George Karpenkov [Fri, 10 Aug 2018 22:27:04 +0000 (22:27 +0000)]
[analyzer] Record nullability implications on getting items from NSDictionary
If we get an item from a dictionary, we know that the item is non-null
if and only if the key is non-null.
This patch is a rather hacky way to record this implication, because
some logic needs to be duplicated from the solver.
And yet, it's pretty simple, performant, and works.
Other possible approaches:
- Record the implication, in future rely on Z3 to pick it up.
- Generalize the current code and move it to the constraint manager.
George Karpenkov [Fri, 10 Aug 2018 21:36:45 +0000 (21:36 +0000)]
[analyzer] [NFC] [tests] Move plist-based diagnostics tests to separate files, use diff instead of a FileCheck
Some of the analyzer tests check the exact plist output, in order to
verify that the diagnostics produced is correct.
Current testing setup has many issues:
plist output clobbers tests, making them harder to read
it is impossible to debug test failures given error messages from FileCheck.
The only recourse is manually creating the files and using the diff
again, it is impossible to update the tests given the error message:
the only process is a tedious manual one,
going from a separate plist file to CHECK directives.
This patch offers a much better approach of using "diff" directly in place of FileCheck,
and moving tests to separate files.
Generated using the following script:
```
import os
import glob
import re
import subprocess
subprocess.check_call(["mkdir", "-p", plist_out_folder])
with open(plist_out_f, 'w') as out_f:
out_f.write(plist_data)
def main():
files = glob.glob("**/*.*")
for f in files:
with open(f) as f_handler:
data = f_handler.read()
if diagnostics_key in data:
print "Converting %s" %f
process_file(f, data)
Clang generates copy and dispose helper functions for each block literal
on the stack. Often these functions are equivalent for different blocks.
This commit makes changes to merge equivalent copy and dispose helper
functions and reduce code size.
To enable merging equivalent copy/dispose functions, the captured object
infomation is encoded into the helper function name. This allows IRGen
to check whether an equivalent helper function has already been emitted
and reuse the function instead of generating a new helper function
whenever a block is defined. In addition, the helper functions are
marked as linkonce_odr to enable merging helper functions that have the
same name across translation units and marked as unnamed_addr to enable
the linker's deduplication pass to merge functions that have different
names but the same content.
David Chisnall [Fri, 10 Aug 2018 12:53:13 +0000 (12:53 +0000)]
Add Windows support for the GNUstep Objective-C ABI V2.
Summary:
Introduces funclet-based unwinding for Objective-C and fixes an issue
where global blocks can't have their isa pointers initialised on
Windows.
After discussion with Dustin, this changes the name mangling of
Objective-C types to prevent a C++ catch statement of type struct X*
from catching an Objective-C object of type X*.
Bruno Ricci [Fri, 10 Aug 2018 11:20:20 +0000 (11:20 +0000)]
Fix a wrong type bug in ParsedAttr::TypeTagForDatatypeData
This patch fixes a wrong type bug inside ParsedAttr::TypeTagForDatatypeData.
The details to the best of my knowledge are as follow. The incredible thing
is that everything works out just fine by chance due to a sequence of lucky
coincidences in the layout of various types.
The struct ParsedAttr::TypeTagForDatatypeData contains among other things
a ParsedType *MatchingCType, where ParsedType is just OpaquePtr<QualType>.
However the member MatchingCType is initialized in the constructor for
type_tag_for_datatype attribute as follows:
new (&ExtraData.MatchingCType) ParsedType(matchingCType);
This results in the ParsedType being constructed in the location of the
ParsedType * Later ParsedAttr::getMatchingCType do return
*getTypeTagForDatatypeDataSlot().MatchingCType; which instead of
dereferencing the ParsedType * will dereference the QualType inside
the ParsedType. Now this QualType in this case contains no qualifiers
and therefore is a valid Type *. Therefore getMatchingCType returns a
Type or at least the stuff that is in the first sizeof(void*) bytes of it,
But it turns out that Type inherits from ExtQualsCommonBase and that the
first member of ExtQualsCommonBase is a const Type *const BaseType. This
Type * in this case points to the original Type pointed to by the
QualType and so everything works fine even though all the types were wrong.
This bug was only found because I changed the layout of Type,
which obviously broke all of this long chain of improbable events.
Hans Wennborg [Fri, 10 Aug 2018 09:49:21 +0000 (09:49 +0000)]
clang-cl: Support /guard:cf,nochecks
This extension emits the guard cf table without inserting the
instrumentation. Currently that's what clang-cl does with /guard:cf
anyway, but this allows a user to request that explicitly.
Richard Trieu [Fri, 10 Aug 2018 01:30:10 +0000 (01:30 +0000)]
Fix size calculation from r339380
r339380 changed the trailing types of ParsedAttr to use llvm::TrailingObjects.
However, it did not copy over one of the size attributes, causing a too
small allocation for this object. The error was detected with
AddressSanitizer use-after-poison
Emmett Neyman [Thu, 9 Aug 2018 21:59:01 +0000 (21:59 +0000)]
Added LLVM metadata to generated IR to increase vectorization width
Summary: Edited `loop_proto_to_llvm` to emit metadata at the end of the generated IR. This metadata will increase the vector width when the IR is optimized.
Erich Keane [Thu, 9 Aug 2018 21:18:13 +0000 (21:18 +0000)]
Fix and recommit r339382 based on solution from RSmith:
As sent on cfe-commits:
"You need to use "friend TrailingObjects;" here, not
"friend class TrailingObjects;", to avoid breaking MSVC
(which doesn't implement injected-class-names quite according to spec)."
Erich Keane [Thu, 9 Aug 2018 20:25:12 +0000 (20:25 +0000)]
[NFC] Convert ParsedAttr to use llvm::TrailingObjects
ParsedAttr is using a hand-rolled trailing-objects
implementation that gets cleaned up quite a bit by
just using llvm::TrailingObjects. This is a large
TrailingObjects list, but most things are length '0'.
Erich Keane [Thu, 9 Aug 2018 13:21:32 +0000 (13:21 +0000)]
Implement diagnostic stream operator for ParsedAttr.
As a part of attempting to clean up the way attributes are
printed, this patch adds an operator << to the diagnostics/
partialdiagnostics so that ParsedAttr can be sent directly.
This patch also rewrites a large amount* of the times when
ParsedAttr was printed using its IdentifierInfo object instead
of being printed itself.
*"a large amount" == "All I could find".
David Chisnall [Thu, 9 Aug 2018 08:02:42 +0000 (08:02 +0000)]
Correctly initialise global blocks on Windows.
Summary:
Windows does not allow globals to be initialised to point to globals in
another DLL. Exported globals may be referenced only from code. Work
around this by creating an initialiser that runs in early library
initialisation and sets the isa pointer.
Petr Hosek [Thu, 9 Aug 2018 02:16:18 +0000 (02:16 +0000)]
[CMake] Use normalized Windows target triples
Changes the default Windows target triple returned by
GetHostTriple.cmake from the old environment names (which we wanted to
move away from) to newer, normalized ones. This also requires updating
all tests to use the new systems names in constraints.
Petr Hosek [Wed, 8 Aug 2018 22:23:57 +0000 (22:23 +0000)]
[ADT] Normalize empty triple components
LLVM triple normalization is handling "unknown" and empty components
differently; for example given "x86_64-unknown-linux-gnu" and
"x86_64-linux-gnu" which should be equivalent, triple normalization
returns "x86_64-unknown-linux-gnu" and "x86_64--linux-gnu". autoconf's
config.sub returns "x86_64-unknown-linux-gnu" for both
"x86_64-linux-gnu" and "x86_64-unknown-linux-gnu". This changes the
triple normalization to behave the same way, replacing empty triple
components with "unknown".
Craig Topper [Wed, 8 Aug 2018 19:55:52 +0000 (19:55 +0000)]
[Builtins] Implement __builtin_clrsb to be compatible with gcc
gcc defines an intrinsic called __builtin_clrsb which counts the number of extra sign bits on a number. This is equivalent to counting the number of leading zeros on a positive number or the number of leading ones on a negative number and subtracting one from the result. Since we can't count leading ones we need to invert negative numbers to count zeros.
This patch will cause the builtin to be expanded inline while gcc uses a call to a function like clrsbdi2 that is implemented in libgcc. But this is similar to what we already do for popcnt. And I don't think compiler-rt supports clrsbdi2.
Craig Topper [Wed, 8 Aug 2018 19:14:23 +0000 (19:14 +0000)]
[CodeGen][Timers] Enable llvm::TimePassesIsEnabled when -ftime-report is specified
r330571 added a new FrontendTimesIsEnabled variable and replaced many usages of llvm::TimePassesIsEnabled. Including the place that set llvm::TimePassesIsEnabled for -ftime-report. The effect of this is that -ftime-report now only contains the timers specifically referenced in CodeGenAction.cpp and none of the timers in the backend.
This commit adds back the assignment, but otherwise leaves everything else unchanged.
Simon Pilgrim [Wed, 8 Aug 2018 15:53:14 +0000 (15:53 +0000)]
[CGObjCGNU] Rename GetSelector helper method to fix -Woverloaded-virtual warning (PR38210)
As suggested by @theraven on PR38210, this patch fixes the gcc -Woverloaded-virtual warnings by renaming the extra CGObjCGNU::GetSelector method to CGObjCGNU::GetTypedSelector
Balazs Keri [Wed, 8 Aug 2018 15:04:27 +0000 (15:04 +0000)]
[AST] Check described template at structural equivalence check.
Summary:
When checking a class or function the described class or function template
is checked too.
Split StructuralEquivalenceContext::Finish into multiple functions.
Improved test with symmetric check, added new tests.
Kristof Umann [Wed, 8 Aug 2018 13:18:53 +0000 (13:18 +0000)]
[analyzer][UninitializedObjectChecker] Pointer/reference objects are dereferenced according to dynamic type
This patch fixed an issue where the dynamic type of pointer/reference
object was known by the analyzer, but wasn't obtained in the checker,
which resulted in false negatives. This should also increase reliability
of the checker, as derefencing is always done now according to the
dynamic type (even if that happens to be the same as the static type).
Special thanks to Artem Degrachev for setting me on the right track.
Kristof Umann [Wed, 8 Aug 2018 12:23:02 +0000 (12:23 +0000)]
[analyzer][UninitializedObjectChecker] Fixed a false negative by no longer filtering out certain constructor calls
As of now, all constructor calls are ignored that are being called
by a constructor. The point of this was not to analyze the fields
of an object, so an uninitialized field wouldn't be reported
multiple times.
This however introduced false negatives when the two constructors
were in no relation to one another -- see the test file for a neat
example for this with singletons. This patch aims so fix this issue.
Balazs Keri [Wed, 8 Aug 2018 09:40:57 +0000 (09:40 +0000)]
[ASTImporter] Load external Decls when getting field index.
Summary:
At equality check of fields without name the index of fields is compared.
At determining the index of a field all fields of the parent context
should be loaded from external source to find the field at all.
Matt Davis [Tue, 7 Aug 2018 23:13:28 +0000 (23:13 +0000)]
[analyzer] Avoid querying this-pointers for static-methods.
Summary:
The loop-widening code processes c++ methods looking for `this` pointers. In
the case of static methods (which do not have `this` pointers), an assertion
was triggering. This patch avoids trying to process `this` pointers for
static methods, and thus avoids triggering the assertion .
JF Bastien [Tue, 7 Aug 2018 22:43:44 +0000 (22:43 +0000)]
[NFC] Improve auto-var-init alignment check
We're not actually testing for alignment, we just want to know that whatever incoming alignment got propagated. Do that by capturing the alignment and checking that it's actually what's passed later, instead of hard-coding an alignment value.
David Greene [Tue, 7 Aug 2018 17:44:43 +0000 (17:44 +0000)]
[WebAssembly] Force use of lld for test/Driver/wasm-toolchain.c(pp)
lld is the only supported linker that works for WebAssembly, so ensure
clang is using it for this test. This gets the tests passing when
configuring clang to use a different linker by default.
Kristof Umann [Tue, 7 Aug 2018 12:55:26 +0000 (12:55 +0000)]
[analyzer][UninitializedObjectChecker] New flag to turn off dereferencing
Even for a checker being in alpha, some reports about pointees held so little
value to the user that it's safer to disable pointer/reference chasing for now.
It can be enabled with a new flag, in which case checker should function as it
has always been. This can be set with `CheckPointeeInitialization`.
David Chisnall [Tue, 7 Aug 2018 12:02:46 +0000 (12:02 +0000)]
[objc-gnustep] Don't emit .guess ivar offset vars.
These were intended to allow non-fragile and fragile ABI code to be
mixed, as long as the fragile classes were higher up the hierarchy than
the non-fragile ones. Unfortunately:
- No one actually wants to do this.
- Recent versions of Linux's run-time linker break it.
[clang-format] comment reflow: add last line's penalty when ending broken
Summary:
This fixes a bug in clang-format where the last line's penalty is not
taken into account when its ending is broken. Usually the last line's penalty
is handled by addNextStateToQueue, but in cases where the trailing `*/` is put
on a newline, the contents of the last line have to be considered for penalizing.
Fix clash of gcc toolchains in driver regression tests
For some regression tests the path to the right toolchain is specified
using the -sysroot switch. However, if clang was configured with a
custom gcc toolchain (either by using GCC_INSTALL_PREFIX in cmake or the
equivalent configure command), the path to the custom gcc toolchain path
takes precedence to the one specified by sysroot. This causes several
regression tests to fail as they will be using an unexpected path. This
patch fixes this issue by adding --gcc-toolchain='' to all tests that
rely on that. The empty string causes the driver to pick the path from
sysroot instead.
This patch contain the same kind of fixes as done in rC225182
JF Bastien [Tue, 7 Aug 2018 04:44:13 +0000 (04:44 +0000)]
Auto var init test fix #2
It turns out that the AVX bots have different alignment for their vectors, and my test mistakenly assumed a particular vector alignent on the stack. Instead, capture the alignment and test for it in subsequent operations.
JF Bastien [Tue, 7 Aug 2018 03:12:52 +0000 (03:12 +0000)]
[NFC] Test automatic variable initialization
Summary:
r337887 started using memset for automatic variable initialization where sensible. A follow-up discussion leads me to believe that we should better test automatic variable initialization, and that there are probably follow-up patches in clang and LLVM to improve codegen. It’ll be important to measure -O0 compile time, and figure out which transforms should be in the frontend versus the backend.
This patch is just a test of the current behavior, no questions asked. Follow-up patches will tune the code generation.
Artem Dergachev [Tue, 7 Aug 2018 02:22:59 +0000 (02:22 +0000)]
[analyzer] NFC: Document that we support implicit argument constructors.
The change in the AST in r338135 caused us to accidentally support
inlining constructors of operator implicit arguments. Previously they were
hard to support because they were treated as arguments in expressions
but not in declarations, but now they can be transparently treated as
simple temporaries.
Add tests and comments to explain how it now works.
[lit, python] Always add quotes around the python path in lit
Summary:
The issue with the python path is that the path to python on Windows can contain spaces. To make the tests always work, the path to python needs to be surrounded by quotes.
This is a companion change to: https://reviews.llvm.org/D50206
For InnerPointerChecker to function properly, both the checker itself
and parts of MallocChecker that handle relevant use-after-free problems
need to be turned on. So far, the latter part has been developed within
MallocChecker's NewDelete sub-checker, often causing warnings to appear
under that name. This patch defines a new CheckKind within MallocChecker
for the inner pointer checking functionality, so that the correct name
is displayed in warnings and in the ExplodedGraph.
Simon Marchi [Mon, 6 Aug 2018 21:48:20 +0000 (21:48 +0000)]
[VirtualFileSystem] InMemoryFileSystem::status: Return a Status with the requested name
Summary:
InMemoryFileSystem::status behaves differently than
RealFileSystem::status. The Name contained in the Status returned by
RealFileSystem::status will be the path as requested by the caller,
whereas InMemoryFileSystem::status returns the normalized path.
For example, when requested the status for "../src/first.h",
RealFileSystem returns a Status with "../src/first.h" as the Name.
InMemoryFileSystem returns "/absolute/path/to/src/first.h".
The reason for this change is that I want to make a unit test in the
clangd testsuite (where we use an InMemoryFileSystem) to reproduce a
bug I get with the clangd program (where a RealFileSystem is used).
This difference in behavior "hides" the bug in the unit test version.
An indirect impact of this change is that a -Wnonportable-include-path
warning is now emitted in test PCH/case-insensitive-include.c. This is
because the real path of the included file (with the wrong case) was not
available previously, whereas it is now.
projects / clang / log