Build up statistics about the work done for analysis based warnings.
Special detail is added for uninitialized variable analysis as this has
serious performance problems than need to be tracked.
Computing some of this data is expensive, for example walking the CFG to
determine its size. To avoid doing that unless the stats data is going
to be used, we thread a bit into the Sema object to track whether
detailed stats should be collected or not. This bit is used to avoid
computations whereever the computations are likely to be more expensive
than checking the state of the flag. Thus, counters are in some cases
unconditionally updated, but the more expensive (and less frequent)
aggregation steps are skipped.
With this patch, we're able to see that for 'gcc.c':
*** Analysis Based Warnings Stats:
232 functions analyzed (0 w/o CFGs).
7151 CFG blocks built.
30 average CFG blocks per function.
1167 max CFG blocks per function.
163 functions analyzed for uninitialiazed variables
640 variables analyzed.
3 average variables per function.
94 max variables per function.
96409 block visits.
591 average block visits per function.
61546 max block visits per function.
And for the reduced testcase in PR10183:
*** Analysis Based Warnings Stats:
98 functions analyzed (0 w/o CFGs).
8526 CFG blocks built.
87 average CFG blocks per function.
7277 max CFG blocks per function.
68 functions analyzed for uninitialiazed variables
1359 variables analyzed.
19 average variables per function.
1196 max variables per function. 2540494 block visits.
37360 average block visits per function. 2536495 max block visits per function.
That last number is the somewhat scary one that indicates the problem in
PR10183.
Douglas Gregor [Wed, 6 Jul 2011 16:00:34 +0000 (16:00 +0000)]
Teach the static analyzer's interpretation of Cocoa conventions to
obey the objc_method_family attribute when provided. Fixes
<rdar://problem/9726279>.
John McCall [Wed, 6 Jul 2011 07:30:07 +0000 (07:30 +0000)]
When tree-transforming an expression sequence, always flag expanded
variadic argument pack expansions as having changed, rather than doing
it for each changed expansion, which leaves out zero-argument packs
with catastrophic consequences.
John McCall [Wed, 6 Jul 2011 06:57:57 +0000 (06:57 +0000)]
Fixed enum types can be complete without actually being valid to use
as scope specifiers; diagnose the attempt, rather than letting it go
to an assert. The rest of PR10264.
John McCall [Wed, 6 Jul 2011 05:58:41 +0000 (05:58 +0000)]
Properly protect colons when parsing a nested-name-specifier as part
of an enum specifier in dialects which permit fixed underlying types.
Fixes the rejects-valid part of PR10264.
Douglas Gregor [Wed, 6 Jul 2011 03:00:34 +0000 (03:00 +0000)]
Improve the Python bindings for libclang in a few ways, from Eli
Bendersky. Specifically:
* Implemented a new function in libclang: clang_isAttribute
* Fixing TranslationUnit.get_includes to only go through the argument
* buffer when it contains something. This fixed a crash on Windows
* clang_getFileName returns CXString, not char*. Made appropriate
* fixes in cindex.py - now the relevant tests pass and we can see the
* full locations correctly again (previously there was garbage in
* place of the file name)
* Exposed clang_getCursorDisplayName to the python bindings
John McCall [Wed, 6 Jul 2011 01:22:26 +0000 (01:22 +0000)]
Call objc_terminate() instead of abort() when a cleanup throws an
exception in Objective-C; in Objective-C++ we still use std::terminate().
This is only available in very recent runtimes.
John McCall [Wed, 6 Jul 2011 00:26:06 +0000 (00:26 +0000)]
Change the driver's logic about Objective-C runtimes: abstract out a
structure to hold inferred information, then propagate each invididual
bit down to -cc1. Separate the bits of "supports weak" and "has a native
ARC runtime"; make the latter a CodeGenOption.
The tool chain is still driving this decision, because it's the place that
has the required deployment target information on Darwin, but at least it's
better-factored now.
Douglas Gregor [Tue, 5 Jul 2011 18:30:26 +0000 (18:30 +0000)]
Look through parenthesized declarators when determining whether an
instantiated function template was written with a prototype or via
some kind of typedef. Fixes PR10273 / <rdar://problem/9723679>.
Switch the Decl and Stmt stats printing to use llvm::errs() instead of
fprintf, and to be more consistent in formatting with the other stats
printing routines.
Eli Friedman [Sat, 2 Jul 2011 00:34:19 +0000 (00:34 +0000)]
Make clang behave in a gcc-compatible way in the presence of multiple flags for the same x86 target feature (e.g. -mno-sse -msse). gcc uses a somewhat unintuitive algorithm here in that the enabled SSE instructions is based on the order of the *last* flag for *each* feature-level, so that "-mno-sse -msse2" only enables SSE2, but "-mno-sse -msse2 -msse" enables all SSE levels.
Move the Sema argument to all of the Sema-using helper functions in
SemaDeclAttr to the first argument. This makes them follow the very
consistent policy elsewhere in Sema for helper functions.
Original patch by Caitlin Sadowski, with some tweaking by me.
Mechanical rename of 'd' Decl pointer parameters to 'D'. This is more
conventional in the rest of Clang's codebase, and closer to the current
style recommendations. It also makes the code more internally consistent
as FD, VD, etc are used frequently for particular decl variables.
-Fix mistake in ASTContext::getInnerObjCOwnership noticed by Doug
-Remove unnecessary 'return'.
-Remove unnecessary 'if' check (llvm_unreachable make sure attrStr will be non-null)
-Add a test of transferring ownership to a reference cast type.
[ARC] When casting from a pointer to an objective-c object with known ownership, if the
cast type has no ownership specified, implicitly "transfer" the ownership of the cast'ed type
to the cast type:
id x;
(NSString**)&x; // Casting as (__strong NSString**).
-Remove Sema::ActOnCastOfParenListExpr and move most of its functionality to
newly introduced Sema::BuildVectorLiteral.
-Make Sema::ActOnCastExpr handle a vector initializer both when the cast'ed expression
is a ParenListExpr and when it is a ParenExpr.
-Ultimately make Sema::ActOnParenOrParenListExpr independent of what the cast type was.
[ARC] When casting from a pointer to an objective-c object with known ownership, if the
cast type has no ownership specified, implicitly "transfer" the ownership of the cast'ed type
to the cast type:
id x;
static_cast<NSString**>(&x); // Casting as (__strong NSString**).
This currently only works for C++ named casts, C casts to follow.
Douglas Gregor [Fri, 1 Jul 2011 21:54:36 +0000 (21:54 +0000)]
Emit guard variables for any weak global that has a run-time
initializer. Previously, we only used guard variables for weak static
data members. Fixes <rdar://problem/9692249>.
Douglas Gregor [Fri, 1 Jul 2011 21:27:45 +0000 (21:27 +0000)]
When adding boolean keywords for typo correction, add either "bool" or
"_Bool" (depending on dialect), but not both, since they have the same
edit distance from "Bool".
Douglas Gregor [Fri, 1 Jul 2011 21:08:19 +0000 (21:08 +0000)]
Don't zero-initialize default-initialized local variables that have
trivial default constructors. This generated-code regression was
caused by r131796, which had simplified the handling of default
initialization in Sema. Fixes <rdar://problem/9694300>.
Richard Trieu [Fri, 1 Jul 2011 20:54:02 +0000 (20:54 +0000)]
For code such as:
int f(int x) {
if (int foo = f(bar)) {}
return 0;
}
Clang produces the following error messages:
paren_imbalance.cc:2:19: error: use of undeclared identifier 'bar'
if (int foo = f(bar)) {}
^
paren_imbalance.cc:2:26: error: expected ')'
if (int foo = f(bar)) {}
^
paren_imbalance.cc:2:6: note: to match this '('
if (int foo = f(bar)) {}
^
The second error is incorrect. This patch will stop Clang from producing an error on parenthesis imbalance during error recovery when there isn't one.
Richard Trieu [Fri, 1 Jul 2011 20:02:53 +0000 (20:02 +0000)]
Fix for PR7410. Allow functions in a derived class that improperly overwrite a virtual function in the base class to be inserted into the derived class function list to prevent extra errors every time the derived class is used.
Richard Smith [Fri, 1 Jul 2011 19:46:12 +0000 (19:46 +0000)]
Fix AST representations of alias-declarations which define tag types. Inside classes, the tag types need to have an associated access specifier, and inside function definitions, they need to be included in the declarations of the DeclStmt. These issues manifested as assertions during template instantiation, and also in a WIP constexpr patch.
Douglas Gregor [Fri, 1 Jul 2011 18:22:13 +0000 (18:22 +0000)]
When we create a precompiled preamble, don't copy the
CompilerInvocation on the stack, because other objects (e.g., the
CompilerInstance) maintain an intrusive reference-counted pointer to
the CompilerInvocation. This doesn't matter in the normal case,
because we take back the CompilerInvocation. However, during crash
recovery, this leads to us trying to free an object on the stack, and
hilarity ensues. Fixes <rdar://problem/9652540>.
John McCall [Fri, 1 Jul 2011 02:19:08 +0000 (02:19 +0000)]
Just mangle substituted template parameter types as unresolved types.
This is kindof questionable but seems to do more-or-less the right thing.
This is not a particularly friendly part of the ABI.
Douglas Gregor [Fri, 1 Jul 2011 01:22:09 +0000 (01:22 +0000)]
Introduce the notion of instantiation dependence into Clang's AST. A
type/expression/template argument/etc. is instantiation-dependent if
it somehow involves a template parameter, even if it doesn't meet the
requirements for the more common kinds of dependence (dependent type,
type-dependent expression, value-dependent expression).
When we see an instantiation-dependent type, we know we always need to
perform substitution into that instantiation-dependent type. This
keeps us from short-circuiting evaluation in places where we
shouldn't, and lets us properly implement C++0x [temp.type]p2.
In theory, this would also allow us to properly mangle
instantiation-dependent-but-not-dependent decltype types per the
Itanium C++ ABI, but we aren't quite there because we still mangle
based on the canonical type in cases like, e.g.,
John McCall [Fri, 1 Jul 2011 00:04:39 +0000 (00:04 +0000)]
Change the mangling of enclosing template template parameters
that serve as the base template name of an unresolved-name to
be mangled as a substitution.
Replace an unreachable error path with an assert
(SourceManager::createFileID cannot return an invalid file ID).
Also update a comment to reflect this.
Chandler Carruth [Thu, 30 Jun 2011 08:14:54 +0000 (08:14 +0000)]
Fix an obvious typo in an attribute's diagnostics.
Patch by Caitlin Sadowski.
Unfortunately, this attribute doesn't seem to have a single test. It is
only mentioned in comments in one test, and as a string literal in
a copy of some Clang code checked in as a test for the Indexer. =[ It
dates from 2009 r74280 as part of OpenCL 1.0.
Introduce a caching mechanism for macro expanded tokens.
Previously macro expanded tokens were added to Preprocessor's bump allocator and never released,
even after the TokenLexer that were lexing them was finished, thus they were wasting memory.
A very "useful" boost library was causing clang to eat 1 GB just for the expanded macro tokens.
Introduce a special cache that works like a stack; a TokenLexer can add the macro expanded tokens
in the cache, and when it finishes, the tokens are removed from the end of the cache.
Now consumed memory by expanded tokens for that library is ~ 1.5 MB.
Douglas Gregor [Wed, 29 Jun 2011 21:22:02 +0000 (21:22 +0000)]
When redeclaring a local extern in the same scope, make sure that we
replace the existing declaration appropriately. Patch by Jordy Rose,
fixes PR10013 / <rdar://problem/9584157>.
David Chisnall [Wed, 29 Jun 2011 13:17:23 +0000 (13:17 +0000)]
Make ARC support default to true. Please can we stop making Objective-C runtime features part of the platform description? This only makes sense for Darwin.
projects / clang / log