:44:56 +0000 (10:44 -0500)] 
c++: duplicated side effects of xobj arg [PR113640]

We miscompile the below testcase because keep_unused_object_arg thinks
the object argument of an xobj member function is unused, and so it ends
up duplicating the argument's side effects.

PR c++/113640

gcc/cp/ChangeLog:

* call.cc (keep_unused_object_arg): Punt for an xobj member
function.

gcc/testsuite/ChangeLog:

* g++.dg/cpp23/explicit-obj-lambda14.C: New test.

Reviewed-by: Jason Merrill <jason@redhat.com>
18 months agoc++: unifying integer parm with type-dep arg [PR113644]
Patrick Palka [Tue, 30 Jan 2024 15:13:41 +0000 (10:13 -0500)] 
c++: unifying integer parm with type-dep arg [PR113644]

Here when trying to unify P=42 A=T::value we ICE due to the latter's
empty type, which same_type_p dislikes.

PR c++/113644

gcc/cp/ChangeLog:

* pt.cc (unify) <case INTEGER_CST>: Handle NULL_TREE type.

gcc/testsuite/ChangeLog:

* g++.dg/template/nontype30.C: New test.

18 months agolibstdc++: Fix check in testsuite/std/time/clock/gps/io.cc
Jonathan Wakely [Mon, 22 Jan 2024 10:48:55 +0000 (10:48 +0000)] 
libstdc++: Fix check in testsuite/std/time/clock/gps/io.cc

The test_format() function contained an incorrect assertion but wasn't
actually being called from main.

libstdc++-v3/ChangeLog:

* testsuite/std/time/clock/gps/io.cc: Fix expected result in
assertion and call test_format() from main.

18 months agoRISC-V: Bugfix for vls mode aggregated in GPR calling convention
Pan Li [Tue, 30 Jan 2024 07:42:06 +0000 (15:42 +0800)] 
RISC-V: Bugfix for vls mode aggregated in GPR calling convention

According to the issue as below.

https://hub.fgit.cf/riscv-non-isa/riscv-elf-psabi-doc/pull/416

When the mode size of vls integer mode is less than 2 * XLEN, we will
take the gpr for both the args and the return values. Instead of the
reference. For example the below code:

typedef short v8hi __attribute__ ((vector_size (16)));

v8hi __attribute__((noinline))
add (v8hi a, v8hi b)
{
  v8hi r = a + b;
  return r;
}

Before this patch:
add:
  vsetivli zero,8,e16,m1,ta,ma
  vle16.v  v1,0(a1) <== arg by reference
  vle16.v  v2,0(a2) <== arg by reference
  vadd.vv  v1,v1,v2
  vse16.v  v1,0(a0) <== return by reference
  ret

After this patch:
add:
  addi     sp,sp,-32
  sd       a0,0(sp)  <== arg by register a0 - a3
  sd       a1,8(sp)
  sd       a2,16(sp)
  sd       a3,24(sp)
  addi     a5,sp,16
  vsetivli zero,8,e16,m1,ta,ma
  vle16.v  v2,0(sp)
  vle16.v  v1,0(a5)
  vadd.vv  v1,v1,v2
  vse16.v  v1,0(sp)
  ld       a0,0(sp)  <== return by a0 - a1.
  ld       a1,8(sp)
  addi     sp,sp,32
  jr       ra

For vls floating point, we take the same rules as integer and passed by
the gpr or reference.

The riscv regression passed for this patch.

gcc/ChangeLog:

* config/riscv/riscv.cc (riscv_v_vls_mode_aggregate_gpr_count): New function to
calculate the gpr count required by vls mode.
(riscv_v_vls_to_gpr_mode): New function convert vls mode to gpr mode.
(riscv_pass_vls_aggregate_in_gpr): New function to return the rtx of gpr
for vls mode.
(riscv_get_arg_info): Add vls mode handling.
(riscv_pass_by_reference): Return false if arg info has no zero gpr count.

gcc/testsuite/ChangeLog:

* gcc.target/riscv/rvv/autovec/vls/def.h: Add new helper macro.
* gcc.target/riscv/rvv/autovec/vls/calling-convention-1.c: New test.
* gcc.target/riscv/rvv/autovec/vls/calling-convention-10.c: New test.
* gcc.target/riscv/rvv/autovec/vls/calling-convention-2.c: New test.
* gcc.target/riscv/rvv/autovec/vls/calling-convention-3.c: New test.
* gcc.target/riscv/rvv/autovec/vls/calling-convention-4.c: New test.
* gcc.target/riscv/rvv/autovec/vls/calling-convention-5.c: New test.
* gcc.target/riscv/rvv/autovec/vls/calling-convention-6.c: New test.
* gcc.target/riscv/rvv/autovec/vls/calling-convention-7.c: New test.
* gcc.target/riscv/rvv/autovec/vls/calling-convention-8.c: New test.
* gcc.target/riscv/rvv/autovec/vls/calling-convention-9.c: New test.
* gcc.target/riscv/rvv/autovec/vls/calling-convention-run-1.c: New test.
* gcc.target/riscv/rvv/autovec/vls/calling-convention-run-2.c: New test.
* gcc.target/riscv/rvv/autovec/vls/calling-convention-run-3.c: New test.
* gcc.target/riscv/rvv/autovec/vls/calling-convention-run-4.c: New test.
* gcc.target/riscv/rvv/autovec/vls/calling-convention-run-5.c: New test.
* gcc.target/riscv/rvv/autovec/vls/calling-convention-run-6.c: New test.

Signed-off-by: Pan Li <pan2.li@intel.com>
18 months agoanalyzer: fix -Wanalyzer-allocation-size false +ve on Linux kernel's round_up macro...
David Malcolm [Tue, 30 Jan 2024 13:17:47 +0000 (08:17 -0500)] 
analyzer: fix -Wanalyzer-allocation-size false +ve on Linux kernel's round_up macro [PR113654]

gcc/analyzer/ChangeLog:
PR analyzer/113654
* region-model.cc (is_round_up): New.
(is_multiple_p): New.
(is_dubious_capacity): New.
(region_model::check_region_size): Move usage of size_visitor into
is_dubious_capacity.

gcc/testsuite/ChangeLog:
PR analyzer/113654
* c-c++-common/analyzer/allocation-size-pr113654-1.c: New test.

Signed-off-by: David Malcolm <dmalcolm@redhat.com>
18 months agoanalyzer: add SARIF property bag to -Wanalyzer-allocation-size
David Malcolm [Tue, 30 Jan 2024 13:17:47 +0000 (08:17 -0500)] 
analyzer: add SARIF property bag to -Wanalyzer-allocation-size

This is useful for debugging the analyzer.

gcc/analyzer/ChangeLog:
* region-model.cc
(dubious_allocation_size::dubious_allocation_size): Add
"capacity_sval" param.  Drop unused ctor.
(dubious_allocation_size::maybe_add_sarif_properties): New.
(dubious_allocation_size::m_capacity_sval): New field.
(region_model::check_region_size): Pass capacity svalue to
dubious_allocation_size ctor.

Signed-off-by: David Malcolm <dmalcolm@redhat.com>
18 months agogccrs: Fix output line ending patterns.
Arthur Cohen [Mon, 22 Jan 2024 13:04:11 +0000 (14:04 +0100)] 
gccrs: Fix output line ending patterns.

gcc/testsuite/ChangeLog:

* rust/execute/torture/builtin_macros1.rs: Fix output pattern.
* rust/execute/torture/coercion3.rs: Likewise.
* rust/execute/torture/issue-2080.rs: Likewise.
* rust/execute/torture/issue-2179.rs: Likewise.
* rust/execute/torture/issue-2180.rs: Likewise.
* rust/execute/torture/iter1.rs: Likewise.

18 months agogccrs: Remove TraitImplItem
Owen Avery [Wed, 20 Dec 2023 08:04:36 +0000 (03:04 -0500)] 
gccrs: Remove TraitImplItem

gcc/rust/ChangeLog:

* ast/rust-ast-full-decls.h
(class TraitImplItem): Remove forward declaration.
(class AssociatedItem): Add forward declaration.
* ast/rust-ast.h
(class TraitImplItem): Remove.
(class TraitItem): Inherit from AssociatedItem.
(SingleASTNode::take_trait_impl_item):
Return std::unique_ptr<AssociatedItem> instead of
std::unique_ptr<TraitImplItem>.
* ast/rust-item.h
(class Function): Inherit from AssociatedItem instead of
TraitImplItem.
(class TypeAlias): Likewise.
(class ConstantItem): Likewise.
(class TraitImpl): Store items as AssociatedItem.
* expand/rust-derive-clone.cc
(DeriveClone::clone_fn): Return std::unique_ptr<AssociatedItem>.
(DeriveClone::clone_impl): Take std::unique_ptr<AssociatedItem>.
* expand/rust-derive-clone.h
(DeriveClone::clone_fn): Return std::unique_ptr<AssociatedItem>.
(DeriveClone::clone_impl): Take std::unique_ptr<AssociatedItem>.
* expand/rust-expand-visitor.cc
(ExpandVisitor::visit): Handle changes to
SingleASTNode::take_trait_impl_item.
* parse/rust-parse-impl.h
(Parser::parse_impl): Parse TraitImpl as containing AssociatedItem.
(Parser::parse_trait_impl_item): Return
std::unique_ptr<AssociatedItem>.
(Parser::parse_trait_impl_function_or_method): Likewise.
* parse/rust-parse.h
(Parser::parse_trait_impl_item): Return
std::unique_ptr<AssociatedItem>.
(Parser::parse_trait_impl_function_or_method): Likewise.

Signed-off-by: Owen Avery <powerboat9.gamer@gmail.com>
18 months agogccrs: Add improved error when no fields in initializer
Robert Goss [Sun, 14 Jan 2024 17:34:22 +0000 (17:34 +0000)] 
gccrs: Add improved error when no fields in initializer

If a struct type with a variant that has fields is initialized with some fields the expression  HIR StructExprStructFields is checked that all the fields are assigned. However, if no fields are initialized the HIR StructExprStruct is generated. This doesn't check if the struct is a unit during typechekc and only fails at the compile stage with a ICE.

Add a check at the typecheck stage that makes sure the struct does not have a variant with fields and give an error message based on the rustc one.

We have also updated the message given in the case where one field was present to list the missing fields and match more closely the new message.

gcc/rust/ChangeLog:

* typecheck/rust-hir-type-check-expr.cc (TypeCheckExpr::visit) Add additional check
* typecheck/rust-hir-type-check-struct-field.h: A helper method to make error added
* typecheck/rust-hir-type-check-struct.cc (TypeCheckStructExpr::resolve) Update message

gcc/testsuite/ChangeLog:

* rust/compile/missing_constructor_fields.rs: Added case with no initializers

Signed-off-by: Robert Goss <goss.robert@gmail.com>
18 months agogccrs: Test: check implemented for lifetime handling
Jakub Dupak [Thu, 18 Jan 2024 13:37:58 +0000 (14:37 +0100)] 
gccrs: Test: check implemented for lifetime handling

gcc/testsuite/ChangeLog:

* rust/compile/for_lifetimes.rs: New test.

Signed-off-by: Jakub Dupak <dev@jakubdupak.com>
18 months agogccrs: AST: Fix for lifetime lowering
Jakub Dupak [Tue, 9 Jan 2024 14:46:41 +0000 (15:46 +0100)] 
gccrs: AST: Fix for lifetime lowering

gcc/rust/ChangeLog:

* hir/rust-ast-lower-type.cc (ASTLoweringTypeBounds::visit): fix for lifetimes
(ASTLowerWhereClauseItem::visit): fix for lifetimes

Signed-off-by: Jakub Dupak <dev@jakubdupak.com>
18 months agogccrs: AST: Fix for lifetime parsing
Jakub Dupak [Tue, 9 Jan 2024 14:45:12 +0000 (15:45 +0100)] 
gccrs: AST: Fix for lifetime parsing

gcc/rust/ChangeLog:

* parse/rust-parse-impl.h (Parser::parse_where_clause): fix parsing
(Parser::parse_where_clause_item): fix parsing
(Parser::parse_type_bound_where_clause_item): fix parsing
(Parser::parse_trait_bound): fix parsing
* parse/rust-parse.h: fix parsing

Signed-off-by: Jakub Dupak <dev@jakubdupak.com>
18 months agogccrs: Test: fix missing lifetime in a test
Jakub Dupak [Tue, 9 Jan 2024 14:43:01 +0000 (15:43 +0100)] 
gccrs: Test: fix missing lifetime in a test

This test did not compile with rustc.

gcc/testsuite/ChangeLog:

* rust/compile/torture/utf8_identifiers.rs: add mising lifetime

Signed-off-by: Jakub Dupak <dev@jakubdupak.com>
18 months agogccrs: Added newline to get more readable lexdump
Kushal Pal [Sat, 30 Dec 2023 04:10:19 +0000 (09:40 +0530)] 
gccrs: Added newline to get more readable lexdump

Fixes #2783

gcc/rust/ChangeLog:

* lex/rust-lex.cc (Lexer::dump_and_skip):
Changed " " to '\n'

Signed-off-by: Kushal Pal <kushalpal109@gmail.com>
18 months agogccrs: Unify storage of associated items in SingleASTNode
Owen Avery [Wed, 20 Dec 2023 07:30:20 +0000 (02:30 -0500)] 
gccrs: Unify storage of associated items in SingleASTNode

gcc/rust/ChangeLog:

* ast/rust-ast-fragment.cc
(Fragment::assert_single_fragment): Update.

* ast/rust-ast.h
(class TraitImplItem): Move definition before that of TraitItem.
(class TraitItem):
Inherit from TraitImplItem instead of AssociatedItem.

(class SingleASTNode): Unify handling of associated items.
(SingleASTNode::take_assoc_item): Move from...
(SingleASTNode::take_impl_item): ...here, but leave stub calling
take_assoc_item behind.
(SingleASTNode::take_trait_item):
Cast associated item to TraitItem.
(SingleASTNode::take_trait_impl_item):
Cast associated item to TraitImplItem.

* ast/rust-ast.cc
(SingleASTNode::SingleASTNode):
Unify handling of associated items.
(SingleASTNode::operator=): Likewise.
(SingleASTNode::accept_vis): Likewise.
(SingleASTNode::is_error): Likewise.
(SingleASTNode::as_string): Likewise.

* ast/rust-item.h
(class Function): Remove direct inheritence from AssociatedItem.
(class ConstantItem): Likewise.
* ast/rust-macro.h
(class MacroInvocation):
Remove direct inheritence from AssociatedItem and TraitImplItem.

18 months agogccrs: Add improved error when a field is redefined in a struct constructor
Robert Goss [Tue, 16 Jan 2024 22:50:51 +0000 (22:50 +0000)] 
gccrs: Add improved error when a field is redefined in a struct constructor

Fixes #2381

If a struct type is initialized with one of it's fields repeated it will currently issue an error at the use site. However we would like to give the rust error code and (like rustc) show both the specifications for the field to help the user diagnose the issue.

We move the check after the index for the field has been established so we can look up if the field has already been defined and get it's location.

We update the visit method to return if it has handled an error otherwise we then output a second fatal error as not all the field in the specification have been processed.

gcc/rust/ChangeLog:

* typecheck/rust-hir-type-check-struct-field.h: Allow visit to return a bool
* typecheck/rust-hir-type-check-struct.cc: Improve check of repeat fields

gcc/testsuite/ChangeLog:

* rust/compile/repeated_constructor_fields.rs: Added case with constructor field repeated

Signed-off-by: Robert Goss <goss.robert@gmail.com>
18 months agogccrs: Handle `async` keyword for regular implementations
Kushal Pal [Sat, 6 Jan 2024 05:59:14 +0000 (11:29 +0530)] 
gccrs: Handle `async` keyword for regular implementations

Fixes #2788

gcc/rust/ChangeLog:

* parse/rust-parse-impl.h (Parser::parse_inherent_impl_item):
Added switch-case for ASYNC token.

gcc/testsuite/ChangeLog:

* rust/compile/issue-2788.rs: New test.

Signed-off-by: Kushal Pal <kushalpal109@gmail.com>
18 months agogccrs: Fix inconsistent formatting
Kushal Pal [Tue, 16 Jan 2024 13:36:35 +0000 (19:06 +0530)] 
gccrs: Fix inconsistent formatting

gcc/rust/ChangeLog:

* checks/errors/rust-ast-validation.cc (ASTValidation::visit):
Enclose const in single quotes.

gcc/testsuite/ChangeLog:

* rust/compile/const_trait_fn.rs:
Enclose const in single quotes.

Signed-off-by: Kushal Pal <kushalpal109@gmail.com>
18 months agogccrs: Handle `async` functions in traits
Kushal Pal [Wed, 3 Jan 2024 10:32:07 +0000 (16:02 +0530)] 
gccrs: Handle `async` functions in traits

Fixes #2785

gcc/rust/ChangeLog:

* checks/errors/rust-ast-validation.cc (ASTValidation::visit):
Added check for `async` functions inside trait.
* parse/rust-parse-impl.h (Parser::parse_trait_item):
Added switch-case for ASYNC token.

gcc/testsuite/ChangeLog:

* rust/compile/issue-2785.rs: New test.

Signed-off-by: Kushal Pal <kushalpal109@gmail.com>
18 months agogccrs: Handle newlines during string parsing while lexing
Nirmal Patel [Wed, 27 Dec 2023 22:49:58 +0000 (17:49 -0500)] 
gccrs: Handle newlines during string parsing while lexing

If newline strings are encountered while lexing, the lexer now handles
newline characters by incrementing current line number. This provides
correct line number when displaying errors. If the lexer encounters end
of file before string end, then it will use the start of the string as
the location to an report error.

gcc/rust/ChangeLog:
* lex/rust-lex.cc (Lexer::parse_byte_string): Handle newline
while parsing byte strings
(Lexer::parse_string): Handle newline while parsing strings

Signed-off-by: Nirmal Patel <nirmal@nirmal.dev>
18 months agogccrs: TyTy: Refactor FnType deprecated API
Jakub Dupak [Thu, 14 Dec 2023 11:54:11 +0000 (12:54 +0100)] 
gccrs: TyTy: Refactor FnType deprecated API

gcc/rust/ChangeLog:

* backend/rust-compile-expr.cc (CompileExpr::visit): Use new API.
* typecheck/rust-tyty-call.cc (TypeCheckCallExpr::visit): Use new API.
* typecheck/rust-tyty-cmp.h: Remove old API.
* typecheck/rust-tyty.cc (FnPtr::is_equal): Use new API.
* typecheck/rust-tyty.h: Remove old API.
* typecheck/rust-unify.cc (UnifyRules::expect_fnptr): Use new API.

Signed-off-by: Jakub Dupak <dev@jakubdupak.com>
18 months agogccrs: ast: Lower 'for' lifetimes
Jakub Dupak [Sun, 3 Dec 2023 11:32:55 +0000 (12:32 +0100)] 
gccrs: ast: Lower 'for' lifetimes

gcc/rust/ChangeLog:

* hir/rust-ast-lower-type.cc (ASTLoweringType::visit): For lifetimes.

Signed-off-by: Jakub Dupak <dev@jakubdupak.com>
18 months agogccrs: ast: Infer static lifetime for const and static items
Jakub Dupak [Sun, 3 Dec 2023 11:30:08 +0000 (12:30 +0100)] 
gccrs: ast: Infer static lifetime for const and static items

(probably incomplete propagation)

gcc/rust/ChangeLog:

* hir/rust-ast-lower-base.cc (ASTLoweringBase::lower_lifetime): Propagate static
requirement.
* hir/rust-ast-lower-base.h: Propagate static requirement.
* hir/rust-ast-lower-implitem.h: Propagate static requirement.
* hir/rust-ast-lower-item.cc (ASTLoweringItem::visit): Propagate static requirement.
* hir/rust-ast-lower-type.cc (ASTLoweringType::translate): Propagate static requirement.
(ASTLoweringType::visit): Propagate static requirement.
* hir/rust-ast-lower-type.h: Propagate static requirement.

Signed-off-by: Jakub Dupak <dev@jakubdupak.com>
18 months agogccrs: ast: Full lifetime elision handling
Jakub Dupak [Sun, 3 Dec 2023 11:28:07 +0000 (12:28 +0100)] 
gccrs: ast: Full lifetime elision handling

gcc/rust/ChangeLog:

* parse/rust-parse-impl.h (Parser::parse_generic_param): Lifetime elision control.
(Parser::parse_lifetime_where_clause_item): Lifetime elision control.
(Parser::parse_type_param_bound): Lifetime elision control.
(Parser::parse_lifetime_bounds): Lifetime elision control.
(Parser::parse_lifetime): Lifetime elision control.
(Parser::parse_path_generic_args): Lifetime elision control.
(Parser::parse_self_param): Lifetime elision control.
(Parser::parse_break_expr): Lifetime elision control.
(Parser::parse_continue_expr): Lifetime elision control.
(Parser::parse_reference_type_inner): Lifetime elision control.
* parse/rust-parse.h: Lifetime elision control.

Signed-off-by: Jakub Dupak <dev@jakubdupak.com>
18 months agogccrs: ast: Unify explicitly and implicitly elided lifettimes
Jakub Dupak [Sun, 3 Dec 2023 11:25:23 +0000 (12:25 +0100)] 
gccrs: ast: Unify explicitly and implicitly elided lifettimes

gcc/rust/ChangeLog:

* ast/rust-ast.h: Elided lifetime static constructor
* ast/rust-type.h: Default lifetime to elided.
* parse/rust-parse-impl.h (Parser::parse_lifetime_param): Use elided lifetime.
(Parser::parse_lifetime): Use elided lifetime/
(Parser::lifetime_from_token): Use elided lifetime.
(Parser::parse_self_param): Use elided lifetime.
(Parser::parse_reference_type_inner): Use elided lifetime.

Signed-off-by: Jakub Dupak <dev@jakubdupak.com>
18 months agogccrs: ast: Fix lifetime type parsing
Jakub Dupak [Sun, 3 Dec 2023 11:23:17 +0000 (12:23 +0100)] 
gccrs: ast: Fix lifetime type parsing

There was a mismatch whether lifetime 'static is parsed as "static"
or "'static".

gcc/rust/ChangeLog:

* parse/rust-parse-impl.h (Parser::lifetime_from_token): Fix matched pattern.

Signed-off-by: Jakub Dupak <dev@jakubdupak.com>
18 months agogccrs: Generate error for `async` trait fucntions
Kushal Pal [Sat, 16 Dec 2023 14:37:23 +0000 (20:07 +0530)] 
gccrs: Generate error for `async` trait fucntions

Fixes #2767

gcc/rust/ChangeLog:

* checks/errors/rust-ast-validation.cc (ASTValidation::visit):
Added check for `async` function inside trait.

gcc/testsuite/ChangeLog:

* rust/compile/issue-2767.rs: New test.

Signed-off-by: Kushal Pal <kushalpal109@gmail.com>
18 months agogccrs: Handle `async` qualifier inside trait
Kushal Pal [Fri, 15 Dec 2023 19:11:09 +0000 (00:41 +0530)] 
gccrs: Handle `async` qualifier inside trait

Fixes #2778

gcc/rust/ChangeLog:

* parse/rust-parse-impl.h (Parser::parse_trait_impl_item):
Handled `async` items

Signed-off-by: Kushal Pal <kushalpal109@gmail.com>
18 months agogccrs: split rust-mangle.cc into two files
Raiki Tamura [Tue, 26 Dec 2023 04:08:42 +0000 (13:08 +0900)] 
gccrs: split rust-mangle.cc into two files

gcc/rust/ChangeLog:

* Make-lang.in: Add .o files
* backend/rust-mangle.cc (struct V0Path): moved to splitted files
(v0_path): Likewise.
(legacy_mangle_name): Likewise.
(legacy_mangle_canonical_path): Likewise.
(legacy_hash): Likewise.
(v0_tuple_prefix): Likewise.
(v0_numeric_prefix): Likewise.
(v0_simple_type_prefix): Likewise.
(v0_complex_type_prefix): Likewise.
(v0_integer_62): Likewise.
(v0_opt_integer_62): Likewise.
(v0_disambiguator): Likewise.
(v0_type_prefix): Likewise.
(v0_generic_args): Likewise.
(v0_identifier): Likewise.
(v0_type_path): Likewise.
(v0_function_path): Likewise.
(v0_scope_path): Likewise.
(v0_crate_path): Likewise.
(v0_inherent_or_trait_impl_path): Likewise.
(v0_closure): Likewise.
(legacy_mangle_item): Likewise.
(v0_mangle_item): Likewise.
* backend/rust-mangle.h (legacy_mangle_item): Likewise.
(v0_mangle_item): Likewise.
* backend/rust-mangle-legacy.cc: New file.
* backend/rust-mangle-v0.cc: New file.

Signed-off-by: Raiki Tamura <tamaron1203@gmail.com>
18 months agogccrs: BIR: Cleanup
Jakub Dupak [Mon, 11 Dec 2023 21:09:07 +0000 (22:09 +0100)] 
gccrs: BIR: Cleanup

gcc/rust/ChangeLog:

* checks/errors/borrowck/rust-bir-place.h: Cleanup.
* checks/errors/borrowck/rust-borrow-checker.h: Cleanup.

Signed-off-by: Jakub Dupak <dev@jakubdupak.com>
18 months agogccrs: TyTy: SubstitutionRef cast specialization
Jakub Dupak [Mon, 11 Dec 2023 22:11:34 +0000 (23:11 +0100)] 
gccrs: TyTy: SubstitutionRef cast specialization

Allows skipping parent check when casting.

gcc/rust/ChangeLog:

* typecheck/rust-tyty.h (BaseType::is): Cast API.
(SubstitutionRef>): Cast API.
(BaseType::as): Cast API.
(BaseType::try_as): Cast API.

Signed-off-by: Jakub Dupak <dev@jakubdupak.com>
18 months agogccrs: TyTy: Common interface for fucntion-like types
Jakub Dupak [Sun, 3 Dec 2023 10:16:59 +0000 (11:16 +0100)] 
gccrs: TyTy: Common interface for fucntion-like types

gcc/rust/ChangeLog:

* typecheck/rust-tyty.h (class ClosureType): Inherit interface.
(class FnPtr): Inherit interface.
(class FnType): Inherit interface.
(class CallableTypeInterface): New interface.
(BaseType::is): Detect interface members API.

Signed-off-by: Jakub Dupak <dev@jakubdupak.com>
18 months agogccrs: TyTy: refactor to new API
Jakub Dupak [Sun, 3 Dec 2023 10:15:06 +0000 (11:15 +0100)] 
gccrs: TyTy: refactor to new API

gcc/rust/ChangeLog:

* typecheck/rust-hir-type-check-type.cc (TypeCheckType::resolve_root_path): Refactor.

Signed-off-by: Jakub Dupak <dev@jakubdupak.com>
18 months agogccrs: BIR: Fix missed nodiscard
Jakub Dupak [Mon, 11 Dec 2023 20:52:48 +0000 (21:52 +0100)] 
gccrs: BIR: Fix missed nodiscard

gcc/rust/ChangeLog:

* checks/errors/borrowck/rust-bir-builder-internal.h: Replace nodiscard.
* checks/errors/borrowck/rust-bir-place.h: Replace nodiscard.

Signed-off-by: Jakub Dupak <dev@jakubdupak.com>
18 months agogccrs: TyTy: Fix missed nodiscard
Jakub Dupak [Sun, 3 Dec 2023 10:13:04 +0000 (11:13 +0100)] 
gccrs: TyTy: Fix missed nodiscard

gcc/rust/ChangeLog:

* typecheck/rust-tyty.h: Fix nodiscard to warn unused.

Signed-off-by: Jakub Dupak <dev@jakubdupak.com>
18 months agogccrs: HIR: add missing getters
Jakub Dupak [Sun, 3 Dec 2023 10:07:04 +0000 (11:07 +0100)] 
gccrs: HIR: add missing getters

gcc/rust/ChangeLog:

* hir/tree/rust-hir-item.h: Ad lifetime getter.
* hir/tree/rust-hir-path.h: Make getter const ref.
* hir/tree/rust-hir.h: Const ref and new getter.

Signed-off-by: Jakub Dupak <dev@jakubdupak.com>
18 months agogccrs: refactor builtins initialization and attributes
Arthur Cohen [Thu, 3 Nov 2022 14:02:37 +0000 (15:02 +0100)] 
gccrs: refactor builtins initialization and attributes

This commit performs builtin initialization in a more "GCC-y" way,
similarly to what the D frontend is doing. This way, we no longer have
to worry about invalid attributes or types when initializing them by
hand.

Also add attributes support through LANG_HOOKS_COMMON_ATTRIBUTE_TABLE
lang hook.

Most of these changes are based on D frontend.

gcc/rust/ChangeLog:

* Make-lang.in (GRS_OBJS): Add rust-attribs.o.
* backend/rust-builtins.cc (builtin_const, builtin_noreturn)
(builtin_novops): Remove.
(BuiltinsContext::lookup_simple_builtin): Adjust.
(BuiltinsContext::setup_overflow_fns): Remove.
(BuiltinsContext::define_function_type): Set builtin type to
errormark so the builtin is considered unavailable.
(BuiltinsContext::setup_math_fns): Remove.
(BuiltinsContext::setup_atomic_fns): Remove.
(build_c_type_nodes): Refactor based on D frontend.
(BuiltinsContext::define_builtin_types): Likewise.
(DEF_PRIMITIVE_TYPE): New.
(DEF_FUNCTION_TYPE_0): New.
(DEF_FUNCTION_TYPE_1): New.
(DEF_FUNCTION_TYPE_2): New.
(DEF_FUNCTION_TYPE_3): New.
(DEF_FUNCTION_TYPE_4): New.
(DEF_FUNCTION_TYPE_5): New.
(DEF_FUNCTION_TYPE_6): New.
(DEF_FUNCTION_TYPE_7): New.
(DEF_FUNCTION_TYPE_8): New.
(DEF_FUNCTION_TYPE_9): New.
(DEF_FUNCTION_TYPE_10): New.
(DEF_FUNCTION_TYPE_11): New.
(DEF_FUNCTION_TYPE_VAR_0): New.
(DEF_FUNCTION_TYPE_VAR_1): New.
(DEF_FUNCTION_TYPE_VAR_2): New.
(DEF_FUNCTION_TYPE_VAR_3): New.
(DEF_FUNCTION_TYPE_VAR_4): New.
(DEF_FUNCTION_TYPE_VAR_5): New.
(DEF_FUNCTION_TYPE_VAR_6): New.
(DEF_FUNCTION_TYPE_VAR_7): New.
(DEF_FUNCTION_TYPE_VAR_11): New.
(DEF_POINTER_TYPE): New.
(BuiltinsContext::setup): Adjust.
(BuiltinsContext::define_builtin_attributes): New.
(DEF_ATTR_NULL_TREE): New.
(DEF_ATTR_INT): New.
(DEF_ATTR_STRING): New.
(DEF_ATTR_IDENT): New.
(DEF_ATTR_TREE_LIST): New.
(handle_flags): Remove.
(BuiltinsContext::define_builtins): New.
(DEF_BUILTIN): New.
(BuiltinsContext::define_builtin): Remove.
(BuiltinsContext::register_rust_mappings): New. Add all missing
builtins.
(BuiltinsContext::lookup_gcc_builtin): Adjust.
* backend/rust-builtins.h (DEF_PRIMITIVE_TYPE): New.
(DEF_FUNCTION_TYPE_0): New.
(DEF_FUNCTION_TYPE_1): New.
(DEF_FUNCTION_TYPE_2): New.
(DEF_FUNCTION_TYPE_3): New.
(DEF_FUNCTION_TYPE_4): New.
(DEF_FUNCTION_TYPE_5): New.
(DEF_FUNCTION_TYPE_6): New.
(DEF_FUNCTION_TYPE_7): New.
(DEF_FUNCTION_TYPE_8): New.
(DEF_FUNCTION_TYPE_9): New.
(DEF_FUNCTION_TYPE_10): New.
(DEF_FUNCTION_TYPE_11): New.
(DEF_FUNCTION_TYPE_VAR_0): New.
(DEF_FUNCTION_TYPE_VAR_1): New.
(DEF_FUNCTION_TYPE_VAR_2): New.
(DEF_FUNCTION_TYPE_VAR_3): New.
(DEF_FUNCTION_TYPE_VAR_4): New.
(DEF_FUNCTION_TYPE_VAR_5): New.
(DEF_FUNCTION_TYPE_VAR_6): New.
(DEF_FUNCTION_TYPE_VAR_7): New.
(DEF_FUNCTION_TYPE_VAR_11): New.
(DEF_POINTER_TYPE): New.
(DEF_ATTR_NULL_TREE): New.
(DEF_ATTR_INT): New.
(DEF_ATTR_STRING): New.
(DEF_ATTR_IDENT): New.
(DEF_ATTR_TREE_LIST): New.
* backend/rust-compile-intrinsic.cc (Intrinsics::compile): Add
comment.
(op_with_overflow_inner): Adjust.
(copy_handler_inner): Adjust.
(prefetch_data_handler): Adjust.
(build_atomic_builtin_name): Adjust.
(atomic_load_handler_inner): Adjust.
(uninit_handler): Adjust.
(move_val_init_handler): Adjust.
(expect_handler_inner): Adjust.
* rust-gcc.cc (fetch_overflow_builtins): Adjust.
* rust-lang.cc (rust_localize_identifier): Adjust.
(LANG_HOOKS_COMMON_ATTRIBUTE_TABLE): New.
* rust-attribs.cc: New file.

gcc/testsuite/ChangeLog:

* rust/compile/torture/intrinsics-4.rs: Adjust.
* rust/compile/torture/intrinsics-math.rs: Adjust.
* rust/execute/torture/atomic_load.rs: Adjust.
* rust/execute/torture/atomic_store.rs: Adjust.
* rust/compile/torture/intrinsics-1.rs: Removed.
* rust/compile/torture/builtin_abort.rs: New test.
* rust/execute/torture/builtin_abort.rs: New test.

Signed-off-by: Marc Poulhiès <dkm@kataplop.net>
Co-authored-by: Arthur Cohen <arthur.cohen@embecosm.com>
18 months agogccrs: Change cfg stripper to use default visitor
Pierre-Emmanuel Patry [Thu, 23 Nov 2023 19:22:29 +0000 (20:22 +0100)] 
gccrs: Change cfg stripper to use default visitor

Cfg visitor used it's own visit patterns, this made the code longer than
required.

gcc/rust/ChangeLog:

* expand/rust-cfg-strip.cc (CfgStrip::visit): Change calls from visitor
to default visitor.
(CfgStrip::go): Add call to visit crate.
* expand/rust-cfg-strip.h (class CfgStrip): Update prototypes and
remove empty ones.
* ast/rust-ast-visitor.cc: add WhereClause condition check.

Signed-off-by: Pierre-Emmanuel Patry <pierre-emmanuel.patry@embecosm.com>
18 months agogccrs: Make expand visitor inherit from default visitor
Pierre-Emmanuel Patry [Mon, 27 Nov 2023 13:12:20 +0000 (14:12 +0100)] 
gccrs: Make expand visitor inherit from default visitor

Many visit functions in the expand visitor simply visit their components
like the default visitor. Making the expand visitor inherit from the
default visitor allows us to keep all visitor in sync without having to
change every visitor.

gcc/rust/ChangeLog:

* expand/rust-expand-visitor.cc (ExpandVisitor::go): Add call to visit
on the crate.
(ExpandVisitor::visit): Remove some visit functions in favor of their
default visitor counterpart.
* expand/rust-expand-visitor.h (class ExpandVisitor): Inherit from
default visitor and remove now useless function prototypes.

Signed-off-by: Pierre-Emmanuel Patry <pierre-emmanuel.patry@embecosm.com>
18 months agogccrs: Make default resolver inherit from default visitor
Pierre-Emmanuel Patry [Mon, 27 Nov 2023 16:00:56 +0000 (17:00 +0100)] 
gccrs: Make default resolver inherit from default visitor

The default resolver put some scope in place but mostly has traversal
functions similar to the default ast visitor, making it inherit from the
default visitor allows us to avoid code duplication.

gcc/rust/ChangeLog:

* resolve/rust-default-resolver.cc (DefaultResolver::visit): Remove
duplicated functions.
* resolve/rust-default-resolver.h (class DefaultResolver): Make the
default resolver inherit from the default visitor.

Signed-off-by: Pierre-Emmanuel Patry <pierre-emmanuel.patry@embecosm.com>
18 months agogccrs: Allow enabling lang_items and no_core features
Owen Avery [Sat, 9 Dec 2023 06:22:02 +0000 (01:22 -0500)] 
gccrs: Allow enabling lang_items and no_core features

gcc/rust/ChangeLog:

* checks/errors/rust-feature.cc
(Feature::name_hash_map):
Add entries for Name::LANG_ITEMS and Name::NO_CORE.
* checks/errors/rust-feature.h
(Feature::Name::LANG_ITEMS): New.
(Feature::Name::NO_CORE): New.

gcc/testsuite/ChangeLog:

* rust/compile/sized-stub.rs: New test.

Signed-off-by: Owen Avery <powerboat9.gamer@gmail.com>
18 months agogccrs: Renamed `WIN64` to `WIN_64`
Kushal Pal [Sat, 9 Dec 2023 06:34:01 +0000 (12:04 +0530)] 
gccrs: Renamed `WIN64` to `WIN_64`

Fixes issue #2768

gcc/rust/ChangeLog:

* backend/rust-compile-base.cc (HIRCompileBase::setup_abi_options):
Renamed `WIN64` to `WIN_64`
* util/rust-abi.cc (get_abi_from_string): Likewise
(get_string_from_abi): Likewise
* util/rust-abi.h (enum ABI): Likewise

Signed-off-by: Kushal Pal <kushalpal109@gmail.com>
18 months agogccrs: Generate error for const trait functions
Nobel Singh [Tue, 5 Dec 2023 12:27:29 +0000 (18:12 +0545)] 
gccrs: Generate error for const trait functions

Fixes issue #2040

Add check to assure that a function cant be declared const inside trait impl
blocks.

gcc/rust/ChangeLog:

* checks/errors/rust-ast-validation.cc (ASTValidation::visit): Add
check for const funtion.

gcc/testsuite/ChangeLog:

* rust/compile/issue-2040.rs: New test.

Signed-off-by: Nobel Singh <nobel2073@gmail.com>
18 months agogccrs: Add a regression test for function body check
Pierre-Emmanuel Patry [Thu, 23 Nov 2023 13:08:50 +0000 (14:08 +0100)] 
gccrs: Add a regression test for function body check

Function body check emit an error message when a required function body
is missing.

gcc/testsuite/ChangeLog:

* rust/compile/functions_without_body.rs: New test.

Signed-off-by: Pierre-Emmanuel Patry <pierre-emmanuel.patry@embecosm.com>
18 months agogccrs: Add validation for functions without body
Pierre-Emmanuel Patry [Wed, 22 Nov 2023 15:37:17 +0000 (16:37 +0100)] 
gccrs: Add validation for functions without body

Add checks in the ast validation pass to error out with functions
(either free or associated) without a definition.

gcc/rust/ChangeLog:

* checks/errors/rust-ast-validation.cc (ASTValidation::visit): Add
a validation check and emit an error depending on the context.

Signed-off-by: Pierre-Emmanuel Patry <pierre-emmanuel.patry@embecosm.com>
18 months agogccrs: Make function bodies truly optional
Pierre-Emmanuel Patry [Wed, 22 Nov 2023 14:15:29 +0000 (15:15 +0100)] 
gccrs: Make function bodies truly optional

Missing body on a function should be rejected at a later stage in the
compiler, not during parsing.

gcc/rust/ChangeLog:

* ast/rust-ast-collector.cc (TokenCollector::visit): Adapt defintion
getter.
* ast/rust-ast-visitor.cc (DefaultASTVisitor::visit): Likewise.
* expand/rust-cfg-strip.cc (CfgStrip::visit): Likewise.
* expand/rust-expand-visitor.cc (ExpandVisitor::visit): Likewise.
* hir/rust-ast-lower-implitem.h: Likewise.
* hir/rust-ast-lower-item.cc (ASTLoweringItem::visit): Likewise.
* resolve/rust-ast-resolve-item.cc (ResolveItem::visit): Likewise.
* resolve/rust-ast-resolve-stmt.h: Likewise.
* resolve/rust-default-resolver.cc (DefaultResolver::visit): Likewise.
* util/rust-attributes.cc (AttributeChecker::visit):  Likewise.
* parse/rust-parse-impl.h: Allow empty function body during parsing.
* ast/rust-ast.cc (Function::Function): Constructor now take an
optional for the body.
(Function::operator=): Adapt to new optional member.
(Function::as_string): Likewise.
* ast/rust-item.h (class Function): Make body optional and do not
rely on nullptr anymore.

Signed-off-by: Pierre-Emmanuel Patry <pierre-emmanuel.patry@embecosm.com>
18 months agogccrs: Add execution test for name resolution 2.0
Pierre-Emmanuel Patry [Thu, 30 Nov 2023 12:27:01 +0000 (13:27 +0100)] 
gccrs: Add execution test for name resolution 2.0

We already have some compile tests but it lacked an execution test to
ensure everything compiles correctly to the correct value.

gcc/testsuite/ChangeLog:

* rust/execute/torture/name_resolution.rs: New test.

Signed-off-by: Pierre-Emmanuel Patry <pierre-emmanuel.patry@embecosm.com>
18 months agogccrs: Add multiple regression test in name resolution
Pierre-Emmanuel Patry [Thu, 30 Nov 2023 09:43:36 +0000 (10:43 +0100)] 
gccrs: Add multiple regression test in name resolution

Local variables and functions or global variables have different
resolution when binded to a variable. This was not covered before, even
though it was handled by the new name resolution. This commit highlight
this behavior from the new name resolution mechanism.

gcc/testsuite/ChangeLog:

* rust/compile/name_resolution11.rs: New test.
* rust/compile/name_resolution12.rs: New test.
* rust/compile/name_resolution13.rs: New test.

Signed-off-by: Pierre-Emmanuel Patry <pierre-emmanuel.patry@embecosm.com>
18 months agogccrs: Make early name resolver inherit from default one
Pierre-Emmanuel Patry [Mon, 27 Nov 2023 14:39:24 +0000 (15:39 +0100)] 
gccrs: Make early name resolver inherit from default one

Many visit functions in the early name resolver are just plain old
traversal visit functions like the ones from the default visitor.

gcc/rust/ChangeLog:

* resolve/rust-early-name-resolver.cc (EarlyNameResolver::resolve_generic_args):
Move function.
(EarlyNameResolver::resolve_qualified_path_type): Likewise.
(EarlyNameResolver::visit): Add a top level visit function for crate
and remove duplicated code.
* resolve/rust-early-name-resolver.h (class EarlyNameResolver): Update
overriden function list.

Signed-off-by: Pierre-Emmanuel Patry <pierre-emmanuel.patry@embecosm.com>
18 months agogccrs: Change the attribute checker visitor to default one
Pierre-Emmanuel Patry [Fri, 24 Nov 2023 13:09:23 +0000 (14:09 +0100)] 
gccrs: Change the attribute checker visitor to default one

Make the attribute checker visitor inherit from the default visitor
in order to keep visit behavior shared.

gcc/rust/ChangeLog:

* util/rust-attributes.cc (AttributeChecker::visit): Add visit function
for crates.
* util/rust-attributes.h (class AttributeChecker): Update function
prototypes.

Signed-off-by: Pierre-Emmanuel Patry <pierre-emmanuel.patry@embecosm.com>
18 months agogccrs: Make feature gate visitor inherit from default one
Pierre-Emmanuel Patry [Fri, 24 Nov 2023 10:17:52 +0000 (11:17 +0100)] 
gccrs: Make feature gate visitor inherit from default one

The feature gating behavior may be shortened and kept cleaner using the
default visitor. This means less maintenance on visit functions as the
traversal is shared by multiple visitors.

gcc/rust/ChangeLog:

* checks/errors/rust-feature-gate.cc (FeatureGate::visit): Add a visit
function for the crate level.
(FeatureGate::check): Add call to crate visit.
* checks/errors/rust-feature-gate.h (class FeatureGate): Remove now
useless visit functions (traversal only).

Signed-off-by: Pierre-Emmanuel Patry <pierre-emmanuel.patry@embecosm.com>
18 months agogccrs: Add regression test for const fn in trait
Pierre-Emmanuel Patry [Wed, 22 Nov 2023 12:22:42 +0000 (13:22 +0100)] 
gccrs: Add regression test for const fn in trait

Const fn declaration in trait declarations shall emit an error. This new
test highlight this behavior.

gcc/testsuite/ChangeLog:

* rust/compile/const_trait_fn.rs: New test.

Signed-off-by: Pierre-Emmanuel Patry <pierre-emmanuel.patry@embecosm.com>
18 months agogccrs: Add AST validation check for const in trait
Pierre-Emmanuel Patry [Wed, 22 Nov 2023 12:08:13 +0000 (13:08 +0100)] 
gccrs: Add AST validation check for const in trait

Add a new check in AST validation pass that checks that no function
declaration in traits are declared const.

gcc/rust/ChangeLog:

* checks/errors/rust-ast-validation.cc (ASTValidation::visit): Add
const check.
* checks/errors/rust-ast-validation.h: Add visit function prototype.

Signed-off-by: Pierre-Emmanuel Patry <pierre-emmanuel.patry@embecosm.com>
18 months agogccrs: Add a regression test for async const functions
Pierre-Emmanuel Patry [Wed, 22 Nov 2023 10:12:05 +0000 (11:12 +0100)] 
gccrs: Add a regression test for async const functions

Functions that are both async and const shall be rejected during the
AST validation pass. This new test highlight this behavior.

gcc/testsuite/ChangeLog:

* rust/compile/const_async_function.rs: New test.

Signed-off-by: Pierre-Emmanuel Patry <pierre-emmanuel.patry@embecosm.com>
18 months agogccrs: Add async const function ast validation pass
Pierre-Emmanuel Patry [Wed, 22 Nov 2023 09:44:08 +0000 (10:44 +0100)] 
gccrs: Add async const function ast validation pass

Add a check during AST validation pass to ensure functions are either
const or async but not both.

gcc/rust/ChangeLog:

* checks/errors/rust-ast-validation.cc (ASTValidation::visit): Add
async const check.

Signed-off-by: Pierre-Emmanuel Patry <pierre-emmanuel.patry@embecosm.com>
18 months agogccrs: Allow const and async specifiers in functions
Pierre-Emmanuel Patry [Wed, 22 Nov 2023 09:35:44 +0000 (10:35 +0100)] 
gccrs: Allow const and async specifiers in functions

We need to account for const specifiers in async parsing as const
can be used in the syntax before the async keyword.

gcc/rust/ChangeLog:

* parse/rust-parse-impl.h (Parser::parse_vis_item): Allow parsing async
items in const.
(Parser::parse_async_item): Account for const offset during async
lookahead.

Signed-off-by: Pierre-Emmanuel Patry <pierre-emmanuel.patry@embecosm.com>
18 months agogccrs: Split async and const function qualifiers
Pierre-Emmanuel Patry [Tue, 21 Nov 2023 09:29:35 +0000 (10:29 +0100)] 
gccrs: Split async and const function qualifiers

A function cannot be both async and const, however this should not be
handled in the parser but rather at a later stage in the compiler. This
commit change the AsyncConstStatus in the AST and the HIR to allows a
function to be both async and const.

gcc/rust/ChangeLog:

* ast/rust-ast-builder.cc (AstBuilder::fn_qualifiers): Change
constructor to match the new arguments.
* ast/rust-ast-collector.cc (TokenCollector::visit): Change behavior
to handle both const and async specifiers at the same time.
* ast/rust-ast.cc (FunctionQualifiers::as_string): Likewise.
* ast/rust-item.h (class FunctionQualifiers): Remove AsyncConstStatus
and replace it with both Async and Const status. Also change the safety
arguments to use an enum instead of a boolean.
* hir/rust-ast-lower-base.cc (ASTLoweringBase::lower_qualifiers):
Update constructor call.
* hir/tree/rust-hir-item.h: Add Const and Async status, remove
AsyncConstStatus, update the constructor.
* hir/tree/rust-hir.cc (FunctionQualifiers::as_string): Update with
the new status.
* parse/rust-parse-impl.h (Parser::parse_function_qualifiers): Update
constructor call.
* util/rust-common.h (enum Mutability): Make an enum class.
(enum class): Add Async and Const enum class to avoid booleans.
(enum Unsafety): Change to an enum class.

Signed-off-by: Pierre-Emmanuel Patry <pierre-emmanuel.patry@embecosm.com>
18 months agogccrs: Remove class AST::InherentImplItem
Owen Avery [Tue, 28 Nov 2023 06:07:00 +0000 (01:07 -0500)] 
gccrs: Remove class AST::InherentImplItem

gcc/rust/ChangeLog:

* ast/rust-ast-full-decls.h
(class InherentImplItem): Remove.
* ast/rust-ast.h
(class InherentImplItem): Remove.
(class SingleASTNode):
Store pointer to AssociatedItem instead of InherentImplItem.

* ast/rust-ast.cc
(SingleASTNode::SingleASTNode):
Use clone_associated_item instead of clone_inherent_impl_item.
(SingleASTNode::operator=): Likewise.

* ast/rust-item.h
(class InherentImpl):
Use AssociatedItem rather than InherentImplItem.
(class Function): Likewise.
(class ConstantItem): Likewise.
* ast/rust-macro.h
(class MacroInvocation): Likewise.
* expand/rust-expand-visitor.cc
(ExpandVisitor::visit): Likewise.
* parse/rust-parse-impl.h
(Parser::parse_impl): Likewise.
(Parser::parse_inherent_impl_item): Likewise.
(Parser::parse_inherent_impl_function_or_method): Likewise.
* parse/rust-parse.h
(Parser::parse_inherent_impl_item): Likewise.
(Parser::parse_inherent_impl_function_or_method): Likewise.

Signed-off-by: Owen Avery <powerboat9.gamer@gmail.com>
18 months agogccrs: Remove backend dependancy on resolution rib information
Philip Herron [Mon, 2 Oct 2023 17:41:33 +0000 (18:41 +0100)] 
gccrs: Remove backend dependancy on resolution rib information

When making more desugaring for the HIR we can need to add new Let bindings
which will require namesolution information but also rib information for
which block the let binding is associated which was very unnessecary. This
patch simply updates the BLOCK_CONTEXT of the current scope as we are
compiling and encounter HIR::LetStmts instead of trying to do it all
upfront which lots of wierd checks

gcc/rust/ChangeLog:

* backend/rust-compile-base.cc (HIRCompileBase::compile_locals_for_block): removed
* backend/rust-compile-base.h: update header
* backend/rust-compile-block.cc (CompileBlock::visit): remove old logic
* backend/rust-compile-expr.cc (CompileExpr::generate_closure_function): likewise
* backend/rust-compile-stmt.cc (CompileStmt::visit): likewise
* backend/rust-compile-var-decl.h: ensure we setup tuple bindings correctly

Signed-off-by: Philip Herron <herron.philip@googlemail.com>
18 months agogccrs: Add a regression test for unsafe module validation
Pierre-Emmanuel Patry [Tue, 21 Nov 2023 18:34:36 +0000 (19:34 +0100)] 
gccrs: Add a regression test for unsafe module validation

Add a new test to check for unsafe modules during AST validation pass.

gcc/testsuite/ChangeLog:

* rust/compile/unsafe_module.rs: New test.

Signed-off-by: Pierre-Emmanuel Patry <pierre-emmanuel.patry@embecosm.com>
18 months agogccrs: Emit an error on unsafe modules
Pierre-Emmanuel Patry [Tue, 21 Nov 2023 17:51:21 +0000 (18:51 +0100)] 
gccrs: Emit an error on unsafe modules

An error should be emitted on unsafe modules during the AST validation
pass as the syntax allows those even though they're not alowed later down
the line.

gcc/rust/ChangeLog:

* ast/rust-item.h: Add safety getter to modules.
* checks/errors/rust-ast-validation.cc (ASTValidation::visit): Check
a module's safety and emit an error when meeting an unsafe module.
* checks/errors/rust-ast-validation.h: Add function prototype.
* parse/rust-parse-impl.h (Parser::parse_module): Move the module locus
to the first token instead of the mod keyword.

Signed-off-by: Pierre-Emmanuel Patry <pierre-emmanuel.patry@embecosm.com>
18 months agogccrs: Parse module safety
Pierre-Emmanuel Patry [Tue, 21 Nov 2023 17:28:30 +0000 (18:28 +0100)] 
gccrs: Parse module safety

Unsafe module are rejected at a later stage but could not be parsed
properly. This commit changes the parser in order to allow unsafe module
in the AST.

gcc/rust/ChangeLog:

* parse/rust-parse-impl.h (Parser::parse_vis_item): Dispatch to parse
module when meeting an unsafe module.
(Parser::parse_module): Set unsafe status when the parser encounter an
unsafe keyword.

Signed-off-by: Pierre-Emmanuel Patry <pierre-emmanuel.patry@embecosm.com>