Douglas Gregor [Fri, 18 Sep 2009 17:42:29 +0000 (17:42 +0000)]
For code completion in C++ member access expressions and tag names,
look into the current scope for anything that could start a
nested-names-specifier. These results are ranked worse than any of the
results actually found in the lexical scope.
Perform a little more pruning of the result set, eliminating
constructors, __va_list_tag, and any duplication of declarations in
the result set. For the latter, implemented
NamespaceDecl::getCanonicalDecl.
Douglas Gregor [Fri, 18 Sep 2009 15:51:54 +0000 (15:51 +0000)]
When gathering results for code completion, only include hidden
results when there is some way to refer to them in the language, such
as with a qualified name in C++.
Douglas Gregor [Fri, 18 Sep 2009 15:37:17 +0000 (15:37 +0000)]
Implement code completion for tags, e.g., code completion after "enum"
will provide the names of various enumerations currently
visible. Introduced filtering of code-completion results when we build
the result set, so that we can identify just the kinds of declarations
we want.
This implementation is incomplete for C++, since we don't consider
that the token after the tag keyword could start a
nested-name-specifier.
Revert most of r82198, which was causing a large number of crashes
when running the analyzer on real projects. We'll keep the change to
AnalysisManager.cpp in r82198 so that -fobjc-gc analyzes code
correctly in both GC and non-GC modes, although this may emit two
diagnostics for each bug in some cases (a better solution will come
later).
Introduce caching of diagnostics in BugReporter. This provides extra
pruning of diagnostics that may be emitted multiple times. This is
accomplished by adding FoldingSet profiling support to PathDiagnostic,
and then having BugReporter record what diagnostics have been issued.
This was motived to a serious bug introduced by moving the
'divide-by-zero' checking outside of GRExprEngine into a separate
'Checker' class. When analyzing code using the '-fobjc-gc' option, a
given function would be analyzed twice, but the second time various
"internal checks" would be disabled to avoid emitting multiple
diagnostics (e.g., "null dereference") for the same issue. The
problem is that such checks also effect path pruning and don't just
emit diagnostics. This resulted in an assertion failure involving a
real divide-by-zero in some analyzed code where we would get an
assertion failure in APInt because the 'DivZero' check was disabled
and didn't prune the logic that resulted in the divide-by-zero in the
analyzer.
The implemented solution is somewhat of a hack, and may not perform
extremely well. This will need to be cleaned up over time.
As a regression test, 'misc-ps.m' has been modified so that its tests
are run using -fobjc-gc to test this diagnostic pruning behavior.
Douglas Gregor [Thu, 17 Sep 2009 21:32:03 +0000 (21:32 +0000)]
Initial implementation of a code-completion interface in Clang. In
essence, code completion is triggered by a magic "code completion"
token produced by the lexer [*], which the parser recognizes at
certain points in the grammar. The parser then calls into the Action
object with the appropriate CodeCompletionXXX action.
Sema implements the CodeCompletionXXX callbacks by performing minimal
translation, then forwarding them to a CodeCompletionConsumer
subclass, which uses the results of semantic analysis to provide
code-completion results. At present, only a single, "printing" code
completion consumer is available, for regression testing and
debugging. However, the design is meant to permit other
code-completion consumers.
This initial commit contains two code-completion actions: one for
member access, e.g., "x." or "p->", and one for
nested-name-specifiers, e.g., "std::". More code-completion actions
will follow, along with improved gathering of code-completion results
for the various contexts.
[*] In the current -code-completion-dump testing/debugging mode, the
file is truncated at the completion point and EOF is translated into
"code completion".
Daniel Dunbar [Thu, 17 Sep 2009 19:55:53 +0000 (19:55 +0000)]
Tweak clang testing.
- Move CMake to using the new test runner.
- Switch Makefiles to use the lit.site.cfg.in template.
- Remove explicit --path arguments, instead this gets written into the site
configuration. This means running lit from the command line should use the
exact same configuration as is used in 'make test', assuming it can find the
site configuration file. You still need to run 'make test' (or the cmake
build target equivalent) at least once.
Douglas Gregor [Thu, 17 Sep 2009 19:51:30 +0000 (19:51 +0000)]
Merge uninstantiated default arguments more carefully, and try not to
complain about specializations of member functions that are not
definitions. Fixes PR4995.
Anders Carlsson [Thu, 17 Sep 2009 03:46:43 +0000 (03:46 +0000)]
When mangling function template specialization, mangle the type encoding of the original function template. Also, port mangle.cpp and function-template-specialization.cpp over to using FileCheck.
Daniel Dunbar [Thu, 17 Sep 2009 03:06:51 +0000 (03:06 +0000)]
PCH: When writing PCH files, tweak the predicate function deciding whether to mark a decl as "external" to be closer to reality.
This still isn't perfect, but I believe it is conservatively accurate at marking decls which IRgen needs to see, while still keeping the "deserialization footprint" on Cocoa.h.
Daniel Dunbar [Thu, 17 Sep 2009 03:06:44 +0000 (03:06 +0000)]
PCH: When deserializing an explicit "external definition", don't pass it to HandleTopLevelDecl -- this is already being done inside the reader.
This is something of a hack, since whether the reader actually did this depends on the "isConsumerInterestedIn" predicate. I think we need to rework how this works, but I need to discuss with Doug.
Daniel Dunbar [Thu, 17 Sep 2009 00:48:13 +0000 (00:48 +0000)]
Initial support for code generation from .ast files.
- Doesn't actually work yet because only module level asm's get correctly marked as externally visible in the PCH.
- Other things like 'clang-cc foo.ast -ast-dump' now work, as well.
Starting patch to generate more specific API for objc's
GC. Currently, new API will be generated under
clang-cc's -fobjc-newgc-api flag which will eventually
become the default. WIP.
Douglas Gregor [Wed, 16 Sep 2009 18:34:49 +0000 (18:34 +0000)]
Teach Sema::FindInstantiatedDecl to find instantiated RecordDecls even
when we are not instantiating the corresponding "current
instantiation." This happens, e.g., when we are instantiating a
declaration reference that refers into the "current instantiation" but
occurs in a default function argument. The libstdc++ vector default
constructor now instantiates properly.
Anders Carlsson [Wed, 16 Sep 2009 15:53:40 +0000 (15:53 +0000)]
x86-64 ABI: If a type is a C++ record with either a non-trivial destructor or a non-trivial copy constructor, it should be passed in a pointer. Daniel, plz review.
Daniel Dunbar [Wed, 16 Sep 2009 01:40:13 +0000 (01:40 +0000)]
A toy for Doug: Add some 'lit' tests which just run clang++ -fsyntax-only over all the files in a directory.
- 'lit utils/C++Tests' to run, you may need to fix the stdc++ include path, and your mileage may vary.
- A handful of the LLVM headers are XFAIL because they don't include the right headers.
Douglas Gregor [Tue, 15 Sep 2009 22:30:29 +0000 (22:30 +0000)]
When implicitly declaring operators new, new[], delete, and delete[],
give them the appropriate exception specifications. This,
unfortunately, requires us to maintain and/or implicitly generate
handles to namespace "std" and the class "std::bad_alloc". However,
every other approach I've come up with was more hackish, and this
standard requirement itself is quite the hack.
1) don't do overload resolution in selecting conversion
to pointer function for delete expression. 2)
Treat type conversion function and its 'const' version
as identical in building the visible conversion list.
projects / clang / log