Nico Weber [Thu, 31 Aug 2017 06:17:08 +0000 (06:17 +0000)]
Suppress -Wdelete-non-virtual-dtor warnings about classes defined in system headers.
r312167 made it so that we emit Wdelete-non-virtual-dtor from delete statements
that are in system headers (e.g. std::unique_ptr). That works great on Linux
and macOS, but on Windows there are non-final classes that are defined in
system headers that have virtual methods but non-virtual destructors and yet
get deleted through a base class pointer (e.g. ATL::CAccessToken::CRevert). So
paddle back a bit and don't emit the warning if it's about a class defined in a
system header.
Nico Weber [Wed, 30 Aug 2017 20:25:22 +0000 (20:25 +0000)]
Let -Wdelete-non-virtual-dtor fire in system headers too.
Makes the warning useful again in a std::unique_ptr world, PR28460.
Also make the warning not fire in unevaluated contexts, since system libraries
(e.g. libc++) do do that. This would've been a good change before we started
emitting this warning in system headers too, but "normal" code seems to be less
template-heavy, so we didn't notice until now.
Alex Lorenz [Wed, 30 Aug 2017 15:28:01 +0000 (15:28 +0000)]
Recommit r312127: [refactor] AST selection tree should contain syntactic
form of PseudoObjectExpr
The new commit adjusts unittest test code compilation options so that the
Objective-C code in the unittest can be parsed on non-macOS platforms.
Original message:
The AST selection finder now constructs a selection tree that contains only the
syntactic form of PseudoObjectExpr. This form of selection tree is more
meaningful when doing downstream analysis as we're interested in the syntactic
features of the AST and the correct lexical parent relation.
Alex Lorenz [Wed, 30 Aug 2017 15:00:27 +0000 (15:00 +0000)]
[refactor] AST selection tree should contain syntactic form
of PseudoObjectExpr
The AST selection finder now constructs a selection tree that contains only the
syntactic form of PseudoObjectExpr. This form of selection tree is more
meaningful when doing downstream analysis as we're interested in the syntactic
features of the AST and the correct lexical parent relation.
Summary:
This is an implementation for [bug 17362](https://bugs.llvm.org/attachment.cgi?bugid=17362) which adds support for indenting preprocessor statements inside if/ifdef/endif. This takes previous work from fmauch (https://github.com/fmauch/clang/tree/preprocessor_indent) and makes it into a full feature.
The context of this patch is that I'm a VMware intern, and I implemented this because VMware needs the feature. As such, some decisions were made based on what VMware wants, and I would appreciate suggestions on expanding this if necessary to use-cases other people may want.
This adds a new enum config option, `IndentPPDirectives`. Values are:
* `PPDIS_None` (in config: `None`):
```
#if FOO
#if BAR
#include <foo>
#endif
#endif
```
* `PPDIS_AfterHash` (in config: `AfterHash`):
```
#if FOO
# if BAR
# include <foo>
# endif
#endif
```
This is meant to work whether spaces or tabs are used for indentation. Preprocessor indentation is independent of indentation for non-preprocessor lines.
Preprocessor indentation also attempts to ignore include guards with the checks:
1. Include guards cover the entire file
2. Include guards don't have `#else`
3. Include guards begin with
```
#ifndef <var>
#define <var>
```
This patch allows `UnwrappedLineParser::PPBranchLevel` to be decremented to -1 (the initial value is -1) so the variable can be used for indent tracking.
Defects:
* This patch does not handle the case where there's code between the `#ifndef` and `#define` but all other conditions hold. This is because when the #define line is parsed, `UnwrappedLineParser::Lines` doesn't hold the previous code line yet, so we can't detect it. This is out of the scope of this patch.
* This patch does not handle cases where legitimate lines may be outside an include guard. Examples are `#pragma once` and `#pragma GCC diagnostic`, or anything else that does not change the meaning of the file if it's included multiple times.
* This does not detect when there is a single non-preprocessor line in front of an include-guard-like structure where other conditions hold because `ScopedLineState` hides the line.
* Preprocessor indentation throws off `TokenAnnotator::setCommentLineLevels` so the indentation of comments immediately before indented preprocessor lines is toggled on each run. Fixing this issue appears to be a major change and too much complexity for this patch.
Boris Kolpackov [Wed, 30 Aug 2017 08:45:59 +0000 (08:45 +0000)]
[modules] Add ability to specify module name to module file mapping
Extend the -fmodule-file option to support the [<name>=]<file> value format.
If the name is omitted, then the old semantics is preserved (the module file
is loaded whether needed or not). If the name is specified, then the mapping
is treated as just another prebuilt module search mechanism, similar to
-fprebuilt-module-path, and the module file is only loaded if actually used
(e.g., via import). With one exception: this mapping also overrides module
file references embedded in other modules (which can be useful if module files
are moved/renamed as often happens during remote compilation).
This override semantics requires some extra work: we now store the module name
in addition to the file name in the serialized AST representation.
Extract the argument forwarding for OpenCL arguments. Make this more
data driven as we are just repeating the argument name and spelling.
This costs a slight bit more memory due to the string duplication, but
makes it easier to follow. It should be possible to forward the
internal string representation from the TableGen data to avoid this.
But, this makes the code simpler to follow for now.
Richard Smith [Tue, 29 Aug 2017 22:14:43 +0000 (22:14 +0000)]
PR10147: When substituting a template template argument, substitute in the most
recent (non-friend) declaration to pick up the right set of default template
arguments.
Evgeniy Stepanov [Tue, 29 Aug 2017 20:03:51 +0000 (20:03 +0000)]
Minimal runtime for UBSan.
Summary:
An implementation of ubsan runtime library suitable for use in production.
Minimal attack surface.
* No stack traces.
* Definitely no C++ demangling.
* No UBSAN_OPTIONS=log_file=/path (very suid-unfriendly). And no UBSAN_OPTIONS in general.
* as simple as possible
Minimal CPU and RAM overhead.
* Source locations unnecessary in the presence of (split) debug info.
* Values and types (as in A+B overflows T) can be reconstructed from register/stack dumps, once you know what type of error you are looking at.
* above two items save 3% binary size.
When UBSan is used with -ftrap-function=abort, sometimes it is hard to reason about failures. This library replaces abort with a slightly more informative message without much extra overhead. Since ubsan interface in not stable, this code must reside in compiler-rt.
Reid Kleckner [Tue, 29 Aug 2017 17:40:04 +0000 (17:40 +0000)]
[ms] Fix vbtable index for covariant overrides of vbase methods
Overriding a method from a virtual base with a covariant return type
consumes a slot from the vftable in the virtual base. This can make it
impossible to implement certain diamond inheritance hierarchies, but we
have to follow along for compatibility in the simple cases.
This patch only affects our vtable dumper and member pointer function
mangling, since all other callers of getMethodVFTableLocation seem to
recompute VBTableIndex instead of using the one in the method location.
Boris Kolpackov [Tue, 29 Aug 2017 15:30:18 +0000 (15:30 +0000)]
[modules-ts] Omit submodule semantics for TS modules
If a TS module name has more than one component (e.g., foo.bar) then we
erroneously activated the submodule semantics when encountering a module
declaration in the module implementation unit (e.g., 'module foo.bar;').
Summary:
This patch detects the leading '<' in likely xml files and stops formatting in
that case. A recent use of a Qt xml file with a .ts extension triggered this:
http://doc.qt.io/qt-4.8/linguist-ts-file-format.html
Clang format does not allow the flag **BraceWrapping.AfterEnum** control the case when our **enum** is preceded by **typedef** keyword (what is common in C language).
**Patch description:**
Added case to the **"AfterEnum"** flag when our enum does not start a line - is preceded by **typedef** keyword.
Raphael Isemann [Tue, 29 Aug 2017 09:27:41 +0000 (09:27 +0000)]
[modules] Add test for using declaration in classes.
Summary:
This adds a test that checks if the using declaration in classes still works as intended with modules.
The motivation for this is that we tried to add a shortcut to `removeDecl` that would skip the removal of declarations from the lookup table if they are hidden. This optimization passed the clang test suite but actually broke the using declaration in combination with -fmodules-local-submodule-visibility. In this mode we hide all decls from other modules such as by chance the parent method, in which case don't remove the parent method from the lookup table and get ambiguous lookup errors. After this patch we now correctly see if this behavior is broken by a patch like this in the test suite.
Summary:
Previously, clang-format would try to wrap template string substitutions
by indenting relative to the openening `${`. This helped with
indenting structured strings, such as strings containing HTML, as the
substitutions would be aligned according to the structure of the string.
However it turns out that the overwhelming majority of template string +
substitution usages are for substitutions into non-structured strings,
e.g. URLs or just plain messages. For these situations, clang-format
would often produce very ugly indents, in particular for strings
containing no line breaks:
return `<a href='http://google3/${file}?l=${row}'>${file}</a>(${
row
},${
col
}): `;
This change makes clang-format indent template string substitutions as
if they were string concatenation operations. It wraps +4 on overlong
lines and keeps all operands on the same line:
While this breaks some lexical continuity between the `${` and `row}`
here, the overall effects are still a huge improvement, and users can
still manually break the string using `+` if desired.
Serge Pavlov [Tue, 29 Aug 2017 05:22:26 +0000 (05:22 +0000)]
Use class to pass information about executable name
Information about clang executable name components, such as target and
driver mode, was passes in std::pair. With this change it is passed in
a special structure. It improves readability and makes access to this
information more convenient.
Yuka Takahashi [Tue, 29 Aug 2017 02:01:56 +0000 (02:01 +0000)]
[Bash-autocompletion] Add support for -std=
Summary:
Add support for autocompleting values of -std= by including
LangStandards.def. This patch relies on D36782, and is using two-stage
code generation.
Richard Smith [Tue, 29 Aug 2017 01:52:13 +0000 (01:52 +0000)]
Improve constant expression evaluation of arrays of unknown bound.
The standard is not clear on how these are supposed to be handled, so we
conservatively treat as non-constant any cases whose value is unknown or whose
evaluation might result in undefined behavior.
Replace "long" with __UINTPTR_TYPE__
to make the test added in rL311935 Windows-friendly.
Caught by the buildbot llvm-clang-x86_64-expensive-checks-win.
This diff fixes modeling of arithmetic
expressions where pointers are treated as integers
(i.e. via C-style / reinterpret casts).
For now we return UnknownVal unless the operation is a comparison.
Michal Gorny [Mon, 28 Aug 2017 20:29:52 +0000 (20:29 +0000)]
Reland r311836 - [Driver] Use arch type to find compiler-rt libraries (on Linux)
Use llvm::Triple::getArchTypeName() when looking for compiler-rt
libraries, rather than the exact arch string from the triple. This is
more correct as it matches the values used when building compiler-rt
(builtin-config-ix.cmake) which are the subset of the values allowed
in triples.
For example, this fixes an issue when the compiler set for
i686-pc-linux-gnu triple would not find an i386 compiler-rt library,
while this is the exact arch that is detected by compiler-rt. The same
applies to any other i?86 variant allowed by LLVM.
This also makes the special case for MSVC unnecessary, since now i386
will be used reliably for all 32-bit x86 variants.
Erich Keane [Mon, 28 Aug 2017 18:53:17 +0000 (18:53 +0000)]
Change Diagnostic Category size error from runtime to compiletime
Diagnostic Categories are fairly annoying, and are only enforced
by a runtime-debug-only assert. This puts in a touch more work
to get this all done at compile-time with static asserts
Hans Wennborg [Mon, 28 Aug 2017 17:53:00 +0000 (17:53 +0000)]
Revert r311857 "Emit static constexpr member as available_externally definition"
It caused PR759744.
> Emit static constexpr member as available_externally definition
>
> By exposing the constant initializer, the optimizer can fold many
> of these constructs.
>
> Differential Revision: https://reviews.llvm.org/D34992
Alex Lorenz [Mon, 28 Aug 2017 11:12:05 +0000 (11:12 +0000)]
[refactor] initial support for refactoring action rules
This patch implements the initial support for refactoring action rules. The
first rule that's supported is a "source change" rule that returns a set of
atomic changes. This patch is based on the ideas presented in my RFC:
Peter Szecsi [Mon, 28 Aug 2017 10:50:28 +0000 (10:50 +0000)]
[StaticAnalyzer] LoopUnrolling: Keep track the maximum number of steps for each loop
This way the unrolling can be restricted for loops which will take at most a
given number of steps. It is defined as 128 in this patch and it seems to have
a good number for that purpose.
Peter Szecsi [Mon, 28 Aug 2017 10:34:50 +0000 (10:34 +0000)]
[StaticAnalyzer] LoopUnrolling: Excluding loops which splits the state
Added check if the execution of the last step of the given unrolled loop has
generated more branches. If yes, than treat it as a normal (non-unrolled) loop
in the remaining part of the analysis.
Peter Szecsi [Mon, 28 Aug 2017 10:21:24 +0000 (10:21 +0000)]
[StaticAnalyzer] LoopUnrolling fixes
1. The LoopUnrolling feature needs the LoopExit included in the CFG so added this
dependency via the config options
2. The LoopExit element can be encountered even if we haven't encountered the
block of the corresponding LoopStmt. So the asserts were not right.
3. If we are caching out the Node then we get a nullptr from generateNode which
case was not handled.
Faisal Vali [Sun, 27 Aug 2017 16:49:47 +0000 (16:49 +0000)]
Don't see through 'using member-declarations' when determining the relation of any potential implicit object expression to the parent class of the member function containing the function call.
Prior to this patch clang would not error here:
template <class T> struct B;
template <class T> struct A {
void foo();
void foo2();
void test1() {
B<T>::foo(); // OK, foo is declared in A<int> - matches type of 'this'.
B<T>::foo2(); // This should be an error!
// foo2 is found in B<int>, 'base unrelated' to 'this'.
}
};
template <class T> struct B : A<T> {
using A<T>::foo2;
};
Vassil Vassilev [Sun, 27 Aug 2017 10:58:03 +0000 (10:58 +0000)]
D34444: Teach codegen to work in incremental processing mode.
When isIncrementalProcessingEnabled is on we might want to produce multiple
llvm::Modules. This patch allows the clients to start a new llvm::Module,
allowing CodeGen to continue working after a HandleEndOfTranslationUnit call.
This should give the necessary facilities to write a unittest for D34059.
As discussed in the review this is meant to give us a way to proceed forward
in our efforts to upstream our interpreter-related patches. The design of this
will likely change soon.
struct B {
void f() {
A::bar(3); // selects (double) ??!!
A::g((int*)0); // Instead of no object argument, states conversion error?!!
}
};
The fix is as follows: When we detect that what appears to be an implicit member function call (A::bar) is actually a call to a member of a class (A) unrelated to the type (B) that contains the member function (B::f) from which the call is being made, don't treat it (A::bar) as an Implicit Member Call Expression.
P.S. I wonder if there is an existing bug report related to this? (Surprisingly, a cursory search did not find one).
Michal Gorny [Sat, 26 Aug 2017 21:35:11 +0000 (21:35 +0000)]
[Driver] Use arch type to find compiler-rt libraries (on Linux)
Use llvm::Triple::getArchTypeName() when looking for compiler-rt
libraries, rather than the exact arch string from the triple. This is
more correct as it matches the values used when building compiler-rt
(builtin-config-ix.cmake) which are the subset of the values allowed
in triples.
For example, this fixes an issue when the compiler set for
i686-pc-linux-gnu triple would not find an i386 compiler-rt library,
while this is the exact arch that is detected by compiler-rt. The same
applies to any other i?86 variant allowed by LLVM.
This also makes the special case for MSVC unnecessary, since now i386
will be used reliably for all 32-bit x86 variants.
Richard Smith [Sat, 26 Aug 2017 01:04:35 +0000 (01:04 +0000)]
Add flag to request Clang is ABI-compatible with older versions of itself
This patch adds a flag -fclang-abi-compat that can be used to request that
Clang attempts to be ABI-compatible with some older version of itself.
This is provided on a best-effort basis; right now, this can be used to undo
the ABI change in r310401, reverting Clang to its prior C++ ABI for pass/return
by value of class types affected by that change, and to undo the ABI change in
r262688, reverting Clang to using integer registers rather than SSE registers
for passing <1 x long long> vectors. The intent is that we will maintain this
backwards compatibility path as we make ABI-breaking fixes in future.
The reversion to the old behavior for r310401 is also applied to the PS4 target
since that change is not part of its platform ABI (which is essentially to do
whatever Clang 3.2 did).
Daniel Jasper [Fri, 25 Aug 2017 19:14:53 +0000 (19:14 +0000)]
[Format] Invert nestingAndIndentLevel pair in WhitespaceManager used for
alignments
Indent should be compared before nesting level to determine if a token
is on the same scope as the one we align with. Because it was inverted,
clang-format sometimes tried to align tokens with tokens from outer
scopes, causing the assert(Shift >= 0) to fire.
This fixes bug #33507. Patch by Beren Minor, thank you!
Vedant Kumar [Fri, 25 Aug 2017 18:07:03 +0000 (18:07 +0000)]
[Frontend] Fix printing policy for AST context loaded from file
In ASTUnit::LoadFromASTFile, the context object is set up using
default-constructed LangOptions (which only later get populated). As the
language options are used in the constructor of PrintingPolicy, this
needs to be updated explicitly after the language options are available.
Alex Lorenz [Fri, 25 Aug 2017 16:12:17 +0000 (16:12 +0000)]
[ObjC] Add a -Wobjc-messaging-id warning
-Wobjc-messaging-id is a new, non-default warning that warns about
message sends to unqualified id in Objective-C. This warning is useful
for projects that would like to avoid any potential future compiler
errors/warnings, as the system frameworks might add a method with the same
selector which could make the message send to id ambiguous.
Alex Lorenz [Fri, 25 Aug 2017 10:07:00 +0000 (10:07 +0000)]
[IRGen] Evaluate constant static variables referenced through member
expressions
C++ allows us to reference static variables through member expressions. Prior to
this commit, non-integer static variables that were referenced using a member
expression were always emitted using lvalue loads. The old behaviour introduced
an inconsistency between regular uses of static variables and member expressions
uses. For example, the following program compiled and linked successfully:
This commit ensures that constant static variables referenced through member
expressions are emitted in the same way as ordinary static variable references.
Gor Nishanov [Fri, 25 Aug 2017 04:46:54 +0000 (04:46 +0000)]
[coroutines] Support coroutine-handle returning await-suspend (i.e symmetric control transfer)
Summary:
If await_suspend returns a coroutine_handle, as in the example below:
```
coroutine_handle<> await_suspend(coroutine_handle<> h) {
coro.promise().waiter = h;
return coro;
}
```
suspensionExpression processing will resume the coroutine pointed at by that handle.
Related LLVM change rL311751 makes resume calls of this kind `musttail` at any optimization level.
This enables unlimited symmetric control transfer from coroutine to coroutine without blowing up the stack.
projects / clang / log