Manman Ren [Wed, 20 Mar 2013 16:59:38 +0000 (16:59 +0000)]
Exploit this-return of a callsite in a this-return function.
For constructors/desctructors that return 'this', if there exists a callsite
that returns 'this' and is immediately before the return instruction, make
sure we are using the return value from the callsite.
We don't need to keep 'this' alive through the callsite. It also enables
optimizations in the backend, such as tail call optimization.
Rafael Espindola [Wed, 20 Mar 2013 03:05:54 +0000 (03:05 +0000)]
Remove the Tools DensMap from the toolchain.
Each toolchain has a set of tools, but they are all of known types. It can
have a linker, an assembler, a "clang" (compile, analyze, ...) a non-clang
compiler, etc.
Instead of keeping a map, just have member variable for each type of tool.
Nick Lewycky [Wed, 20 Mar 2013 02:14:38 +0000 (02:14 +0000)]
The flag "-coverage-function-names-in-data" is actually backwards -- we do
emit function names in .gcda files by default, and the flag turns that off!
Rename the flag to make it match what it actually does. This keeps the default
format compatible with gcc 4.2.
Reid Kleckner [Wed, 20 Mar 2013 01:40:23 +0000 (01:40 +0000)]
[ms-cxxabi] Fix assertion on unhandled function template arg types
This code was changed in r158376 to get template argument source info
for better diagnostics, but the current code asserts for any kind of
unsupported template argument before it can issue a diagnostic. This change
goes back to the Itanium implementation of isTemplate() and puts the argument
index into the diagnostic instead of a source location.
Nick Lewycky [Wed, 20 Mar 2013 01:38:16 +0000 (01:38 +0000)]
Make clang emit linkage names in debug info for subprograms when coverage info
is enabled. Also add a new -test-coverage cc1 flag which makes testing coverage
possible and add our first clang-side coverage test.
Jordan Rose [Wed, 20 Mar 2013 00:35:37 +0000 (00:35 +0000)]
[analyzer] Re-apply "Do part of the work to find shortest bug paths up front".
With the assurance that the trimmed graph does not contain cycles,
this patch is safe (with a few tweaks), and provides the performance
boost it was intended to.
Part of performance work for <rdar://problem/13433687>.
Jordan Rose [Wed, 20 Mar 2013 00:35:31 +0000 (00:35 +0000)]
[analyzer] Break cycles (optionally) when trimming an ExplodedGraph.
Having a trimmed graph with no cycles (a DAG) is much more convenient for
trying to find shortest paths, which is exactly what BugReporter needs to do.
Part of the performance work for <rdar://problem/13433687>.
Douglas Gregor [Wed, 20 Mar 2013 00:22:05 +0000 (00:22 +0000)]
<rdar://problem/10796651> Introduce configuration macros into module maps.
Configuration macros are macros that are intended to alter how a
module works, such that we need to build different module variants
for different values of these macros. A module can declare its
configuration macros, in which case we will complain if the definition
of a configation macro on the command line (or lack thereof) differs
from the current preprocessor state at the point where the module is
imported. This should eliminate some surprises when enabling modules,
because "#define CONFIG_MACRO ..." followed by "#include
<module/header.h>" would silently ignore the CONFIG_MACRO setting. At
least it will no longer be silent about it.
Configuration macros are eventually intended to help reduce the number
of module variants that need to be built. When the list of
configuration macros for a module is exhaustive, we only need to
consider the settings for those macros when building/finding the
module, which can help isolate modules for various project-specific -D
flags that should never affect how modules are build (but currently do).
David Blaikie [Tue, 19 Mar 2013 23:10:14 +0000 (23:10 +0000)]
Simplify/generalize some debug info test cases
Mostly, try to depend on the annotation comments more so these tests are more
legible, brief, and agnostic to schema changes in the future (sure, they're not
agnostic to changes to the comment annotations but since they're easier to read
they should be easier to update if that happens).
Anna Zaks [Tue, 19 Mar 2013 22:38:09 +0000 (22:38 +0000)]
[analyzer] Do not believe lazy binding when symbolic region types do not match
This fixes a crash when analyzing LLVM that was exposed by r177220 (modeling of
trivial copy/move assignment operators).
When we look up a lazy binding for “Builder”, we see the direct binding of Loc at offset 0.
Previously, we believed the binding, which led to a crash. Now, we do not believe it as
the types do not match.
Jordan Rose [Tue, 19 Mar 2013 22:10:35 +0000 (22:10 +0000)]
Revert "[analyzer] Do part of the work to find shortest bug paths up front."
The whole reason we were doing a BFS in the first place is because an
ExplodedGraph can have cycles. Unfortunately, my removeErrorNode "update"
doesn't work at all if there are cycles.
I'd still like to be able to avoid doing the BFS every time, but I'll come
back to it later.
Objective-C [qoi]: Provide improved parse diagnostics when
closing rbrace is missing in an ObjC class declaration.
Can do beter than this, but it involves addition of
overhead which will be present in correct code.
// rdar://6854840
Tobias Grosser [Tue, 19 Mar 2013 15:30:48 +0000 (15:30 +0000)]
cindex.py: Handle NULL pointers when parsing translation units
The code inside cindex.py was comparing NULL pointer returned by
clang_parseTranslationUnit and clang_createTranslationUnit with None.
However, as illustrated by the two tests I've added, those conditions
were ineffective which resulted in assert triggering later on.
Instead, a pointer is now treated as a boolean value, a NULL pointer being
False.
Douglas Gregor [Tue, 19 Mar 2013 00:28:20 +0000 (00:28 +0000)]
<rdar://problem/13363214> Eliminate race condition between module rebuild and the global module index.
The global module index was querying the file manager for each of the
module files it knows about at load time, to prune out any out-of-date
information. The file manager would then cache the results of the
stat() falls used to find that module file.
Later, the same translation unit could end up trying to import one of the
module files that had previously been ignored by the module cache, but
after some other Clang instance rebuilt the module file to bring it
up-to-date. The stale stat() results in the file manager would
trigger a second rebuild of the already-up-to-date module, causing
failures down the line.
The global module index now lazily resolves its module file references
to actual AST reader module files only after the module file has been
loaded, eliminating the stat-caching race. Moreover, the AST reader
can communicate to its caller that a module file is missing (rather
than simply being out-of-date), allowing us to simplify the
module-loading logic and allowing the compiler to recover if a
dependent module file ends up getting deleted.
Richard Smith [Tue, 19 Mar 2013 00:01:12 +0000 (00:01 +0000)]
PR15383: When -fsanitize=float-cast-overflow checks a float-to-int conversion,
it wasn't taking into account that the float should be truncated *before* the
range check happens. Thus (unsigned)-0.99 and (unsigned char)255.9 have defined
behavior and should not be trapped.
Jordan Rose [Mon, 18 Mar 2013 23:34:37 +0000 (23:34 +0000)]
[analyzer] Do part of the work to find shortest bug paths up front.
Splitting the graph trimming and the path-finding (r177216) already
recovered quite a bit of performance lost to increased suppression.
We can still do better by also performing the reverse BFS up front
(needed for shortest-path-finding) and only walking the shortest path
for each report. This does mean we have to walk back up the path and
invalidate all the BFS numbers if the report turns out to be invalid,
but it's probably still faster than redoing the full BFS every time.
More performance work for <rdar://problem/13433687>
Richard Smith [Mon, 18 Mar 2013 22:52:47 +0000 (22:52 +0000)]
Add missing diagnostic for a nested-name-specifier on a free-standing type definition. Bump some related diagnostics from warning to extension in C++, since they're errors there. Add some missing checks for function specifiers on non-function declarations.
Reed Kotler [Mon, 18 Mar 2013 22:18:00 +0000 (22:18 +0000)]
This code works around what appears to be a bug in another part of clang.
I have filed http://llvm.org/bugs/show_bug.cgi?id=15538 against clang.
This code is safer anyway because "cast" assumes you really know that
it's okay to make the cast. In this case isa should not be false and
dyn_cast should not return null as far as I understand. But everything
else is valid so I did not want to revert my previous patch for attributes
mips16/nomips16 or use an llvm_unreachable here which would make a number
of our tests fail for mips.
Richard Smith [Mon, 18 Mar 2013 21:44:56 +0000 (21:44 +0000)]
C++11 status: mark 'extended integral types' as N/A, since we don't support
any, and add a note that we cannot treat __int128 as an extended integral type.
Objective-C modern translator. Don't put line info
into the pre-preprocessed file to be passed to
modern translator when compiling in no debug mode.
// rdar://13138170
David Blaikie [Sun, 17 Mar 2013 20:29:22 +0000 (20:29 +0000)]
Generalize a few debug info test cases
Checking for the annotation comment rather than the metadata values makes these
tests resilient to a coming refactor that will pull these fields out into a
separate metadata node.
Jordan Rose [Sat, 16 Mar 2013 02:14:06 +0000 (02:14 +0000)]
[analyzer] Model trivial copy/move assignment operators with a bind as well.
r175234 allowed the analyzer to model trivial copy/move constructors as
an aggregate bind. This commit extends that to trivial assignment
operators as well. Like the last commit, one of the motivating factors here
is not warning when the right-hand object is partially-initialized, which
can have legitimate uses.
Jordan Rose [Sat, 16 Mar 2013 01:07:58 +0000 (01:07 +0000)]
[analyzer] Separate graph trimming from creating the single-path graph.
When we generate a path diagnostic for a bug report, we have to take the
full ExplodedGraph and limit it down to a single path. We do this in two
steps: "trimming", which limits the graph to all the paths that lead to
this particular bug, and "creating the report graph", which finds the
shortest path in the trimmed path to any error node.
With BugReporterVisitor false positive suppression, this becomes more
expensive: it's possible for some paths through the trimmed graph to be
invalid (i.e. likely false positives) but others to be valid. Therefore
we have to run the visitors over each path in the graph until we find one
that is valid, or until we've ruled them all out. This can become quite
expensive.
This commit separates out graph trimming from creating the report graph,
performing the first only once per bug equivalence class and the second
once per bug report. It also cleans up that portion of the code by
introducing some wrapper classes.
This seems to recover most of the performance regression described in my
last commit.
Jordan Rose [Sat, 16 Mar 2013 01:07:47 +0000 (01:07 +0000)]
[analyzer] Don't repeat a bug equivalence class if every report is invalid.
I removed this check in the recursion->iteration commit, but forgot that
generatePathDiagnostic may be called multiple times if there are multiple
PathDiagnosticConsumers.
Manman Ren [Sat, 16 Mar 2013 00:11:09 +0000 (00:11 +0000)]
Exploit this-return of a callsite in a this-return function.
For constructors/desctructors that return 'this', if there exists a callsite
that returns 'this' and is immediately before the return instruction, make
sure we are using the return value from the callsite.
We don't need to keep 'this' alive through the callsite. It also enables
optimizations in the backend, such as tail call optimization.
Richard Trieu [Fri, 15 Mar 2013 23:55:09 +0000 (23:55 +0000)]
Improve template diffing handling of default integer values.
When the template argument is both default and value dependent, the expression
retrieved for the default argument cannot be evaluated, thus never matching
any argument value. To get the proper value, get the template argument
from the desugared template specialization. Also, output the original
expression to provide more information about the argument mismatch.
Anna Zaks [Fri, 15 Mar 2013 23:34:29 +0000 (23:34 +0000)]
[analyzer] Use isLiveRegion to determine when SymbolRegionValue is dead.
Fixes a FIXME, improves dead symbol collection, suppresses a false positive,
which resulted from reusing the same symbol twice for simulation of 2 calls to the same function.
Fixing this lead to 2 possible false negatives in CString checker. Since the checker is still alpha and
the solution will not require revert of this commit, move the tests to a FIXME section.
Ted Kremenek [Fri, 15 Mar 2013 23:09:37 +0000 (23:09 +0000)]
Fix buffer underrun (invalid read) triggered during diagnostic rendering. The test would overflow when computing '0 - 1'.
I don't have a good testcase for this that does not depend on system headers.
It did not trigger with preprocessed output, and I had trouble reducing the example.
Douglas Gregor [Fri, 15 Mar 2013 22:15:07 +0000 (22:15 +0000)]
<rdar://problem/13426257> Introduce SDKSettings.plist as an input file dependency for PCH/modules.
When we're building a precompiled header or module against an SDK on
Darwin, there will be a file SDKSettings.plist in the sysroot. Since
stat()'ing every system header on which a module or PCH file depends
is performance suicide, we instead stat() just SDKSettings.plist. This
hack works well on Darwin; it's unclear how we want to handle this on
other platforms. If there is a canonical file, we should use it; if
not, we either have to take the performance hit of stat()'ing system
headers repeatedly or roll the dice by not checking anything.
Jordan Rose [Fri, 15 Mar 2013 21:41:55 +0000 (21:41 +0000)]
[analyzer] Make GRBugReporter::generatePathDiagnostic iterative, not recursive.
The previous generatePathDiagnostic() was intended to be tail-recursive,
restarting and trying again if a report was marked invalid. However:
(1) this leaked all the cloned visitors, which weren't being deleted, and
(2) this wasn't actually tail-recursive because some local variables had
non-trivial destructors.
This was causing us to overflow the stack on inputs with large numbers of
reports in the same equivalence class, such as sqlite3.c. Being iterative
at least prevents us from blowing out the stack, but doesn't solve the
performance issue: suppressing thousands (yes, thousands) of paths in the
same equivalence class is expensive. I'm looking into that now.