language options. Use that .def file to declare the LangOptions class
and initialize all of its members, eliminating a source of annoying
initialization bugs.
AST serialization changes are next up.
llvm-svn: 139605
declaration was deserialized from an AST file. Use this instead of
Decl::getPCHLevel() wherever possible. This is a simple step toward
killing off Decl::getPCHLevel().
llvm-svn: 139427
'id' that can be used (only!) via a contextual keyword as the result
type of an Objective-C message send. 'instancetype' then gives the
method a related result type, which we have already been inferring for
a variety of methods (new, alloc, init, self, retain). Addresses
<rdar://problem/9267640>.
llvm-svn: 139275
builtin types (When requested). This is another step toward making
ASTUnit build the ASTContext as needed when loading an AST file,
rather than doing so after the fact. No actual functionality change (yet).
llvm-svn: 138985
include guards don't show up as macro definitions in every translation
unit that imports a module. Macro definitions can, however, be
exported with the intentionally-ugly #__export_macro__
directive. Implement this feature by not even bothering to serialize
non-exported macros to a module, because clients of that module need
not (should not) know that these macros even exist.
llvm-svn: 138943
The initial incentive was to fix a crash when PCH chaining categories
to an interface, but the fix was done in the "modules way" that I hear
is popular with the kids these days.
Each module stores the local chain of categories and we combine them
when the interface is loaded. We also warn if non-dependent modules
introduce duplicate named categories.
llvm-svn: 138926
sure that all of the CXXConversionDecls go into the same
bucket. Otherwise, name lookup might not find them all. Fixes
<rdar://problem/10041960>.
llvm-svn: 138824
Empty lookups can occur in the DeclContext map when we are chaining PCHs, where
the empty lookup indicates that we already looked in ExternalASTSource.
llvm-svn: 138816
table when serializing an AST file. This was a holdover from the days
before chained PCH, and is a complete waste of time and storage
now. It's a good thing it's useless, because I have no idea how I
would have implemented MaterializeVisibleDecls efficiently in the
presence of modules.
llvm-svn: 138496
Currently getMacroArgExpandedLocation is very inefficient and for the case
of a location pointing at the main file it will end up checking almost all of
the SLocEntries. Make it faster:
-Use a map of macro argument chunks to their expanded source location. The map
is for a single source file, it's stored in the file's ContentCache and lazily
computed, like the source lines cache.
-In SLocEntry's FileInfo add an 'unsigned NumCreatedFIDs' field that keeps track
of the number of FileIDs (files and macros) that were created during preprocessing
of that particular file SLocEntry. This is useful when computing the macro argument
map in skipping included files while scanning for macro arg FileIDs that lexed from
a specific source file. Due to padding, the new field does not increase the size
of SLocEntry.
llvm-svn: 138225
-import-module) vs. loaded because some other module depends on
them. As part of doing this, pass down the module that caused a module
to be loaded directly, rather than assuming that we're loading a
chain. Finally, write out all of the directly-loaded modules when
serializing an AST file (using the new IMPORTS record), so that an AST
file can depend on more than one other AST file, all of which will be
loaded when that AST file is loaded. This allows us to form and load a
tree of modules, but we can't yet load a DAG of modules.
llvm-svn: 137923
all AST files have a normal METADATA record that has the same form
regardless of whether we refer to a chained PCH or any other kind of
AST file.
Introduce the IMPORTS record, which describes all of the AST files
that are imported by this AST file, and how (as a module, a PCH file,
etc.). Currently, we emit at most one entry to this record, to support
chained PCH.
llvm-svn: 137869
type over into the AST context, then make that declaration a
predefined declaration in the AST format. This ensures that different
AST files will at least agree on the (global) declaration ID for 'id',
and eliminates one of the "special" types in the AST file format.
llvm-svn: 137429
eliminating a pile of redundant code (and probably some bugs in the
process). The variation between chained and non-chained PCH is fairly
small now anyway.
llvm-svn: 137410
declaration that never actually gets serialized. Instead, serialize
the various kinds of update records (lexical decls, visible decls, the
addition of an anonymous namespace) for the translation unit, even if
we're not chaining. This way, we won't have to deal with multiple
loaded translation unit declarations.
llvm-svn: 137395
enumerations from the ASTContext into CodeGen, so that we don't need
to serialize it to AST files. This appears to be the last of the
low-hanging fruit for SpecialTypes.
llvm-svn: 137124
layout of a constant NSString from the ASTContext over to CodeGen,
since this is solely CodeGen's responsibility. Eliminates one of the
unnecessary "special" types that we serialize.
llvm-svn: 137121
the last of the ID/offset/index mappings that I know
of. Unfortunately, the "gap" method of testing doesn't work here due
to the way the preprocessing record performs iteration. We'll do more
testing once multi-AST loading is possible.
llvm-svn: 136902
IDs will never cross module boundaries, since they're tied to the
CXXDefinitionData, so just use a local mapping throughout. Eliminate
the global -> local tables and supporting data.
llvm-svn: 136847
AST file, along with an enumeration naming those predefined
declarations. No functionality change, but this will make it easier to
introduce new predefined declarations, when/if we need them.
llvm-svn: 136781
reader, to allow AST files to be loaded with their declarations
remapped to different ID numbers. Fix a number of places where we were
either failing to map local declaration IDs into global declaration
IDs or where interpreting the local declaration IDs within the wrong
module.
I've tested this via the usual "random gaps" method. It works well
except for the preamble tests, because our handling of the precompiled
preamble requires declaration and preprocessed entity to be stable
when parsing code and then loading that back into memory. This
property will hold in general, but my randomized testing naturally
breaks this property to get more coverage. In the future, I expect
that the precompiled preamble logic won't need this property.
I am very unhappy with the current handling of the translation unit,
which is a rather egregious hack. We're going to have to do something
very different here for loading multiple AST files, because we don't
want to have to cope with merging two translation units. Likely, we'll
just handle translation units entirely via "update" records, and
predefine a single, fixed declaration ID for the translation
unit. That will come later.
llvm-svn: 136779
by eliminating the type ID from constructor, destructor, and
conversion function names. There are several reasons for this change:
- A given type (say, int*) isn't guaranteed to have a single, unique
type ID within a chain of PCH files. Hence, we could end up hashing
based on the wrong type ID, causing name lookup to fail.
- The mapping from types back to type IDs required one DenseMap
entry for every type that was ever deserialized, which was an
unacceptable cost to support just the name lookup of constructors,
destructors, and conversion functions. Plus, this mapping could
never actually work with chained or multiple PCH, based on the first
bullet.
Once we have eliminated the type from the hash function, these
problems go away, as does my horrible "reverse type remap" hack, which
was doomed from the start (see bullet #1 above) and far too
complicated.
However, note that removing the type from the hash function means that
all constructors, destructors, and conversion functions have the same
hash key, so I've updated the caller to double-check that the
declarations found have the appropriate name.
llvm-svn: 136708
reader. This scheme permits an AST file to be loaded with its type IDs
shifted anywhere in the type ID space.
At present, the type indices are still allocated in the same boring
way they always have been, just by adding up the number of types in
each PCH file within the chain. However, I've done testing with this
patch by randomly sliding the base indices at load time, to ensure
that remapping is occurring as expected. I may eventually formalize
this in some testing flag, but loading multiple (non-chained) AST
files at once will eventually exercise the same code.
There is one known problem with this patch, which involves name lookup
of operator names (e.g., "x.operator int*()") in cases where multiple
PCH files in the chain. The hash function itself depends on having a
stable type ID, which doesn't happen with chained PCH and *certainly*
doesn't happen when sliding type IDs around. We'll need another
approach. I'll tackle that next.
llvm-svn: 136693
completely broken deserialization mapping code we had for VTableUses,
which would have broken horribly as soon as our local-to-global ID
mapping became interesting.
llvm-svn: 136371
we could turn this into an on-disk hash table so we don't load the
whole thing the first time we need it. However, it tends to be very,
very small (i.e., empty) for most precompiled headers, so it isn't all
that interesting.
llvm-svn: 136352
- Added LazyVector::erase() to support this use case.
- Factored out the LazyDecl-of-Decls to RecordData translation in
the ASTWriter. There is still a pile of code duplication here to
eliminate.
llvm-svn: 136270
contents are lazily loaded on demand from an external source (e.g., an
ExternalASTSource or ExternalSemaSource). The "loaded" entities are
kept separate from the "local" entities, so that the two can grow
independently.
Switch Sema::TentativeDefinitions from a normal vector that is eagerly
populated by the ASTReader into one of these LazyVectors, making the
ASTReader a bit more like me (i.e., lazy).
llvm-svn: 136262
etc. With this I think essentially all of the SourceManager APIs are
converted. Comments and random other bits of cleanup should be all thats
left.
llvm-svn: 136057
and various other 'expansion' based terms. I've tried to reformat where
appropriate and catch as many references in comments but I'm going to do
several more passes. Also I've tried to expand parameter names to be
more clear where appropriate.
llvm-svn: 136056
so that we have one, simple way to map from global bit offsets to
local bit offsets. Eliminates a number of loops over the chain, and
generalizes for more interesting bit remappings.
Also, as an amusing oddity, we were computing global bit offsets
*backwards* for preprocessed entities (e.g., the directly included PCH
file in the chain would start at offset zero, rather than the original
PCH that occurs first in translation unit). Even more amusingly, it
made precompiled preambles work, because we were forgetting to adjust
the local bit offset to a global bit offset when storing preprocessed
entity offsets in the ASTUnit. Two wrongs made a right, and now
they're both right.
llvm-svn: 135750
entities generated directly by the preprocessor from those loaded from
the external source (e.g., the ASTReader). By separating these two
sets of entities into different vectors, we allow both to grow
independently, and eliminate the need for preallocating all of the
loaded preprocessing entities. This is similar to the way the recent
SourceManager refactoring treats FileIDs and the source location
address space.
As part of this, switch over to building a continuous range map to
track preprocessing entities.
llvm-svn: 135646
source locations from source locations loaded from an AST/PCH file.
Previously, loading an AST/PCH file involved carefully pre-allocating
space at the beginning of the source manager for the source locations
and FileIDs that correspond to the prefix, and then appending the
source locations/FileIDs used for parsing the remaining translation
unit. This design forced us into loading PCH files early, as a prefix,
whic has become a rather significant limitation.
This patch splits the SourceManager space into two parts: for source
location "addresses", the lower values (growing upward) are used to
describe parsed code, while upper values (growing downward) are used
for source locations loaded from AST/PCH files. Similarly, positive
FileIDs are used to describe parsed code while negative FileIDs are
used to file/macro locations loaded from AST/PCH files. As a result,
we can load PCH/AST files even during parsing, making various
improvemnts in the future possible, e.g., teaching #include <foo.h> to
look for and load <foo.h.gch> if it happens to be already available.
This patch was originally written by Sebastian Redl, then brought
forward to the modern age by Jonathan Turner, and finally
polished/finished by me to be committed.
llvm-svn: 135484
variants to 'expand'. This changed a couple of public APIs, including
one public type "MacroInstantiation" which is now "MacroExpansion". The
rest of the codebase was updated to reflect this, especially the
libclang code. Two of the C++ (and thus easily changed) libclang APIs
were updated as well because they pertained directly to the old
MacroInstantiation class.
No functionality changed.
llvm-svn: 135139
for a template template parameter.
Uses to follow.
I've also made the uniquing of SubstTemplateTemplateParmPacks
use a ContextualFoldingSet as a minor space efficiency.
llvm-svn: 134137
vector<int>
to
std::vector<int>
Patch by Kaelyn Uhrain, with minor tweaks + PCH support from me. Fixes
PR5776/<rdar://problem/8652971>.
Thanks Kaelyn!
llvm-svn: 134007
Language-design credit goes to a lot of people, but I particularly want
to single out Blaine Garst and Patrick Beard for their contributions.
Compiler implementation credit goes to Argyrios, Doug, Fariborz, and myself,
in no particular order.
llvm-svn: 133103
Related result types apply Cocoa conventions to the type of message
sends and property accesses to Objective-C methods that are known to
always return objects whose type is the same as the type of the
receiving class (or a subclass thereof), such as +alloc and
-init. This tightens up static type safety for Objective-C, so that we
now diagnose mistakes like this:
t.m:4:10: warning: incompatible pointer types initializing 'NSSet *'
with an
expression of type 'NSArray *' [-Wincompatible-pointer-types]
NSSet *array = [[NSArray alloc] init];
^ ~~~~~~~~~~~~~~~~~~~~~~
/System/Library/Frameworks/Foundation.framework/Headers/NSObject.h:72:1:
note:
instance method 'init' is assumed to return an instance of its
receiver
type ('NSArray *')
- (id)init;
^
It also means that we get decent type inference when writing code in
Objective-C++0x:
auto array = [[NSMutableArray alloc] initWithObjects:@"one", @"two",nil];
// ^ now infers NSMutableArray* rather than id
llvm-svn: 132868
in ASTReader::validateFileEntries().
This avoids going through all source location entries and fixes the performance regression.
Many thanks to Doug for the hint!
(rdar://9530587)
llvm-svn: 132481
type that turns one type into another. This is used as the basis to
implement __underlying_type properly - with TypeSourceInfo and proper
behavior in the face of templates.
llvm-svn: 132017
hasTrivialDefaultConstructor() really really means it now.
Also implement a fun standards bug regarding aggregates. Doug, if you'd
like, I can un-implement that bug if you think it is truly a defect.
The bug is that non-special-member constructors are never considered
user-provided, so the following is an aggregate:
struct foo {
foo(int);
};
It's kind of bad, but the solution isn't obvious - should
struct foo {
foo (int) = delete;
};
be an aggregate or not?
Lastly, add a missing initialization to FunctionDecl.
llvm-svn: 131101
build a precompiled header. Use this information to eliminate the call
to SourceManager::getLocation() while loading a precompiled preamble,
since SourceManager::getLocation() itself causes unwanted
deserialization.
Fixed <rdar://problem/9399352>.
llvm-svn: 131021
Increase robustness of the delegating constructor cycle detection
mechanism. No more infinite loops on invalid or logic errors leading to
false results. Ensure that this is maintained correctly accross
serialization.
llvm-svn: 130887
which determines whether a particular file is actually a header that
is intended to be guarded from multiple inclusions within the same
translation unit.
llvm-svn: 130808
accompanying fixes to make it work today.
The core of this patch is to provide a link from a TemplateTypeParmType
back to the TemplateTypeParmDecl node which declared it. This in turn
provides much more precise information about the type, where it came
from, and how it functions for AST consumers.
To make the patch work almost a year after its first attempt, it needed
serialization support, and it now retains the old getName() interface.
Finally, it requires us to not attempt to instantiate the type in an
unsupported friend decl -- specifically those coming from template
friend decls but which refer to a specific type through a dependent
name.
A cleaner representation of the last item would be to build
FriendTemplateDecl nodes for these, storing their template parameters
etc, and to perform proper instantation of them like any other template
declaration. They can still be flagged as unsupported for the purpose of
access checking, etc.
This passed an asserts-enabled bootstrap for me, and the reduced test
case mentioned in the original review thread no longer causes issues,
likely fixed at somewhere amidst the 24k revisions that have elapsed.
llvm-svn: 130628
type trait. The previous implementation suffered from several problems:
1) It implemented all of the logic in RecordType by walking over every
base and field in a CXXRecordDecl and validating the constraints of
the standard. This made for very straightforward code, but is
extremely inefficient. It also is conceptually wrong, the logic tied
to the C++ definition of standard-layout classes should be in
CXXRecordDecl, not RecordType.
2) To address the performance problems with #1, a cache bit was added to
CXXRecordDecl, and at the completion of every C++ class, the
RecordType was queried to determine if it was a standard layout
class, and that state was cached. Two things went very very wrong
with this. First, the caching version of the query *was never
called*. Even within the recursive steps of the walk over all fields
and bases the caching variant was not called, making each query
a full *recursive* walk. Second, despite the cache not being used, it
was computed for every class declared, even when the trait was never
used in the program. This probably significantly regressed compile
time performance for edge-case files.
3) An ASTContext was required merely to query the type trait because
querying it performed the actual computations.
4) The caching bit wasn't managed correctly (uninitialized).
The new implementation follows the system for all the other traits on
C++ classes by encoding all the state needed in the definition data and
building up the trait incrementally as each base and member are added to
the definition of the class.
The idiosyncracies of the specification of standard-layout classes
requires more state than I would like; currently 5 bits. I could
eliminate one of the bits easily at the expense of both clarity and
resilience of the code. I might be able to eliminate one of the other
bits by computing its state in terms of other state bits in the
definition. I've already done that in one place where there was a fairly
simple way to achieve it.
It's possible some of the bits could be moved out of the definition data
and into some other structure which isn't serialized if the serialized
bloat is a problem. That would preclude serialization of a partial class
declaration, but that's likely already precluded.
Comments on any of these issues welcome.
llvm-svn: 130601
operators in C++ record declarations.
This patch starts off by updating a bunch of the standard citations to
refer to the draft 0x standard so that the semantics intended for move
varianst is clear. Where necessary these are duplicated so they'll be
available in doxygen.
It adds bit fields to keep track of the state for the move constructs,
and updates all the code necessary to track this state (I think) as
members are declared for a class. It also wires the state into the
various trait-like accessors in the AST's API, and tests that the type
trait expressions now behave correctly in the presence of move
constructors and move assignment operators.
This isn't complete yet due to these glaring FIXMEs:
1) No synthesis of implicit move constructors or assignment operators.
2) I don't think we correctly enforce the new logic for both copy and
move trivial checks: that the *selected* copy/move
constructor/operator is trivial. Currently this requires *all* of them
to be trivial.
3) Some of the trait logic needs to be folded into the fine-grained
trivial bits to more closely match the wording of the standard. For
example, many of the places we currently set a bit to track POD-ness
could be removed by querying other more fine grained traits on
demand.
llvm-svn: 130076
language options, and warn when reading an AST with a different value
for the bit.
There doesn't appear to be a good way to test this (commenting out
similar other language options doesn't break anything) but if folks have
suggestions on tests I'm happy to add them.
llvm-svn: 130071
which versions of an OS provide a certain facility. For example,
void foo()
__attribute__((availability(macosx,introduced=10.2,deprecated=10.4,obsoleted=10.6)));
says that the function "foo" was introduced in 10.2, deprecated in
10.4, and completely obsoleted in 10.6. This attribute ties in with
the deployment targets (e.g., -mmacosx-version-min=10.1 specifies that
we want to deploy back to Mac OS X 10.1). There are several concrete
behaviors that this attribute enables, as illustrated with the
function foo() above:
- If we choose a deployment target >= Mac OS X 10.4, uses of "foo"
will result in a deprecation warning, as if we had placed
attribute((deprecated)) on it (but with a better diagnostic)
- If we choose a deployment target >= Mac OS X 10.6, uses of "foo"
will result in an "unavailable" warning (in C)/error (in C++), as
if we had placed attribute((unavailable)) on it
- If we choose a deployment target prior to 10.2, foo() is
weak-imported (if it is a kind of entity that can be weak
imported), as if we had placed the weak_import attribute on it.
Naturally, there can be multiple availability attributes on a
declaration, for different platforms; only the current platform
matters when checking availability attributes.
The only platforms this attribute currently works for are "ios" and
"macosx", since we already have -mxxxx-version-min flags for them and we
have experience there with macro tricks translating down to the
deprecated/unavailable/weak_import attributes. The end goal is to open
this up to other platforms, and even extension to other "platforms"
that are really libraries (say, through a #pragma clang
define_system), but that hasn't yet been designed and we may want to
shake out more issues with this narrower problem first.
Addresses <rdar://problem/6690412>.
As a drive-by bug-fix, if an entity is both deprecated and
unavailable, we only emit the "unavailable" diagnostic.
llvm-svn: 128127
Change the interface to expose the new information and deal with the enormous fallout.
Introduce the new ExceptionSpecificationType value EST_DynamicNone to more easily deal with empty throw specifications.
Update the tests for noexcept and fix the various bugs uncovered, such as lack of tentative parsing support.
llvm-svn: 127537
Allow remapping a file by specifying another filename whose contents should be loaded if the original
file gets loaded. This allows to override files without having to create & load buffers in advance.
llvm-svn: 127052
template arguments. I believe that this is the last place in the AST
where we were storing a source range for a nested-name-specifier
rather than a proper nested-name-specifier location structure. (Yay!)
There is still a lot of cleanup to do in the TreeTransform, which
doesn't take advantage of nested-name-specifiers with source-location
information everywhere it could.
llvm-svn: 126844
template specialization types. There are still a few rough edges to
clean up with some of the parser actions dropping
nested-name-specifiers too early.
llvm-svn: 126776
nested-name-speciciers within elaborated type names, e.g.,
enum clang::NestedNameSpecifier::SpecifierKind
Fixes in this iteration include:
(1) Compute the type-source range properly for a dependent template
specialization type that starts with "template template-id ::", as
in a member access expression
dep->template f<T>::f()
This is a latent bug I triggered with this change (because now we're
checking the computed source ranges for dependent template
specialization types). But the real problem was...
(2) Make sure to set the qualifier range on a dependent template
specialization type appropriately. This will go away once we push
nested-name-specifier locations into dependent template
specialization types, but it was the source of the
valgrind errors on the buildbots.
llvm-svn: 126765
information for qualifier type names throughout the parser to address
several problems.
The commit message from r126737:
Push nested-name-specifier source location information into elaborated
name types, e.g., "enum clang::NestedNameSpecifier::SpecifierKind".
Aside from the normal changes, this also required some tweaks to the
parser. Essentially, when we're looking at a type name (via
getTypeName()) specifically for the purpose of creating an annotation
token, we pass down the flag that asks for full type-source location
information to be stored within the returned type. That way, we retain
source-location information involving nested-name-specifiers rather
than trying to reconstruct that information later, long after it's
been lost in the parser.
With this change, test/Index/recursive-cxx-member-calls.cpp is showing
much improved results again, since that code has lots of
nested-name-specifiers.
llvm-svn: 126748
name types, e.g., "enum clang::NestedNameSpecifier::SpecifierKind".
Aside from the normal changes, this also required some tweaks to the
parser. Essentially, when we're looking at a type name (via
getTypeName()) specifically for the purpose of creating an annotation
token, we pass down the flag that asks for full type-source location
information to be stored within the returned type. That way, we retain
source-location information involving nested-name-specifiers rather
than trying to reconstruct that information later, long after it's
been lost in the parser.
With this change, test/Index/recursive-cxx-member-calls.cpp is showing
much improved results again, since that code has lots of
nested-name-specifiers.
llvm-svn: 126737
DependentNameTypeLoc. Teach the recursive AST visitor and libclang how to
walk DependentNameTypeLoc nodes.
Also, teach libclang about TypedefDecl source ranges, so that we get
those. The massive churn in test/Index/recursive-cxx-member-calls.cpp
is a good thing: we're annotating a lot more of this test correctly
now.
llvm-svn: 126729
UnresolvedUsingValueDecl to use NestedNameSpecifierLoc rather than the
extremely-lossy NestedNameSpecifier/SourceRange pair it used to use,
improving source-location information.
Various infrastructure updates to support NestedNameSpecifierLoc:
- AST/PCH (de-)serialization
- Recursive AST visitor
- libclang traversal (including the first tests of this
functionality)
llvm-svn: 126459
way it keeps track of namespaces. Previously, we would map from the
namespace alias to its underlying namespace when building a
nested-name-specifier, losing source information in the process.
llvm-svn: 126358
Store in PCH the directory that the PCH was originally created in.
If a header file is not found at the path that we expect it to be and the PCH file
was moved from its original location, try to resolve the file by assuming that
header+PCH were moved together and the header is in the same place relative to the PCH.
llvm-svn: 125576
causing the deserialization of a large number of declarations when
writing the visible-updates record for the translation unit in C. This
takes us from:
*** AST File Statistics:
2 stat cache hits
6 stat cache misses
1/64463 source location entries read (0.001551%)
15606/16956 types read (92.038216%)
59266/89334 declarations read (66.342041%)
38952/61393 identifiers read (63.446976%)
0/7778 selectors read (0.000000%)
24192/34644 statements read (69.830276%)
388/8809 macros read (4.404586%)
2095/5189 lexical declcontexts read (40.373867%)
0/4587 visible declcontexts read (0.000000%)
0/7716 method pool entries read (0.000000%)
0 method pool misses
to
*** AST File Statistics:
2 stat cache hits
6 stat cache misses
1/64463 source location entries read (0.001551%)
26/16956 types read (0.153338%)
18/89334 declarations read (0.020149%)
145/61393 identifiers read (0.236183%)
0/7778 selectors read (0.000000%)
21/34644 statements read (0.060617%)
0/8809 macros read (0.000000%)
0/5189 lexical declcontexts read (0.000000%)
0/4587 visible declcontexts read (0.000000%)
0/7716 method pool entries read (0.000000%)
0 method pool misses
when generating a chained PCH for a header that #includes Cocoa.h
(from a PCH file) and adds one simple function declaration. The
generated PCH file is now only 9580 bytes (down from > 2MB).
llvm-svn: 125326
we would deserialize all of the macro definitions we knew about while
serializing the macro definitions at the end of the AST/PCH file. Even
though we skipped most of them (since they were unchanged), it's still
a performance problem.
Now, we do the standard AST/PCH chaining trick: watch what identifiers
are deserialized as macro names, and consider only those identifiers
(along with macro definitions that have been deserialized/written in
the source) when serializing the preprocessor state.
llvm-svn: 125324
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
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
- Add ref-qualifiers to the type system; they are part of the
canonical type. Print & profile ref-qualifiers
- Translate the ref-qualifier from the Declarator chunk for
functions to the function type.
- Diagnose mis-uses of ref-qualifiers w.r.t. static member
functions, free functions, constructors, destructors, etc.
- Add serialization and deserialization of ref-qualifiers.
llvm-svn: 124281
Inheritable attributes on declarations may be inherited by any later
redeclaration at merge time. By contrast, a non-inheritable attribute
will not be inherited by later redeclarations. Non-inheritable
attributes may be semantically analysed early, allowing them to
influence the redeclaration/overloading process.
Before this change, the "overloadable" attribute received special
handling to be treated as non-inheritable, while all other attributes
were treated as inheritable. This patch generalises the concept,
while removing a FIXME. Some CUDA location attributes are also marked
as non-inheritable in order to support special overloading semantics
(to be introduced in a later patch).
The patch introduces a new Attr subclass, InheritableAttr, from
which all inheritable attributes derive. Non-inheritable attributes
simply derive from Attr.
N.B. I did not review every attribute to determine whether it should
be marked non-inheritable. This can be done later on an incremental
basis, as this change does not affect default functionality.
llvm-svn: 123959
template template parameter pack that cannot be fully expanded because
its enclosing pack expansion could not be expanded. This form of
TemplateName plays the same role as SubstTemplateTypeParmPackType and
SubstNonTypeTemplateParmPackExpr do for template type parameter packs
and non-type template parameter packs, respectively.
We should now handle these multi-level pack expansion substitutions
anywhere. The largest remaining gap in our variadic-templates support
is that we cannot cope with non-type template parameter packs whose
type is a pack expansion.
llvm-svn: 123521
expansion, when it is known due to the substitution of an out
parameter pack. This allows us to properly handle substitution into
pack expansions that involve multiple parameter packs at different
template parameter levels, even when this substitution happens one
level at a time (as with partial specializations of member class
templates and the signatures of member function templates).
Note that the diagnostic we provide when there is an arity mismatch
between an outer parameter pack and an inner parameter pack in this
case isn't as clear as the normal diagnostic for an arity
mismatch. However, this doesn't matter because these cases are very,
very rare and (even then) only typically occur in a SFINAE context.
The other kinds of pack expansions (expression, template, etc.) still
need to support optional tracking of the number of expansions, and we
need the moral equivalent of SubstTemplateTypeParmPackType for
substituted argument packs of template template and non-type template
parameters.
llvm-svn: 123448
involve template parameter packs at multiple template levels that
occur within the signatures members of class templates (and partial
specializations thereof). This is a work-in-progress that is deficient
in several ways, notably:
- It only works for template type parameter packs, but we need to
also support non-type template parameter packs and template template
parameter packs.
- It doesn't keep track of the lengths of the substituted argument
packs in the expansion, so it can't properly diagnose length
mismatches.
However, this is a concrete step in the right direction.
llvm-svn: 123425
Douglas Gregor