Jordan Rose [Thu, 12 Jul 2012 00:16:25 +0000 (00:16 +0000)]
[analyzer] Don't inline virtual calls unless we can devirtualize properly.
Previously we were using the static type of the base object to inline
methods, whether virtual or non-virtual. Now, we try to see if the base
object has a known type, and if so ask for its implementation of the method.
Richard Smith [Wed, 11 Jul 2012 22:37:56 +0000 (22:37 +0000)]
Stop instantiating a class if we hit a static_assert failure. Also, if the
static_assert fails when parsing the template, don't diagnose it again on every
instantiation.
Dmitri Gribenko [Wed, 11 Jul 2012 21:38:39 +0000 (21:38 +0000)]
Enable comment parsing and semantic analysis to emit diagnostics. A few
diagnostics implemented -- see testcases.
I created a new TableGen file for comment diagnostics,
DiagnosticCommentKinds.td, because comment diagnostics don't logically
fit into AST diagnostics file. But I don't feel strongly about it.
This also implements support for self-closing HTML tags in comment
lexer and parser (for example, <br />).
In order to issue precise diagnostics CommentSema needs to know the
declaration the comment is attached to. There is no easy way to find a decl by
comment, so we match comments and decls in lockstep: after parsing one
declgroup we check if we have any new, not yet attached comments. If we do --
then we do the usual comment-finding process.
It is interesting that this automatically handles trailing comments.
We pick up not only comments that precede the declaration, but also
comments that *follow* the declaration -- thanks to the lookahead in
the lexer: after parsing the declgroup we've consumed the semicolon
and looked ahead through comments.
Added -Wdocumentation-html flag for semantic HTML errors to allow the user to
disable only HTML warnings (but not HTML parse errors, which we emit as
warnings in -Wdocumentation).
Introduce a flag in SourceManager to treat non-system source files
as "volatile", meaning there's a high enough chance that they may
change while we are trying to use them.
This flag is only enabled by libclang.
Currently "volatile" source files will be stat'ed immediately
before opening them, because the file size stat info
may not be accurate since when we got it (e.g. from the PCH).
This avoids crashes when trying to reference mmap'ed memory
from a file whose size is not what we expect.
Note that there's still a window for a racing issue to occur
but the window for it should be way smaller than before.
We can consider later on to avoid mmap completely on such files.
Jordan Rose [Wed, 11 Jul 2012 19:58:23 +0000 (19:58 +0000)]
Allow -verify directives to be filtered by preprocessing.
This is accomplished by making VerifyDiagnosticsConsumer a CommentHandler,
which then only reads the -verify directives that are actually in live
blocks of code. It also makes it simpler to handle -verify directives that
appear in header files, though we still have to manually reparse some files
depending on how they are generated.
This requires some test changes. In particular, all PCH tests now have their
-verify directives outside the "header" portion of the file, using the @line
syntax added in r159978. Other tests have been modified mostly to make it
clear what is being tested, and to prevent polluting the expected output with
the directives themselves.
Patch by Andy Gibbs! (with slight modifications)
The new Frontend/verify-* tests exercise the functionality of this commit,
as well as r159978, r159979, and r160053 (Andy's other -verify enhancements).
Daniel Jasper [Wed, 11 Jul 2012 19:13:13 +0000 (19:13 +0000)]
Add a hook to supply a custom CompilationDatabase. To add a custom CompilationDatabase, make it implement findCompilationDatabaseForDirectory in CustomCompilationDatabase.h and set USE_COSTUM_COMPILATION_DATABASE.
Chad Rosier [Wed, 11 Jul 2012 19:08:21 +0000 (19:08 +0000)]
[driver] Fix so that clang can find correct path prefix for libc object files
from GNU binutils supporting multi-arch folder for ARM target.
Patch by Jiangning Liu <jiangning.liu@arm.com>.
Jordan Rose [Wed, 11 Jul 2012 16:50:36 +0000 (16:50 +0000)]
Emit -verify diagnostics even when we have a fatal error.
Previously we'd halt at the fatal error as expected, but not actually emit
any -verify-related diagnostics. This lets us catch cases that emit a
/different/ fatal error from the one we expected.
This is implemented by adding a "force emit" mode to DiagnosticBuilder, which
will cause diagnostics to immediately be emitted regardless of current
suppression. Needless to say this should probably be used /very/ sparingly.
Patch by Andy Gibbs! Tests for all of Andy's -verify patches coming soon.
Rafael Espindola [Wed, 11 Jul 2012 02:15:51 +0000 (02:15 +0000)]
Don't process #pragma visibility during instantiation. The visibility of the
instantiation depends on the template, its arguments and parameters, but not
where it is instantiated.
Eric Christopher [Wed, 11 Jul 2012 01:49:26 +0000 (01:49 +0000)]
The end of a block doesn't necessarily need a line table entry unless
there's something going on there. Remove the unconditional line entry
and only add one if we're emitting cleanups (any other statements
would be handled normally).
Jordan Rose [Tue, 10 Jul 2012 22:08:01 +0000 (22:08 +0000)]
[analyzer] Construct stack variables directly in their VarDecl.
Also contains a number of tweaks to inlining that are necessary
for constructors and destructors. (I have this enabled on a private
branch, but it is very much unstable.)
Jordan Rose [Tue, 10 Jul 2012 22:07:57 +0000 (22:07 +0000)]
[analyzer] Use CallEvent for building inlined stack frames.
In order to accomplish this, we now build the callee's stack frame
as part of the CallEnter node, rather than the subsequent BlockEdge node.
This should not have any effect on perceived behavior or diagnostics.
This makes it safe to re-enable inlining of member overloaded operators.
Jordan Rose [Tue, 10 Jul 2012 22:07:47 +0000 (22:07 +0000)]
[analyzer] Add a CXXDestructorCall CallEvent.
While this work is still fairly tentative (destructors are still left out of
the CFG by default), we now handle destructors in the same way as any other
calls, instead of just automatically trying to inline them.
Jordan Rose [Tue, 10 Jul 2012 22:07:42 +0000 (22:07 +0000)]
[analyzer] Add new PreImplicitCall and PostImplicitCall ProgramPoints.
These are currently unused, but are intended to be used in lieu of PreStmt
and PostStmt when the call is implicit (e.g. an automatic object destructor).
This also modifies the Data1 field of ProgramPoints to allow storing any
pointer-sized value, as opposed to only aligned pointers. This is necessary
to store SourceLocations.
There is currently no BugReporter support for these; they should be skipped
over in any diagnostic output.
This commit also tags checkers that currently rely on function calls only
occurring at StmtPoints.
Axel Naumann [Tue, 10 Jul 2012 16:50:27 +0000 (16:50 +0000)]
Improve r159256 following Chandler's comments:
Implement UniqueFileContainer::erase(), camelCase, add comment on future optimizations of the cache versus de-optimizations of invalidations.
Axel Naumann [Tue, 10 Jul 2012 16:24:07 +0000 (16:24 +0000)]
Clear diagnostic counts after processing the diagnostics.
The consumer might see multiple input files (e.g. for cling) and since r159977 the count is maintained across input files.
Manuel Klimek [Tue, 10 Jul 2012 13:10:51 +0000 (13:10 +0000)]
Adds support for auto-detection of compilation databases
from a source file and changes clang-check to make use of this.
This makes clang-check just work on in-tree builds, and allows
easy setup via a symlink per source directory to make clang-check
work without any extra configuration.
Chris Lattner [Tue, 10 Jul 2012 05:03:05 +0000 (05:03 +0000)]
Jordan points out that this was incorrect: clang should recover from
*errors* with fixits on them by following the recovery advised by the
fixit, but if it is a fixit on a warning, then obviously the AST
should be for the code as-written.
Jim Grosbach [Mon, 9 Jul 2012 18:34:21 +0000 (18:34 +0000)]
Tests: check for target availability for target-specific tests.
Lots of tests are using an explicit target triple w/o first checking that the
target is actually available. Add a REQUIRES clause to a bunch of them. This should
hopefully unbreak bots which don't configure w/ all targets enabled.
Jordan Rose [Mon, 9 Jul 2012 16:54:53 +0000 (16:54 +0000)]
Better parser recovery in Objective-C containers.
Previously it was possible to get an infinite-loop-on-invalid with a namespace
decl within @interface. Since 'namespace' is normally a safe place to retry
top-level parsing, we just didn't consume the token.
This adds a flag that tracks whether we have temporarily left Objective-C
scope to parse a C-like declaration, and uses that to better recover from
parse problems by stopping at possible method declarations and at @end. To
fix the original problem, we do /not/ stop at 'namespace' when in an
Objective-C @interface or @protocol context (but still do in @implementation).
Jordan Rose [Mon, 9 Jul 2012 16:54:49 +0000 (16:54 +0000)]
[analyzer] When inlining, make sure we use the definition decl.
This was a regression introduced during the CallEvent changes; a call to
FunctionDecl::hasBody was also being used to replace the decl found by
lookup with the actual definition. To keep from making this mistake again
(particularly if/when we start inlining Objective-C methods), this commit
adds a "getDefinition()" method to CallEvent, which should do the right
thing under any circumstances.
Jordan Rose [Mon, 9 Jul 2012 16:54:44 +0000 (16:54 +0000)]
Downgrade the "direct comparison" error for ObjC literals to a warning.
Chris pointed out that while the comparison is certainly problematic
and does not have well-defined behavior, it isn't any worse than some
of the other abuses that we merely warn about and doesn't need to make
the compilation fail.
Revert the release notes change (r159766) now that this is just a new warning.
Richard Smith [Mon, 9 Jul 2012 03:07:20 +0000 (03:07 +0000)]
PR13136:
* When substituting a reference to a non-type template parameter pack where the
corresponding argument is a pack expansion, transform into an expression
which contains an unexpanded parameter pack rather than into an expression
which contains a pack expansion. This causes the SubstNonTypeTemplateParmExpr
to be inside the PackExpansionExpr, rather than outside, so the expression
still looks like a pack expansion and can be deduced.
* Teach MarkUsedTemplateParameters that we can deduce a reference to a template
parameter if it's wrapped in a SubstNonTypeTemplateParmExpr (such nodes are
added during alias template substitution).
Richard Smith [Sun, 8 Jul 2012 04:37:51 +0000 (04:37 +0000)]
PR13243: When deducing a non-type template parameter which is specified as an
expression, skip over any SubstNonTypeTemplateParmExprs which alias templates
may have inserted before checking for a DeclRefExpr referring to a non-type
template parameter declaration.
Richard Smith [Sun, 8 Jul 2012 02:38:24 +0000 (02:38 +0000)]
PR9793: Treat substitution as an instantiation step for the purpose of the
-ftemplate-depth limit. There are various ways to get an infinite (or merely
huge) stack of substitutions with no intervening instantiations. This is also
consistent with gcc's behavior.
Richard Smith [Sat, 7 Jul 2012 22:48:24 +0000 (22:48 +0000)]
PR13290: Constant-evaluation support for CXXConstructExprs which construct a
multidimensional array of class type. Also, preserve zero-initialization when
evaluating an initializer list for an array, in case the initializers refer to
later elements (which have preceding zero-initialization).
Richard Smith [Sat, 7 Jul 2012 08:35:56 +0000 (08:35 +0000)]
PR12670: Support for initializing an array of non-aggregate class type from an
initializer list. Patch by Olivier Goffart, with extra testcases by Meador Inge
and Daniel Lunow.
Richard Smith [Sat, 7 Jul 2012 06:59:51 +0000 (06:59 +0000)]
When marking virtual functions as used for a class' vtable, mark all functions
which will appear in the vtable as used, not just those ones which were
declared within the class itself. Fixes an issue reported as comment#3 in
PR12763 -- we sometimes assert in codegen if we try to emit a reference to a
function declaration which we've not marked as referenced. This also matches
gcc's observed behavior.
John McCall [Sat, 7 Jul 2012 06:41:13 +0000 (06:41 +0000)]
Distinguish more carefully between free functions and C++ instance methods
in the ABI arrangement, and leave a hook behind so that we can easily
tweak CCs on platforms that use different CCs by default for C++
instance methods.
Enhance 'diagtool list-warnings' to report number of diagnostics covered directly under -Wpedantic, and enhance warning-flags.c test to test that this set does not grow.
- Split pedantic driver flag test into separate test file, and XFAIL on cygwin,mingw32
- Fix bug in tablegen logic where a missing '{' caused errors to be included in -Wpedantic.
Implement -Wpedantic and --no-pedantic to complement -Weverything.
This patch introduces some magic in tablegen to create a "Pedantic" diagnostic
group which automagically includes all warnings that are extensions. This
allows a user to suppress specific warnings traditionally under -pedantic used
an ordinary warning flag. This also allows users to use #pragma to silence
specific -pedantic warnings, or promote them to errors, within blocks of text
(just like any other warning).
-Wpedantic is NOT an alias for -pedantic. Instead, it provides another way
to (a) activate -pedantic warnings and (b) disable them. Where they differ
is that -pedantic changes the behavior of the preprocessor slightly, whereas
-Wpedantic does not (it just turns on the warnings).
The magic in the tablegen diagnostic emitter has to do with computing the minimal
set of diagnostic groups and diagnostics that should go into -Wpedantic, as those
diagnostics that already members of groups that themselves are (transitively) members
of -Wpedantic do not need to be included in the Pedantic group directly. I went
back and forth on whether or not to magically generate this group, and the invariant
was that we always wanted extension warnings to be included in -Wpedantic "some how",
but the bookkeeping would be very onerous to manage by hand.
-no-pedantic (and --no-pedantic) is included for completeness, and matches many of the
same kind of flags the compiler already supports. It does what it says: cancels out
-pedantic. One discrepancy is that if one specifies --no-pedantic and -Weverything or
-Wpedantic the pedantic warnings are still enabled (essentially the -W flags win). We
can debate the correct behavior here.
Along the way, this patch nukes some code in TextDiagnosticPrinter.cpp and CXStoredDiagnostic.cpp
that determine whether to include the "-pedantic" flag in the warning output. This is
no longer needed, as all extensions now have a -W flag.
This patch also significantly reduces the number of warnings not under flags from 229
to 158 (all extension warnings). That's a 31% reduction.
Jordan Rose [Fri, 6 Jul 2012 22:00:04 +0000 (22:00 +0000)]
Constify the argument to SourceLocation::getFromPtrEncoding.
This allows SourceLocations to be stored in generic "data" fields
that are typed as "const void *" and are also used to point to
const objects.
Really we should probably be returning a const pointer from
getPtrEncoding as well, but in some places we want to store
SourceLocations in the same generic "data" field as proper
pointers to /mutable/ objects. Oh well.
projects / clang / log