Artem Dergachev [Sat, 20 Aug 2016 17:35:53 +0000 (17:35 +0000)]
[analyzer] Use faster hashing (MD5) in CloneDetector.
This replaces the old approach of fingerprinting every AST node into a string,
which avoided collisions and was simple to implement, but turned out to be
extremely ineffective with respect to both performance and memory.
The collisions are now dealt with in a separate pass, which no longer causes
performance problems because collisions are rare.
Benjamin Kramer [Sat, 20 Aug 2016 16:51:33 +0000 (16:51 +0000)]
[Sema] Don't crash on scanf on forward-declared enums.
This is valid in GNU C, which allows pointers to incomplete enums. GCC
just pretends that the underlying type is 'int' in those cases, follow
that behavior.
Artem Dergachev [Sat, 20 Aug 2016 10:06:59 +0000 (10:06 +0000)]
[analyzer] Make CloneDetector consider macro expansions.
So far macro-generated code was treated by the CloneDetector as normal code.
This caused that some macros where reported as false-positive clones because
large chunks of code coming from otherwise concise macro expansions were treated
as copy-pasted code.
This patch ensures that macros are treated in the same way as literals/function
calls. This prevents macros that expand into multiple statements
from being reported as clones.
Manman Ren [Sat, 20 Aug 2016 00:04:21 +0000 (00:04 +0000)]
[NFC] Add a class ObjCProtocolQualifiers to wrap APIs for ObjC protocol list.
This is in preparation of adding a new type class ObjCTypeParamType that
can take protocol qualifiers. ObjCProtocolQualifiers will be shared between
ObjCObjectType and ObjCTypeParamType.
Vedant Kumar [Fri, 19 Aug 2016 23:15:35 +0000 (23:15 +0000)]
[Driver] Remove unused #include of llvm/Support/config.h
This is a follow-up to r279112 (which removed the need for the header)
and in the same spirit as r279035 (which attempted to un-break
out-of-tree builds).
Richard Smith [Fri, 19 Aug 2016 01:43:06 +0000 (01:43 +0000)]
C++ Modules TS: support parsing the 'module' declaration (including extensions
from p0273r0 approved by EWG). We'll eventually need to handle this from the
lexer as well, in order to disallow preprocessor directives preceding the
module declaration and to support macro import.
Devin Coughlin [Fri, 19 Aug 2016 01:05:31 +0000 (01:05 +0000)]
[analyzer] Weaken assertion in trackNullOrUndefValue()
We should ignore paren casts when making sure that the semantic expression
in a PseudoObjectExpr for an ObjC getter is a message send.
This has no other intended functionality change.
Adding a test for this exposed an interesting issue in another test case
that only manifests under ARC. trackNullOrUndefValue() is not properly
suppressing for nil values that are the result of nil propagation from a nil
receiver when the nil is returned from a function. I've added a FIXME for that
missing suppression.
Justin Lebar [Thu, 18 Aug 2016 20:43:13 +0000 (20:43 +0000)]
[CUDA] Improve handling of math functions.
Summary:
A bunch of related changes here to our CUDA math headers.
- The second arg to nexttoward is a double (well, technically, long
double, but we don't have that), not a float.
- Add a forward-declare of llround(float), which is defined in the CUDA
headers. We need this for the same reason we need most of the other
forward-declares: To prevent a constexpr function in our standard
library from becoming host+device.
- Add nexttowardf implementation.
- Pull "foobarf" functions defined by the CUDA headers in the global
namespace into namespace std. This lets you do e.g. std::sinf.
- Add overloads for math functions accepting integer types. This lets
you do e.g. std::sin(0) without having an ambiguity between the
overload that takes a float and the one that takes a double.
With these changes, we pass testcases derived from libc++ for cmath and
math.h. We can check these testcases in to the test-suite once support
for CUDA lands there.
Yaxun Liu [Thu, 18 Aug 2016 19:34:04 +0000 (19:34 +0000)]
[OpenCL] AMDGCN: Fix size_t type
Pointers of certain GPUs in AMDGCN target in private address space is 32 bit but pointers in other address spaces are 64 bit. size_t type should be defined as 64 bit for these GPUs so that it could hold pointers in all address spaces. Also fixed issues in pointer arithmetic codegen by using pointer specific intptr type.
Zachary Turner [Thu, 18 Aug 2016 19:31:48 +0000 (19:31 +0000)]
Resubmit "[Tooling] Parse compilation database command lines on Windows."
This patch introduced the ability to decide at runtime whether to parse
JSON compilation database command lines using Gnu syntax or Windows
syntax. However, there were many existing unit tests written that
hardcoded Gnu-specific paths. These tests were now failing because
the auto-detection logic was choosing to parse them using Windows
rules.
This resubmission of the patch fixes this by introducing an enum
which defines the syntax mode, which defaults to auto-detect, but
for which the unit tests force Gnu style parsing.
Reid Kleckner [Thu, 18 Aug 2016 18:45:07 +0000 (18:45 +0000)]
[MS] Silence -Wextern-init on const selectany variables
In C, 'extern' is typically used to avoid tentative definitions when
declaring variables in headers, but adding an intializer makes it a
defintion. This is somewhat confusing, so GCC and Clang both warn on it.
In C++, 'extern' is often used to give implictly static 'const'
variables external linkage, so don't warn in that case. If selectany is
present, this might be header code intended for C and C++ inclusion, so
apply the C++ rules.
Richard Smith [Thu, 18 Aug 2016 18:22:22 +0000 (18:22 +0000)]
Use __has_include rather than a configure-time macro to determine if
<sys/resource.h> is available. This should fix out-of-tree builds, at the cost
of not providing the higher rlimits to stage 1 clang when built with an old
host compiler not implementing this feature yet (bootstrap builds should be
fine, though).
Manman Ren [Thu, 18 Aug 2016 17:42:15 +0000 (17:42 +0000)]
Module: add -fprebuilt-module-path to support loading prebuilt modules.
In this mode, there is no need to load any module map and the programmer can
simply use "@import" syntax to load the module directly from a prebuilt
module path. When loading from prebuilt module path, we don't support
rebuilding of the module files and we ignore compatible configuration
mismatches.
This complements the clang_getSkippedRanges function which returns skipped ranges filtered by a specific file.
This function is useful when all the ranges are desired (and a lot more efficient than the equivalent of asking for the ranges file by file, since the implementation of clang_getSkippedRanges iterates over all ranges anyway).
Artem Dergachev [Thu, 18 Aug 2016 12:29:41 +0000 (12:29 +0000)]
[analyzer] Teach CloneDetector to find clones that look like copy-paste errors.
The original clone checker tries to find copy-pasted code that is exactly
identical to the original code, up to minor details.
As an example, if the copy-pasted code has all references to variable 'a'
replaced with references to variable 'b', it is still considered to be
an exact clone.
The new check finds copy-pasted code in which exactly one variable seems
out of place compared to the original code, which likely indicates
a copy-paste error (a variable was forgotten to be renamed in one place).
Vedant Kumar [Thu, 18 Aug 2016 06:43:07 +0000 (06:43 +0000)]
[Driver] Use llvm-config.h, not config.h to unbreak out-of-tree builds
llvm/Config/config.h has intentionally been excluded from llvm
installations (see: llvm/CMakeLists.txt). Un-break out-of-tree builds
post-r278882 by switching to llvm-config.h, which is exported.
Richard Smith [Thu, 18 Aug 2016 01:16:55 +0000 (01:16 +0000)]
PR28438: Update the information on an identifier with local definitions before
trying to write out its macro graph, in case we imported a module that added
another module macro between the most recent local definition and the end of
the module.
Chris Bieneman [Wed, 17 Aug 2016 21:54:30 +0000 (21:54 +0000)]
[Darwin] Stop linking libclang_rt.eprintf.a
Summary:
The eprintf library was added before the general OS X builtins library existed as a place to store one builtin function. Since we have for several years had an actual mandated builtin library for OS X > 10.5, we should just merge eprintf into the main library.
This change will resolve PR28855.
As a follow up I'll also patch compiler-rt to not generate the eprintf library anymore.
Richard Smith [Wed, 17 Aug 2016 21:41:45 +0000 (21:41 +0000)]
PR18417: Increase -ftemplate-depth to the value 1024 recommended by the C++
standard's Annex B. We now attempt to increase the process's stack rlimit to
8MiB on startup, which appears to be enough to allow this to work reliably.
(And if it turns out not to be, we can investigate increasing it further.)
After the introduction of windows command line parsing, some unit tests
began failing that expect to test gnu style command line quirks. The
fix is mechanical but time consuming, so revert this for now.
Zachary Turner [Wed, 17 Aug 2016 20:04:35 +0000 (20:04 +0000)]
[Tooling] Parse compilation database command lines on Windows.
When a compilation database is used on Windows, the command lines cannot
be parsed using the standard GNU style syntax. LLVM provides functions for
parsing Windows style command lines, so use them where appropriate.
After this patch, clang-tidy runs correctly on Windows.
Adrian Prantl [Wed, 17 Aug 2016 18:27:24 +0000 (18:27 +0000)]
Module debug info: Fix a bug in handling record decls without fields.
The previous condition would erroneously mark all CXXRecordDecls
that didn't have any fields as being defined in a clang module.
This patch fixes the condition to only apply to explicit template
instantiations.
Artem Dergachev [Wed, 17 Aug 2016 16:02:45 +0000 (16:02 +0000)]
[analyzer] Add a checker for loss of sign or precision in integral casts.
This new checker tries to find execution paths on which implicit integral casts
cause definite loss of information: a certainly-negative integer is converted
to an unsigned integer, or an integer is definitely truncated to fit into
a smaller type.
Being implicit, such casts are likely to produce unexpected results.
Artem Dergachev [Wed, 17 Aug 2016 15:37:52 +0000 (15:37 +0000)]
[analyzer] Add LocationContext information to SymbolMetadata.
Like SymbolConjured, SymbolMetadata also needs to be uniquely
identified by the moment of its birth.
Such moments are coded by the (Statement, LocationContext, Block count) triples.
Each such triple represents the moment of analyzing a statement with a certain
call backtrace, with corresponding CFG block having been entered a given amount
of times during analysis of the current code body.
The LocationContext information was accidentally omitted for SymbolMetadata,
which leads to reincarnation of SymbolMetadata upon re-entering a code body
with a different backtrace; the new symbol is incorrectly unified with
the old symbol, which leads to unsound assumptions.
CodeGen: Avoid dereferencing end() in ScalarExprEmitter::EmitOverflowCheckedBinOp
Use BB.getNextNode(), which returns nullptr on end(), instead of
&*BB.getIterator(), which is UB on end().
CodeGenFunction::createBasicBlock expects nullptr in this case already.
Richard Smith [Wed, 17 Aug 2016 01:05:07 +0000 (01:05 +0000)]
If possible, set the stack rlimit to at least 8MiB on cc1 startup, and work
around a Linux kernel bug where the actual amount of available stack may be a
*lot* lower than the rlimit.
GCC also sets a higher stack rlimit on startup, but it goes all the way to
64MiB. We can increase this limit if it proves necessary.
The kernel bug is as follows: Linux kernels prior to version 4.1 may choose to
map the process's heap as little as 128MiB before the process's stack for a PIE
binary, even in a 64-bit virtual address space. This means that allocating more
than 128MiB before you reach the process's stack high water mark can lead to
crashes, even if you don't recurse particularly deeply.
We work around the kernel bug by touching a page deep within the stack (after
ensuring that we know how big it is), to preallocate virtual address space for
the stack so that the kernel doesn't allow the brk() area to wander into it,
when building clang as a Linux PIE binary.
Chris Bieneman [Tue, 16 Aug 2016 22:16:29 +0000 (22:16 +0000)]
[CMake] Workflow improvements to PGO generation
This patch adds a few new convenience options used by the PGO CMake cache to setup options on bootstrap stages. The new options are:
PGO_INSTRUMENT_LTO - Builds the instrumented and final builds with LTO
PGO_BUILD_CONFIGURATION - Accepts a CMake cache script that can be used for complex configuration of the stage2-instrumented and stage2 builds.
The patch also includes a fix for bootstrap dependencies so that the instrumented LTO tools don't get used when building the final stage, and it adds distribution targets to the passthrough.
Adrian McCarthy [Tue, 16 Aug 2016 22:11:18 +0000 (22:11 +0000)]
Emit debug info for dynamic classes if they are imported from a DLL.
With -debug-info-kind=limited, we omit debug info for dynamic classes that live in other TUs. This reduces duplicate type information. When statically linked, the type information comes together. But if your binary has a class derived from a base in a DLL, the base class info is not available to the debugger.
The decision is made in shouldOmitDefinition (CGDebugInfo.cpp). Per a suggestion from rnk, I've tweaked the decision so that we do include definitions for classes marked as DLL imports. This should be a relatively small number of classes, so we don't pay a large price for duplication of the type info, yet it should cover most cases on Windows.
Essentially this makes debug info for DLLs independent, but we still assume that all TUs within the same DLL will be consistently built with (or without) debug info and the debugger will be able to search across the debug info within that scope to resolve any declarations into definitions, etc.
Erik Pilkington [Tue, 16 Aug 2016 17:44:11 +0000 (17:44 +0000)]
[ObjC] Warn on unguarded use of partial declaration
This commit adds a traversal of the AST after Sema of a function that diagnoses
unguarded references to declarations that are partially available (based on
availability attributes). This traversal is only done when we would otherwise
emit -Wpartial-availability.
This commit is part of a feature I proposed here:
http://lists.llvm.org/pipermail/cfe-dev/2016-July/049851.html
Aaron Ballman [Tue, 16 Aug 2016 14:48:39 +0000 (14:48 +0000)]
Reduce the number of allocations required for AST attributes. In test cases, the max resident memory changed from 65760k to 64476k which is 1.9% improvement. Allocations in grow_pod changed from 8847 to 4872 according to tcmalloc heap profiler. Overall running time remained the same.
Samuel Antao [Tue, 16 Aug 2016 14:31:39 +0000 (14:31 +0000)]
Add empty --gcc-toolchain empty to cuda-detect test.
Unless we overload the default gcc toolchain with an empty string
the system root used in the tests will be ignored if the user builds
clang with a custom gcc toolchain.
Richard Smith [Tue, 16 Aug 2016 00:13:47 +0000 (00:13 +0000)]
PR28978: If we need overload resolution for the move constructor of an
anonymous union member of a class, we need overload resolution for the move
constructor of the class itself too; we can't rely on Sema to do the right
thing for us for anonymous union types.
Justin Lebar [Mon, 15 Aug 2016 23:00:49 +0000 (23:00 +0000)]
[CUDA] Raise an error if a wrong-side call is codegen'ed.
Summary:
Some function calls in CUDA are allowed to appear in
semantically-correct programs but are an error if they're ever
codegen'ed. Specifically, a host+device function may call a host
function, but it's an error if such a function is ever codegen'ed in
device mode (and vice versa).
Previously, clang made no attempt to catch these errors. For the most
part, they would be caught by ptxas, and reported as "call to unknown
function 'foo'".
Now we catch these errors and report them the same as we report other
illegal calls (e.g. a call from a host function to a device function).
This has a small change in error-message behavior for calls that were
previously disallowed (e.g. calls from a host to a device function).
Previously, we'd catch disallowed calls fairly early, before doing
additional semantic checking e.g. of the call's arguments. Now we catch
these illegal calls at the very end of our semantic checks, so we'll
only emit a "illegal CUDA call" error if the call is otherwise
well-formed.
Manman Ren [Mon, 15 Aug 2016 21:05:00 +0000 (21:05 +0000)]
Objective-C diagnostics: isObjCNSObjectType should check through AttributedType.
For the following example:
typedef __attribute__((NSObject)) CGColorRef ColorAttrRef;
@property (strong, nullable) ColorAttrRef color;
The property type should be ObjC NSObject type and the compiler should not emit
error: property with 'retain (or strong)' attribute must be of object type
Justin Lebar [Mon, 15 Aug 2016 20:38:56 +0000 (20:38 +0000)]
Add the notion of deferred diagnostics.
Summary:
This patch lets you create diagnostics that are emitted if and only if a
particular FunctionDecl is codegen'ed.
This is necessary for CUDA, where some constructs -- e.g. calls from
host+device functions to host functions when compiling for device -- are
allowed to appear in semantically-correct programs, but only if they're
never codegen'ed.
Justin Lebar [Mon, 15 Aug 2016 20:38:52 +0000 (20:38 +0000)]
[CUDA] Include CUDA headers before anything else.
Summary:
There's no point to --cuda-path if we then go and include /usr/include
first. And if you install the right packages, Ubuntu will install (very
old) CUDA headers there.
projects / clang / log