AST/PCH files more lazy:
- Don't preload all of the file source-location entries when reading
the AST file. Instead, load them lazily, when needed.
- Only look up header-search information (whether a header was already
#import'd, how many times it's been included, etc.) when it's needed
by the preprocessor, rather than pre-populating it.
Previously, we would pre-load all of the file source-location entries,
which also populated the header-search information structure. This was
a relatively minor performance issue, since we would end up stat()'ing
all of the headers stored within a AST/PCH file when the AST/PCH file
was loaded. In the normal PCH use case, the stat()s were cached, so
the cost--of preloading ~860 source-location entries in the Cocoa.h
case---was relatively low.
However, the recent optimization that replaced stat+open with
open+fstat turned this into a major problem, since the preloading of
source-location entries would now end up opening those files. Worse,
those files wouldn't be closed until the file manager was destroyed,
so just opening a Cocoa.h PCH file would hold on to ~860 file
descriptors, and it was easy to blow through the process's limit on
the number of open file descriptors.
By eliminating the preloading of these files, we neither open nor stat
the headers stored in the PCH/AST file until they're actually needed
for something. Concretely, we went from
*** HeaderSearch Stats:
835 files tracked.
364 #import/#pragma once files.
823 included exactly once.
6 max times a file is included.
3 #include/#include_next/#import.
0 #includes skipped due to the multi-include optimization.
1 framework lookups.
0 subframework lookups.
*** Source Manager Stats:
835 files mapped, 3 mem buffers mapped.
37460 SLocEntry's allocated, 11215575B of Sloc address space used.
62 bytes of files mapped, 0 files with line #'s computed.
with a trivial program that uses a chained PCH including a Cocoa PCH
to
*** HeaderSearch Stats:
4 files tracked.
1 #import/#pragma once files.
3 included exactly once.
2 max times a file is included.
3 #include/#include_next/#import.
0 #includes skipped due to the multi-include optimization.
1 framework lookups.
0 subframework lookups.
*** Source Manager Stats:
3 files mapped, 3 mem buffers mapped.
37460 SLocEntry's allocated, 11215575B of Sloc address space used.
62 bytes of files mapped, 0 files with line #'s computed.
for the same program.
llvm-svn: 125286
linkage into Decl.cpp. Disable this logic for extern "C" functions, because
the operative rule there is weaker. Fixes rdar://problem/8898466
llvm-svn: 125268
there were only three virtual methods of any significance.
The primary way to grab child iterators now is with
Stmt::child_range children();
Stmt::const_child_range children() const;
where a child_range is just a std::pair of iterators suitable for
being llvm::tie'd to some locals. I've left the old child_begin()
and child_end() accessors in place, but it's probably a substantial
penalty to grab the iterators individually now, since the
switch-based dispatch is kindof inherently slower than vtable
dispatch. Grabbing them together is probably a slight win over the
status quo, although of course we could've achieved that with vtables, too.
I also reclassified SwitchCase (correctly) as an abstract Stmt
class, which (as the first such class that wasn't an Expr subclass)
required some fiddling in a few places.
There are somewhat gross metaprogramming hooks in place to ensure
that new statements/expressions continue to implement
getSourceRange() and children(). I had to work around a recent clang
bug; dgregor actually fixed it already, but I didn't want to
introduce a selfhosting dependency on ToT.
llvm-svn: 125183
and we later find the definition, make sure that we add the definition
(not the declaration) to the list of deferred definitions to
emit. Fixes PR8864.
Thanks to Nick Lewycky for testing this patch out
llvm-svn: 125157
The optimization involves eagerly pruning ExplodedNodes from the ExplodedGraph that contain
practically no difference between the predecessor and successor nodes. For example, if
the state is different between a predecessor and a node, the node is left in. Only for
the 'environment' component of the state do we not care if the ExplodedNodes are different.
This paves the way for future optimizations where we can reclaim the environment objects.
llvm-svn: 125154
Eventually there will also be a lib/StaticAnalyzer/Frontend that will handle initialization and checker registration.
Yet another library to avoid cyclic dependencies between Core and Checkers.
llvm-svn: 125124
record away from the core processor record. The tangling of these two
data structures led to some inefficiencies (e.g., deserializing all
of the detailed preprocessing record when we didn't need it, such as
while performing code completion) along with some unnecessary
ugliness.
llvm-svn: 125117
instead from the Scope; Inner scopes in bodies don't have DeclContexts associated with them.
Fixes http://llvm.org/PR9160 & rdar://problem/8966163.
llvm-svn: 125097
a glvalue as a temporary. Previously, we were enumerating all of the
cases that coul return glvalues and might be called with
Sema::MaybeBindToTemporary(), but that was gross and we missed the
Objective-C property reference case.
llvm-svn: 125070
- BlockDeclRefExprs always store VarDecls
- BDREs no longer store copy expressions
- BlockDecls now store a list of captured variables, information about
how they're captured, and a copy expression if necessary
With that in hand, change IR generation to use the captures data in
blocks instead of walking the block independently.
Additionally, optimize block layout by emitting fields in descending
alignment order, with a heuristic for filling in words when alignment
of the end of the block header is insufficient for the most aligned
field.
llvm-svn: 125005
overridden via remapping. Thus, when we create a "virtual" file in the
file manager, we still stat() the real file that lives behind it so
that we can provide proper uniquing based on inodes. This helps keep
the file manager much more consistent.
To take advantage of this when reparsing files in libclang, we disable
the use of the stat() cache when reparsing or performing code
completion, since the stat() cache is very likely to be out of date in
this use case.
llvm-svn: 124971
A common pattern in classes with multiple initializers is to put the
subclass's common initialization bits into a static function that receives
the value of 'self', e.g:
if (!(self = [super init]))
return nil;
if (!(self = _commonInit(self)))
return nil;
It was reported that 'self' was not set to the result of [super init].
Until we can use inter-procedural analysis, in such a call, transfer the
ObjCSelfInitChecker flags associated with 'self' to the result of the call.
Fixes rdar://8937441 & http://llvm.org/PR9094
llvm-svn: 124940
say "out-of-line definition differ from the declaration in the return type" instead of
the silly "functions that differ only in their return type cannot be overloaded".
Addresses rdar://7980179.
llvm-svn: 124939
when selector metadata is generated, which is triggered
by at least on class implementation. This is to match gcc's
behavior. // rdar://8851684.
llvm-svn: 124909
id-expression, e.g.,
CurrentClass<T>::member
Previously, if CurrentClass<T> was dependent and not complete, we
would treat it as a dependent-scoped declaration reference expression,
even if CurrentClass<T> referred to the current instantiation.
Fixes PR8966 and improves type checking of templates.
llvm-svn: 124867
it's okay for the following template parameters to not have default
arguments (since those template parameters can still be
deduced). Also, downgrade the error about default template arguments
in function templates to an extension warning, since this is a
harmless C++0x extension.
llvm-svn: 124855
The difference with gcc is that it warns if you overload virtual methods only if
the method doesn't also override any method. This is to cut down on the number of warnings
and make it more useful like reported here: http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20423.
If we want to warn that not all overloads are overriden we can have an additional
warning like -Wpartial-override.
-Woverloaded-virtual, unlike gcc, is added to -Wmost. Addresses rdar://8757630.
llvm-svn: 124805
whose inode has changed since the file was first created and that is
being seen through a different path name (e.g., due to symlinks or
relative path elements), such that its FileEntry pointer doesn't match
a known FileEntry pointer. Since this requires a system call (to
stat()), we only perform this deeper checking if we can't find the
file by comparing FileEntry pointers.
Also, add a micro-optimization where we don't bother to compute line
numbers when given the location (1, 1). This improves the
efficiency of clang_getLocationForOffset().
llvm-svn: 124800
right for anonymous struct/union members led to me discovering some
seemingly broken code in that area of Sema, which I fixed, partly by
changing the representation of member pointer constants so that
IndirectFieldDecls aren't expanded. This led to assorted cleanups with
member pointers in CodeGen, and while I was doing that I saw some random
other things to clean up.
llvm-svn: 124785
the atomic writes option, since the intent is that this option be set for an
entire build, which may have any number of compiler instances writing to the
same output file.
llvm-svn: 124772
is not defined in the current translation unit. Doing so lead to compile errors
such as PR9114.
Instead, when CodeGen is building the vtable, don't try to emit a definition
for functions that aren't marked used in the current translation unit.
Fixes PR9114.
llvm-svn: 124768
on that name. Canonicalization eliminates silliness such as "." and
"foo/.." that breaks the uniquing of files in the presence of virtual
files or files whose inode numbers have changed during
parsing/re-parsing. c-index-test isn't able to create this crazy
situation, so I've resorted to testing outside of the Clang
tree. Fixes <rdar://problem/8928220>.
Note that this hackery will go away once we have a real virtual file
system on which we can layer FileManager; the virtual-files hack is
showing cracks.
llvm-svn: 124754
savings of 25% sounds impressive, except that this amounted to only
about 360k in our standard "large" completion result set (40,000
results). Since code completion is performance-sensitive, the 4%
slowdown due to uniquing outweighs the 360k benefit.
llvm-svn: 124737
callbacks class.
- Aside from being generally cleaner, this also allows -H to work correctly in
modes other than standard preprocessing (e.g., -c, -MM, etc.)
llvm-svn: 124723
Macros frequently contain extra '()' to make instantiation less error prone.
This warning was flagging a ton of times on postgresql because of its use of macros.
llvm-svn: 124695
BumpPtrAllocator, rather than manually new/delete'ing them. This
optimization also allows us to avoid allocating memory for and copying
constant strings (e.g., "return", "class").
This also required embedding the priority and availability of results
within the code completion string, to avoid extra memory allocation
within libclang.
llvm-svn: 124673
Implicit dtors confuse the ad hoc path-sensitivity of UninitializedValuesV2.cpp. This isn't
the ideal solution, as it will directly impact compile time, but should significantly reduce
the noise of -Wuninitialized on some code bases.
This immediately "fixes" the false positive reported in PR 9063, although this
isn't the right fix in the long run.
llvm-svn: 124667
extremely rambunctious, both on parsing and on template instantiation.
Calm it down, fixing an internal consistency assert on anonymous enum
instantiation manglings.
llvm-svn: 124653
compatible, not having the same type.
Fix rdar://8183908 in which compatible vector types weren't initialized properly leading to a crash.
llvm-svn: 124637
* llvm-link would complains about mismatched visibility
* If we produce a relocation with an available_externally, it is good to know that
it is hidden.
llvm-svn: 124633
The rationale is that it is highly likely that the user's getter/setter isn't atomically implemented. Off by default.
Addresses rdar://8782645.
-Wcustom-atomic-properties and -Wimplicit-atomic-properties are under the -Watomic-properties group.
llvm-svn: 124609
conversions (<rdar://problem/8592139>) for overload resolution. The
conversion ranking mirrors C++'s conversion ranking fairly closely,
except that we use a same pseudo-subtyping relationship employed by
Objective-C pointer assignment rather than simple checking
derived-to-base conversions. This change covers:
- Conversions to pointers to a specific object type are better than
conversions to 'id', 'Class', qualified 'id', or qualified 'Class'
(note: GCC doesn't perform this ranking, but it matches C++'s rules
for ranking conversions to void*).
- Conversions to qualified 'id' or qualified 'Class' are better than
conversions to 'id' or 'Class', respectively.
- When two conversion sequences convert to the same type, rank the
conversions based on the relationship between the types we're
converting from.
- When two conversion sequences convert from the same non-id,
non-Class type, rank the conversions based on the relationship of
the types we're converting to. (note: GCC allows this ranking even
when converting from 'id', which is extremeley dangerous).
llvm-svn: 124591
allow ref-qualifiers on function types used as template type
arguments. GNU actually allows cv-qualifiers on function types in many
places where it shouldn't, so we currently categorize this as a GNU
extension.
llvm-svn: 124584
current translation unit as available_externally.
This helps devirtualize the second example in PR3100, comment 18:
struct S { S() {}; virtual void xyzzy(); };
inline void foo(S *s) { s->xyzzy(); }
void bar() { S s; foo(&s); }
This involved four major changes:
1. In DefineUsedVTables, always mark virtual member functions as referenced for
non-template classes and class template specializations.
2. In CodeGenVTables::ShouldEmitVTableInThisTU return true if optimizations are
enabled, even if the key function is not implemented in this translation
unit. We don't ever do this for code compiled with -fapple-kext, because we
don't ever want to devirtualize virtual member function calls in that case.
3. Give the correct linkage for vtables where the key function is not defined.
4. Update the linkage for RTTI structures when necessary.
llvm-svn: 124565
fixing a crash which probably nobody was ever going to see. In doing so,
fix a horrendous number of problems with the conditional-cleanups code.
Also, make conditional cleanups re-use the cleanup's activation variable,
which avoids some unfortunate repetitiveness.
llvm-svn: 124481
might be queried in places where we absolutely require a valid
location (e.g., for template instantiation). Fixes some major
brokenness in the use of __is_convertible_to.
llvm-svn: 124465
non-class prvalues actually require the realization of a
temporary. For everything else, we already have an lvalue (or class
prvalue) in the subexpression.
Note: we're missing some move elision in this case. I'll tackle that next.
llvm-svn: 124453
access control errors into SFINAE errors, so that the trait provides
enough support to implement the C++0x std::is_convertible type trait.
To get there, the SFINAETrap now knows how to set up a SFINAE context
independent of any template instantiations or template argument
deduction steps, and (separately) can set a Sema flag to translate
access control errors into SFINAE errors. The latter can also be
useful if we decide that access control errors during template argument
deduction should cause substitution failure (rather than a hard error)
as has been proposed for C++0x.
llvm-svn: 124446
semantics after the C++0x is_convertible type trait. This
implementation is not 100% complete, because it allows access errors
to be hard errors (rather than just evaluating false).
Original patch by Steven Watanabe!
llvm-svn: 124425
FileManager.cpp: Allow virtual files in nonexistent directories.
FileManager.cpp: Close FileDescriptor for virtual files that correspond to actual files.
FileManager.cpp: Enable virtual files to be created even for files that were flagged as NON_EXISTENT_FILE, e.g. by a prior (unsuccessful) addFile().
ASTReader.cpp: Read a PCH even if the original source files cannot be found.
Add a test for reading a PCH of a file that has been removed and diagnostics referencing that file.
llvm-svn: 124374
Douglas Gregor