Douglas Gregor [Mon, 21 Jan 2013 20:07:12 +0000 (20:07 +0000)]
Give ModuleFiles an index, so that we can use indexed vectors rather
than DenseMaps and SmallPtrSets for module-visitation data. ~2.6%
speedup for modules.
Douglas Gregor [Mon, 21 Jan 2013 19:42:21 +0000 (19:42 +0000)]
Eliminate Sema::CompareProperties(), which was walking over a pile of
lexical declarations looking for properties when we could more
efficiently check for property mismatches at property declaration
time. Good for ~1% of -fsyntax-only time when most of the properties
we're checking against come from an AST file.
Manuel Klimek [Mon, 21 Jan 2013 19:17:52 +0000 (19:17 +0000)]
Fixes various problems around enum parsing.
Very similar to what we do for record definitions:
- tighten down what is an enum definition, so that we don't mistake a
function for an enum
- allow common idioms around declarations (we'll want to handle that
more centrally in the future)
We now correctly format:
enum X f() {
a();
return 42;
}
Douglas Gregor [Mon, 21 Jan 2013 19:05:22 +0000 (19:05 +0000)]
Eliminate the oddly-named Sema::ComparePropertiesInBaseAndSuper, which
did a redundant traversal of the lexical declarations in the
superclass. Instead, when we declare a new property, look into the
superclass to see whether we're redeclaring the property. Goot for 1%
of -fsyntax-only time on Cocoa.h and a little less than 3% on my
modules test case.
[ASTUnit] Unlike LoadFromCommandLine, LoadFromCompilerInvocation causes a crash if Precompilepreamble
is set to true because there is no FileManager at that point.
Douglas Gregor [Mon, 21 Jan 2013 18:35:55 +0000 (18:35 +0000)]
Replace some unnecessary O(N^2) lookups for properties with
DeclContext lookups. The performance win is negligible in my tests,
but it's the Right Thing To Do (TM).
Chris Lattner [Mon, 21 Jan 2013 18:28:26 +0000 (18:28 +0000)]
ReadSourceManagerBlock is skipping over records that can contain Blobs. Not passing
in a StringRef to bind to them forces them to be unpacked into the Record as individual
bytes. This is wasteful, but not likely to be measurable in this instance.
Douglas Gregor [Mon, 21 Jan 2013 16:47:11 +0000 (16:47 +0000)]
When loading an identifier from an AST file solely for the purpose of
forming the identifier, e.g., as part of a selector or a declaration
name, don't actually deserialize any information about the
identifier. Instead, simply mark it "out-of-date" and we'll load the
the information on demand. 2% speedup on the modules testcase I'm
looking at; should also help PCH.
Manuel Klimek [Mon, 21 Jan 2013 16:42:44 +0000 (16:42 +0000)]
Fixes formatting of empty blocks.
We now only put empty blocks into a single line, if all of:
- all tokens of the structural element fit into a single line
- we're not in a control flow statement
Note that we usually don't put record definitions into a single line, as
there's usually at least one more token (the semicolon) after the
closing brace. This doesn't hold when we are in a context where there is
no semicolon, like "enum E {}".
There were some missing tests around joining lines around the corner
cases of the allowed number of columns, so this patch adds some.
Douglas Gregor [Mon, 21 Jan 2013 16:16:40 +0000 (16:16 +0000)]
When deserializing a declaration, don't look for redeclarations if its
kind indicates that it can never be redeclared. Good for a 1% speedup,
and redeclaration searching drops off the profile.
Douglas Gregor [Mon, 21 Jan 2013 15:25:38 +0000 (15:25 +0000)]
Introduce a fast path for the ASTReader's name lookup within a
DeclContext. When the DeclContext is of a kind that can only be
defined once and never updated, we limit the search to the module file
that conatins the lookup table. Provides a 15% speedup in one
modules-heavy source file.
Manuel Klimek [Mon, 21 Jan 2013 10:07:49 +0000 (10:07 +0000)]
Fix parsing of return statements.
Previously, we would not detect brace initializer lists in return
statements, thus:
return (a)(b) { 1, 2, 3 };
would put the semicolon onto the next line.
David Blaikie [Mon, 21 Jan 2013 04:37:12 +0000 (04:37 +0000)]
PR14472: Preserve qualifiers while unwrapping types for debug info
Looks like r161368 fixed this for one case but not all. This change generalizes
the solution over all the unwrapping cases. Now that preserving the qualifiers
is done independent of the particular type being unwrapped I won't bother
adding test cases for each one but at least demonstrate that this change was
necessary & sufficient to fix the bug.
Alexey Samsonov [Sun, 20 Jan 2013 13:12:12 +0000 (13:12 +0000)]
Add top-level Clang flag -f(no-)sanitize-address-zero-base-shadow that makes AddressSanitizer use bottom of the address space for the shadow memory. On Linux it can be used with -fPIE/-pie to improve performance.
Sean Silva [Sun, 20 Jan 2013 01:58:26 +0000 (01:58 +0000)]
Fix what appears to be a copy-paste error.
This looks like it was copied from SetUpBuildDumpLog, which dumps to the
file `DiagOpts->DumpBuildInformation`. There is another member
`DiagOpts->DiagnosticLogFile` which appears to be unused. The fact that
this feature doesn't even print to the output file specified on the
command line makes me think that it should be ripped out.
Chris Lattner [Sat, 19 Jan 2013 21:39:22 +0000 (21:39 +0000)]
switch a bunch of ASTReader to use the new BitstreamCursor::advance* methods,
which hide a bunch of private details of the cursor from clients and simplify
their code. More to come.
Jordan Rose [Sat, 19 Jan 2013 19:52:57 +0000 (19:52 +0000)]
[analyzer] Don't show "Entered 'foo'" if 'foo' is implicit.
Before:
Calling implicit default constructor for 'Foo' (where Foo is constructed)
Entered call from 'test' (at "=default" or 'Foo' declaration)
Calling default constructor for 'Bar' (at "=default" or 'Foo' declaration)
After:
Calling implicit default constructor for 'Foo' (where Foo is constructed)
Calling default constructor for 'Bar' (at "=default" or 'Foo' declaration)
This only affects the plist diagnostics; this note is never shown in the
other diagnostics.
Tobias Grosser [Sat, 19 Jan 2013 11:03:44 +0000 (11:03 +0000)]
[cindex.py] Add cache for CompletionChunk spellings
Most of the CompletionChunks represent braces, colons or other one
character spellings. There is no need to call libclang, to figure out
how to write a colon. Instead we use an internal cache to retrieve the
correct spelling. As function calls from python are very expensive and
this is a performance critical part of auto completion this patch makes
formatting of auto completion results a lot faster.
Tobias Grosser [Sat, 19 Jan 2013 11:03:42 +0000 (11:03 +0000)]
[cindex.py] Replace CachedProperty with our own implementation
This is a very performance critical point for auto completion. The manual
implementation gives a large speedup. As it does not complicate the code a lot,
I figured it is worth the change. If anybody understands why the CachedProperty
is here so much slower, I am very interested in working on an improvement of
CachedProperty.
Formatting time changes from 0.72 to 0.57 seconds.
Chandler Carruth [Sat, 19 Jan 2013 08:09:44 +0000 (08:09 +0000)]
Re-sort all the headers. Lots of regressions have crept in here.
Manually fix the order of UnwrappedLineParser.cpp as that one didn't
have its associated header as the first header.
This also uncovered a subtle inclusion order dependency as CLog.h didn't
include LLVM.h to pick up using declarations it relied upon.
Chandler Carruth [Sat, 19 Jan 2013 06:31:24 +0000 (06:31 +0000)]
Move an input header file under an Inputs directory to be consistent
with other auxilliary test inputs and simplify the identification of
inputs to tests.
Anna Zaks [Sat, 19 Jan 2013 02:18:15 +0000 (02:18 +0000)]
[analyzer] Suppress warnings coming out of macros defined in sys/queue.h
Suppress the warning by just not emitting the report. The sink node
would get generated, which is fine since we did reach a bad state.
Motivation
Due to the way code is structured in some of these macros, we do not
reason correctly about it and report false positives. Specifically, the
following loop reports a use-after-free. Because of the way the code is
structured inside of the macro, the analyzer assumes that the list can
have cycles, so you end up with use-after-free in the loop, that is
safely deleting elements of the list. (The user does not have a way to
teach the analyzer about shape of data structures.)
SLIST_FOREACH_SAFE(item, &ctx->example_list, example_le, tmpitem) {
if (item->index == 3) { // if you remove each time, no complaints
assert((&ctx->example_list)->slh_first == item);
SLIST_REMOVE(&ctx->example_list, item, example_s, example_le);
free(item);
}
}
Douglas Gregor [Fri, 18 Jan 2013 22:27:09 +0000 (22:27 +0000)]
Once we've collected the template arguments for a
partially-substituted parameter pack in a template, forget about the
partially-substituted parameter pack: it is now completed. Fixes
<rdar://problem/12176336>.
Manuel Klimek [Fri, 18 Jan 2013 14:04:34 +0000 (14:04 +0000)]
Fixes problems with line merging in the face of preprocessor directives.
This patch prepares being able to test for and fix more problems (see
FIXME in the test for example).
Previously we would output unwrapped lines for preprocessor directives
at the point where we also parsed the hash token. Since often
projections only terminate (and thus output their own unwrapped line)
after peeking at the next token, this would lead to the formatter seeing
the preprocessor directives out-of-order (slightly earlier). To be able
to correctly identify lines to merge, the formatter needs a well-defined
order of unwrapped lines, which this patch introduces.
projects / clang / log