components. These sometimes get synthetically added, and we don't want -Ifoo
and -I./foo to be treated fundamentally differently here.
llvm-svn: 224055
module, use the path from the module map file in preference to the path from
the .pcm file when resolving relative paths in the .pcm file. This allows
diagnostics (and .d output) to give relative paths if the module was found via
a relative path.
llvm-svn: 223577
the root of the module and use paths relative to that directory wherever
possible. This is a step towards allowing explicit modules to be relocated
without being rebuilt, which is important for some kinds of distributed builds,
for good paths in diagnostics, and for appropriate .d output.
This is a recommit of r223443, reverted in r223465; when joining together
imported file paths, we now use the system's separator rather than always
using '/'. This avoids path mismatches between the original module build and
the module user on Windows (at least, in some cases). A more comprehensive
fix will follow.
llvm-svn: 223539
the root of the module and use paths relative to that directory wherever
possible. This is a step towards allowing explicit modules to be relocated
without being rebuilt, which is important for some kinds of distributed builds,
for good paths in diagnostics, and for appropriate .d output.
llvm-svn: 223443
rather than trying to extract this information from the FileEntry after the
fact.
This has a number of beneficial effects. For instance, diagnostic messages for
failed module builds give a path relative to the "module root" rather than an
absolute file path, and the contents of the module includes file is no longer
dependent on what files the including TU happened to inspect prior to
triggering the module build.
llvm-svn: 223095
to be newer than we were expecting. That happens if .pcm's get moved between
file systems during a distributed build. (It's still not OK for them to actually
be different, though, so we still check the size and signature matches.)
llvm-svn: 222507
special member function.
No test yet: the only testcases we have for this issue are extremely complex.
Testcase will be added once I get a reasonable reduction.
llvm-svn: 222506
According to OpenMP standard, Section 2.12.6, atomic Construct, '#pragma omp atomic read' is allowed to be used only for expression statements of form 'v = x;', where x and v (as applicable) are both l-value expressions with scalar type. Patch adds checks for it.
llvm-svn: 222231
For all threadprivate variables which have constructor/destructor emit call to void __kmpc_threadprivate_register(ident_t * <Current Location>, void *<Original Global Addr>, kmpc_ctor <Constructor>, kmpc_cctor NULL, kmpc_dtor <Destructor>);
In expressions all references to such variables are replaced by calls to void *__kmpc_threadprivate_cached(ident_t *<Current Location>, kmp_int32 <Current Thread Id>, void *<Original Global Addr>, size_t <Size of Data>, void ***<Pointer to autogenerated cache – array of private copies of threadprivate variable>);
Test test/OpenMP/threadprivate_codegen.cpp checks that codegen is correct. Also it checks that codegen is correct after serialization/deserialization and one of passes verifies debug info.
Differential Revision: http://reviews.llvm.org/D4002
llvm-svn: 221663
is enabled. Unlike system headers, we want to be more careful about
modifications to user headers, because it's still easy to edit a header
while you're building.
llvm-svn: 221634
This is a new form of expression of the form:
(expr op ... op expr)
where one of the exprs is a parameter pack. It expands into
(expr1 op (expr2onwards op ... op expr))
(and likewise if the pack is on the right). The non-pack operand can be
omitted; in that case, an empty pack gives a fallback value or an error,
depending on the operator.
llvm-svn: 221573
We may need to verify the signature on subsequent imports as well, just
like we verify the size/modtime:
@import A;
@import B; // imports A
@import C; // imports A
llvm-svn: 221569
Use the bitmask to store the set of enabled sanitizers instead of a
bitfield. On the negative side, it makes syntax for querying the
set of enabled sanitizers a bit more clunky. On the positive side, we
will be able to use SanitizerKind to eventually implement the
new semantics for -fsanitize-recover= flag, that would allow us
to make some sanitizers recoverable, and some non-recoverable.
No functionality change.
llvm-svn: 221558
An updated implemnentation of VLA types capturing based on previously committed solution for Lambdas.
This version captures the whole VLA type instead of particular variables which are part of VLA size expression and allows to use previusly calculated size of VLA type in captured regions. Required for OpenMP.
Differential Revision: http://reviews.llvm.org/D5099
llvm-svn: 220850
explicitly using the resulting .pcm file. Unlike for an implicit module build,
we don't need nor want to require these flags to match between the module
and its users.
llvm-svn: 220780
Since the order of the IDs in the AST file (e.g. DeclIDs, SelectorIDs)
is not stable, it is not safe to load an AST file that depends on
another AST file that has been rebuilt since the importer was built,
even if "nothing changed". We previously used size and modtime to check
this, but I've seen cases where a module rebuilt quickly enough to foil
this check and caused very hard to debug build errors.
To save cycles when we're loading the AST, we just generate a random
nonce value and check that it hasn't changed when we load an imported
module, rather than actually hash the whole file.
This is slightly complicated by the fact that we need to verify the
signature inside addModule, since we might otherwise consider that a
mdoule is "OutOfDate" when really it is the importer that is out of
date. I didn't see any regressions in module load time after this
change.
llvm-svn: 220493
This allows a module to specify that it logically contains a file, but that
said file is non-modular and intended for textual inclusion. This allows
layering checks to work properly in the presence of such files.
llvm-svn: 220448
Implicit module builds are not well-suited to a lot of build systems. In
particular, they fare badly in distributed build systems, and they lead to
build artifacts that are not tracked as part of the usual dependency management
process. This change allows explicitly-built module files (which are already
supported through the -emit-module flag) to be explicitly loaded into a build,
allowing build systems to opt to manage module builds and dependencies
themselves.
This is only the first step in supporting such configurations, and it should
be considered experimental and subject to change or removal for now.
llvm-svn: 220359
This patch generates some helper variables which used as a private copies of the corresponding original variables inside an OpenMP 'parallel' directive. These generated variables are initialized by default (with the default constructor, if any). In outlined function references to original variables are replaced by the references to these private helper variables. At the end of the initialization of the private variables and implicit barier is set by calling __kmpc_barrier(...) runtime function to be sure that all threads were initialized using original values of the variables.
Differential Revision: http://reviews.llvm.org/D4752
llvm-svn: 220262
Now that we no longer add mappings when there are no local entities,
there is no need to always bump the size of the tables that correspond
to ContinuousRangeMaps.
llvm-svn: 220208
This is a better fix for 'duplicate key' problems in module continuous
range maps (vs what I added in r215810) by not adding any mappings at
all when there are no local entities. Now it also covers selectors,
which were not always being bumped because the record SELECTOR_OFFSET is
not always emitted. I'll back out most of r215810 in a future commit,
since it should no longer be needed.
llvm-svn: 220207
Plumb through the full QualType of the TemplateArgument::Declaration, as
it's insufficient to only know whether the type is a reference or
pointer (that was necessary for mangling, but insufficient for debug
info). This shouldn't increase the size of TemplateArgument as
TemplateArgument::Integer is still longer by another 32 bits.
Several bits of code were testing that the reference-ness of the
parameters matched, but this seemed to be insufficient (various other
features of the type could've mismatched and wouldn't've been caught)
and unnecessary, at least insofar as removing those tests didn't cause
anything to fail.
(Richard - perchaps you can hypothesize why any of these checks might
need to test reference-ness of the parameters (& explain why
reference-ness is part of the mangling - I would've figured that for the
reference-ness to be different, a prior template argument would have to
be different). I'd be happy to add them in/beef them up and add test
cases if there's a reason for them)
llvm-svn: 219900
initializers, and captured VLA types so that we can
answer questions like "is this a bit-field" without
looking at the enclosing DeclContext. NFC.
llvm-svn: 219522
Assertion failed: "Computed __func__ length differs from type!"
Reworked PredefinedExpr representation with internal StringLiteral field for function declaration.
Differential Revision: http://reviews.llvm.org/D5365
llvm-svn: 219393
Includes parsing and semantic analysis for 'omp teams' directive support from OpenMP 4.0. Adds additional analysis to 'omp target' directive with 'omp teams' directive.
llvm-svn: 219385
This patch generates some helper variables that used as private copies of the corresponding original variables inside an OpenMP 'parallel' directive. These generated variables are initialized by copy using values of the original variables (with the copy constructor, if any). For arrays, initializator is generated for single element and in the codegen procedure this initial value is automatically propagated between all elements of the private copy.
In outlined function, references to original variables are replaced by the references to these private helper variables. At the end of the initialization of the private variables an implicit barier is generated by calling __kmpc_barrier(...) runtime function to be sure that all threads were initialized using original values of the variables.
Differential Revision: http://reviews.llvm.org/D5140
llvm-svn: 219306
This patch generates some helper variables that used as private copies of the corresponding original variables inside an OpenMP 'parallel' directive. These generated variables are initialized by copy using values of the original variables (with the copy constructor, if any). For arrays, initializator is generated for single element and in the codegen procedure this initial value is automatically propagated between all elements of the private copy.
In outlined function, references to original variables are replaced by the references to these private helper variables. At the end of the initialization of the private variables an implicit barier is generated by calling __kmpc_barrier(...) runtime function to be sure that all threads were initialized using original values of the variables.
Differential Revision: http://reviews.llvm.org/D5140
llvm-svn: 219297
This patch generates some helper variables that used as private copies of the corresponding original variables inside an OpenMP 'parallel' directive. These generated variables are initialized by copy using values of the original variables (with the copy constructor, if any). For arrays, initializator is generated for single element and in the codegen procedure this initial value is automatically propagated between all elements of the private copy.
In outlined function, references to original variables are replaced by the references to these private helper variables. At the end of the initialization of the private variables an implicit barier is generated by calling __kmpc_barrier(...) runtime function to be sure that all threads were initialized using original values of the variables.
Differential Revision: http://reviews.llvm.org/D5140
llvm-svn: 219295
Includes parsing and semantic analysis for 'omp teams' directive support from OpenMP 4.0. Adds additional analysis to 'omp target' directive with 'omp teams' directive.
llvm-svn: 219197
This patch implements collapsing of the loops (in particular, in
presense of clause 'collapse'). It calculates number of iterations N
and expressions nesessary to calculate the nested loops counters
values based on new iteration variable (that goes from 0 to N-1)
in Sema. It also adds Codegen for 'omp simd', which uses
(and tests) this feature.
Differential Revision: http://reviews.llvm.org/D5184
llvm-svn: 218743
We build a NestedNameSpecifier that records the CXXRecordDecl in which
__super appeared. Name lookup is performed in all base classes of the
recorded CXXRecordDecl. Use of __super is allowed only inside class and
member function scope.
llvm-svn: 218484
1. We were hitting the NextIsPrevious assertion because we were trying
to merge decl chains that were independent of each other because we had
no Sema object to allow them to find existing decls. This is fixed by
delaying loading the "preloaded" decls until Sema is available.
2. We were trying to get identifier info from an annotation token, which
asserts. The fix is to special-case the module annotations in the
preprocessed output printer.
Fixed in a single commit because when you hit 1 you almost invariably
hit 2 as well.
llvm-svn: 217550
The warning warns on TypedefNameDecls -- typedefs and C++11 using aliases --
that are !isReferenced(). Since the isReferenced() bit on TypedefNameDecls
wasn't used for anything before this warning it wasn't always set correctly,
so this patch also adds a few missing MarkAnyDeclReferenced() calls in
various places for TypedefNameDecls.
This is made a bit complicated due to local typedefs possibly being used only
after their local scope has closed. Consider:
template <class T>
void template_fun(T t) {
typename T::Foo s3foo; // YYY
(void)s3foo;
}
void template_fun_user() {
struct Local {
typedef int Foo; // XXX
} p;
template_fun(p);
}
Here the typedef in XXX is only used at end-of-translation unit, when YYY in
template_fun() gets instantiated. To handle this, typedefs that are unused when
their scope exits are added to a set of potentially unused typedefs, and that
set gets checked at end-of-TU. Typedefs that are still unused at that point then
get warned on. There's also serialization code for this set, so that the
warning works with precompiled headers and modules. For modules, the warning
is emitted when the module is built, for precompiled headers each time the
header gets used.
Finally, consider a function using C++14 auto return types to return a local
type defined in a header:
auto f() {
struct S { typedef int a; };
return S();
}
Here, the typedef escapes its local scope and could be used by only some
translation units including the header. To not warn on this, add a
RecursiveASTVisitor that marks all delcs on local types returned from auto
functions as referenced. (Except if it's a function with internal linkage, or
the decls are private and the local type has no friends -- in these cases, it
_is_ safe to warn.)
Several of the included testcases (most of the interesting ones) were provided
by Richard Smith.
(gcc's spelling -Wunused-local-typedefs is supported as an alias for this
warning.)
llvm-svn: 217298
on CXXRecordDecls when merging definitions, and make it pass by not trying to
save and restore this flag across AST serialization/deserialization. For
CXXRecordDecls, we have a separate mechanism to manage this.
llvm-svn: 216633
Only those callers who are dynamically passing ownership should need the
3 argument form. Those accepting the default ("do pass ownership")
should do so explicitly with a unique_ptr now.
llvm-svn: 216614
pattern of an alias template declaration. Use this to merge alias templates
properly when they're members of class template specializations.
llvm-svn: 216437
declarations. We can't expect to find them in the canonical definition
of the class, because that's not where they live.
This means we no longer reject real ODR violations with friend declarations,
but we weren't consistently doing so anyway.
llvm-svn: 216369
Predefined decls like 'Protocol' in objc are not loaded from AST files,
so we cannot rely on loading the canonical decl to complete the redecl
chain for redeclarations of these decls. The broken redecl chain was
non-circular, so looping over redecls() would hang.
llvm-svn: 215929
tables that correspond to ContinuousRangeMaps, since the keys to those
maps need to be unique, or we may map to the wrong offset.
This fixes a crash + malformed AST file seen when loading some modules
that import Cocoa on Darwin, which is a module with no contents except
imports of other modules. Unfortunately I have not been able to find a
reduced test case that reproduces this problem.
Also add an assert that we aren't mapping one key to multiple values
in CRM. We ought to be able to say there are no duplicate keys at all,
but there are a bunch of 0 -> 0 mappings that are showing up, probably
coming from the source location table.
llvm-svn: 215810
members from all redefinitions of a class that have them, in case the special
member is defined in one module but only declared in another.
llvm-svn: 215675
definitions (because some other declaration declares a special member that
isn't present in the canonical definition), we need to search *all* of them; we
can't just stop when we find the requested name in any of the definitions,
because that can fail to find things (and in particular, it can fail to find
the member of the canonical declaration and return a bogus ODR failure).
llvm-svn: 215612
redefinitions of that namespace have already been loaded. When writing out the
names in a namespace, if we see a name that is locally declared and had
imported declarations merged on top of it, export the local declaration as the
lookup result, because it will be the most recent declaration of that entity in
the redeclaration chain of an importer of the module.
llvm-svn: 215518
We already verified the primary module map file (either the one that
defines the top-level module, or the one that allows inferring it if it
is an inferred framework module). Now we also verify any other module
map files that define submodules, such as when there is a
module.private.modulemap file.
llvm-svn: 215455
According to the gcc docs, -include uses the current working directory
for the lookup instead of the main source file.
This patch gets rid of NormalizeIncludePath (which relied on an
implementation detail of FileManager / FileEntry for the include path
logic to work), and instead hands the correct lookup information down to
LookupFile.
This will allow us to change the FileEntry's behavior regarding its Name
caching.
llvm-svn: 215433
class Module. It's almost always going to be the same as
getContainingModule() for top-level modules, so just add a map to cover
the remaining cases. This lets us do less bookkeeping to keep the
ModuleMap fields up to date.
llvm-svn: 215268
also emit the updated 'operator delete' looked up for that destructor. Switch
from UpdateDecl to an actual update record when this happens due to implicitly
defining a special member function and unify this code path and the one for
instantiating a function definition.
llvm-svn: 215132
intent when we added remark support, but was never implemented in the general
case, because the first -R flags didn't need it. (-Rpass= had special handling
to accomodate its argument.)
-Rno-foo, -Reverything, and -Rno-everything can be used to turn off a remark,
or to turn on or off all remarks. Per discussion on cfe-commits, -Weverything
does not affect remarks, and -Reverything does not affect warnings or errors.
The only "real" -R flag we have right now is -Rmodule-build; that flag is
effectively renamed from -Wmodule-build to -Rmodule-build by this change.
-Wpass and -Wno-pass (and their friends) are also renamed to -Rpass and
-Rno-pass by this change; it's not completely clear whether we intended to have
a -Rpass (with no =pattern), but that is unchanged by this commit, other than
the flag name. The default pattern is effectively one which matches no passes.
In future, we may want to make the default pattern be .*, so that -Reverything
works for -Rpass properly.
llvm-svn: 215046
they're somehow missing a body. Looks like this was left behind when the loop
was generalized, and it's not been problematic before because without modules,
a used, implicit special member function declaration must be a definition.
This was resulting in us trying to emit a constructor declaration rather than
a definition, and producing a constructor missing its member initializers.
llvm-svn: 214473
of a function has a resolved exception specification, then all declarations of
the function do.
We should probably improve the AST representation to make this implicit (perhaps
only store the exception specification on the canonical declaration), but this
fixes things for now.
The testcase for this (which used to assert) also exposes the actual bug I was
trying to reduce here: we sometimes fail to emit the body of an imported
special member function definition. Fix for that to follow.
llvm-svn: 214458
FunctionProtoType::ExtProtoInfo. Most of the users of these fields don't care
about the other ExtProtoInfo bits and just want to talk about the exception
specification.
llvm-svn: 214450
* Add abbreviation for CXXMethodDecl and for FunctionProtoType. These come up
a *lot* in C++ modules.
* Allow typedef declarations to use the abbreviation if they're class members,
or if they're used.
In passing, add more record name records for Clang AST node kinds.
The downside is that we had already used up our allotment of 12 abbreviations,
so this pushes us to an extra bit on each record to support the extra abbrev
kinds, which increases file size by ~1%. This patch *barely* pays for that
through the other improvements, but we've got room for another 18 abbrevs,
so we should be able to make it much more profitable with future changes.
llvm-svn: 214024
* Track override set across module load and save
* Track originating module to allow proper re-export of #undef
* Make override set properly transitive when it picks up a #undef
This fixes nearly all of the remaining macro issues with self-host.
llvm-svn: 213922
thorough tests.
Original commit message:
[modules] Fix macro hiding bug exposed if:
* A submodule of module A is imported into module B
* Another submodule of module A that is not imported into B exports a macro
* Some submodule of module B also exports a definition of the macro, and
happens to be the first submodule of B that imports module A.
In this case, we would incorrectly determine that A's macro redefines B's
macro, and so we don't need to re-export B's macro at all.
This happens with the 'assert' macro in an LLVM self-host. =(
llvm-svn: 213416
This is breaking the system modules on Darwin, because something that
was defined and re-exported no longer is. Might be this patch, or might
just be a really poor interaction with an existing visibility bug.
This reverts commit r213348.
llvm-svn: 213395
* A submodule of module A is imported into module B
* Another submodule of module A that is not imported into B exports a macro
* Some submodule of module B also exports a definition of the macro, and
happens to be the first submodule of B that imports module A.
In this case, we would incorrectly determine that A's macro redefines B's
macro, and so we don't need to re-export B's macro at all.
This happens with the 'assert' macro in an LLVM self-host. =(
llvm-svn: 213348
-- a constructor list initialization that unpacked an initializer list into
constructor arguments and
-- a list initialization that created as std::initializer_list and passed it
as the first argument to a constructor
in the AST. Use this flag while instantiating templates to provide the right
semantics for the resulting initialization.
llvm-svn: 213224
redeclaration chains when pulling in a declaration. We need the redecl chain
unless we know some other declaration will trigger it to be pulled in; that
happens if our originally-canonical declaration had all the knowledge that
we have (and isn't us).
llvm-svn: 213043
into their container; we won't find them there. These things are already being
merged when they're added to their primary template's folding set, so this
merging is redundant (and causes us to reject-valid because we think we've
found an odr violation).
llvm-svn: 212788
Successfully loaded module files may be referenced in other
ModuleManagers, so don't invalidate them. Two related things are fixed:
1) I thought the last module in the manager was always the one that
failed, but it isn't. So check explicitly against the list of
vetted modules from ReadASTCore.
2) We now keep the file descriptor of pcm file open, which avoids the
possibility of having two different pcms for the same module loaded when
building in parallel with headers being modified during a build.
<rdar://problem/16835846>
llvm-svn: 211330
This reverts commit r211096. Looks like it broke the msvc build:
SemaOpenMP.cpp(140) : error C4519: default template arguments are only allowed on a class template
llvm-svn: 211113
This begins to address cognitive dissonance caused by treating the Note
diagnostic level as a severity in the diagnostic engine.
No change in functionality.
llvm-svn: 210758
Diagnostic mappings are used to calculate the final severity of diagnostic
instances.
Detangle the implementation to reflect the terminology used in documentation
and bindings.
No change in functionality.
llvm-svn: 210518
This reapplies r209910 with a fix for the assertion failures hit on the
buildbots.
original commit message:
I thought we could get away without this, but it means that the
FileEntry objects actually refer to the wrong files, since pcms are not
updated inplace, they are atomically renamed into place after compiling
a module.
So we are close to the original behaviour of invalidating the cache for
all modules being removed, but now we should only invalidate the ones
that depend on whichever module failed to load.
Unfortunately I haven't come up with a new test that didn't require
a race between parallel invocations of clang.
<rdar://problem/17038180>
llvm-svn: 209922
I thought we could get away without this, but it means that the
FileEntry objects actually refer to the wrong files, since pcms are not
updated inplace, they are atomically renamed into place after compiling
a module.
So we are close to the original behaviour of invalidating the cache for
all modules being removed, but now we should only invalidate the ones
that depend on whichever module failed to load.
Unfortunately I haven't come up with a new test that didn't require
a race between parallel invocations of clang.
<rdar://problem/17038180>
llvm-svn: 209910
member functions), ensure that the redecl chain never transitions from 'inline'
to 'not inline', since that violates an AST invariant.
llvm-svn: 209794
This patch implements support for selectively disabling optimizations on a
range of function definitions through a pragma. The implementation is that
all function definitions in the range are decorated with attribute
'optnone'.
#pragma clang optimize off
// All function definitions in here are decorated with 'optnone'.
#pragma clang optimize on
// Compilation resumes as normal.
llvm-svn: 209510
instantiated in another module, and the instantiation uses a partial
specialization, include the partial specialization and its template arguments
in the update record. We'll need them if someone imports the second module and
tries to instantiate a member of the template.
llvm-svn: 209472
ASTReaderListener's documentation states that visitInputFile will be
called based on the return values of needsInputFileVisitation and
needsSystemInputFileVisitation, but ChainedASTReaderListener may call
these methods on a child listener based on the values returned by the
other child.
Even worse, the calls to visitInputFile may be short-circuited due to
the use of the boolean or, so the calls to visit may not occur at all
for the second listener.
This updates ChainedASTReaderListener::visitInputFile to propagate the
ASTReaderListener behaviour to both children.
llvm-svn: 209394
It appears that Windows doesn't like renaming over open files, which we
do in clearOutputFiles. The file being compiled should be safe to
removed, but this isn't very satisfying - we don't want to manually
manage the lifetime of files we cannot prove have no references.
llvm-svn: 209195
declaration of that entity in from one of those modules, keep track of the fact
that we've not completed the redeclaration chain yet so that we can pull the
remaining declarations in from the other module if they're needed.
llvm-svn: 209161
Follow-up fix for 209138. Actually, since we already have this file
open, we don't want to refresh the stat() info, since that might be
newer than what we have open (bad!).
llvm-svn: 209143
FileManager::invalidateCache is not safe to call when there may be
existing references to the file. What module load failure needs is
to refresh so stale stat() info isn't stored.
This may be the last user of invalidateCache; I'll take a look and
remove it if possible in a future commit.
This caused a use-after-free error as well as a spurious error message
that a module was "found in both 'X.pcm' and 'X.pcm'" in some cases.
llvm-svn: 209138
ensure that querying the first declaration for its most recent declaration
checks for redeclarations from the imported module.
This works as follows:
* The 'most recent' pointer on a canonical declaration grows a pointer to the
external AST source and a generation number (space- and time-optimized for
the case where there is no external source).
* Each time the 'most recent' pointer is queried, if it has an external source,
we check whether it's up to date, and update it if not.
* The ancillary data stored on the canonical declaration is allocated lazily
to avoid filling it in for declarations that end up being non-canonical.
We'll still perform a redundant (ASTContext) allocation if someone asks for
the most recent declaration from a decl before setPreviousDecl is called,
but such cases are probably all bugs, and are now easy to find.
Some finessing is still in order here -- in particular, we use a very general
mechanism for handling the DefinitionData pointer on CXXRecordData, and a more
targeted approach would be more compact.
Also, the MayHaveOutOfDateDef mechanism should now be expunged, since it was
addressing only a corner of the full problem space here. That's not covered
by this patch.
Early performance benchmarks show that this makes no measurable difference to
Clang performance without modules enabled (and fixes a major correctness issue
with modules enabled). I'll revert if a full performance comparison shows any
problems.
llvm-svn: 209046
Use this to fix the leak of DeserializedDeclsDumper and DeserializedDeclsChecker
in FrontendAction (found by LSan), PR19560.
The "delete this" bool is necessary because both PCHGenerator and ASTUnit
return the same object from both getDeserializationListener() and
getASTMutationListener(), so ASTReader can't just have a unique_ptr.
It's also not possible to just let FrontendAction (or CompilerInstance) own
these listeners due to lifetime issues (see comments on PR19560).
Finally, ASTDeserializationListener can't easily be refcounted, since several of
the current listeners are allocated on the stack.
Having this bool isn't ideal, but it's a pattern that's used in other places in
the codebase too, and it seems better than leaking.
llvm-svn: 208277
Summary:
Previously, we would generate a single name for all reference
temporaries and allow LLVM to rename them for us. Instead, number the
reference temporaries as we build them in Sema.
Reviewers: rsmith
Subscribers: cfe-commits
Differential Revision: http://reviews.llvm.org/D3554
llvm-svn: 207776
We need to open an ASTFile while checking its expected size and
modification time, or another clang instance can modify the file between
the stat() and the open().
llvm-svn: 207735
This fixes a bug where an update record causes us to load an entity that refers
to an entity we've not finished loading yet, resulting in badness.
llvm-svn: 207603
Fixed by moving ProcessWarningOptions from Frontend into Basic. All of
the dependencies for ProcessWarningOptions were already in Basic, so
this was a small change.
llvm-svn: 207549
This patch checks whether the diagnostic options that could lead to
errors (principally -Werror) are consistent between when a module was
built and when it is loaded. If there are new -Werror flags, then the
module is rebuilt. In order to canonicalize the options we do this
check at the level of the constructed DiagnosticsEngine, which contains
the final set of diag to diagnostic level mappings. Currently we only
rebuild with the new diagnostic options, but we intend to refine this in
the future to include the union of the new and old flags, since we know
the old ones did not cause errors. System modules are only rebuilt when
-Wsystem-headers is enabled.
One oddity is that unlike checking language options, we don’t perform
this diagnostic option checking when loading from a precompiled header.
The reason for this is that the compiler cannot rebuild the PCH, so
anything that requires it to be rebuilt effectively leaks into the build
system. And in this case, that would mean the build system
understanding the complex relationship between diagnostic options and
the underlying diagnostic mappings, which is unreasonable. Skipping the
check is safe, because these options do not affect the generated AST.
You simply won’t get new build errors due to changed -Werror options
automatically, which is also true for non-module cases.
llvm-svn: 207477
together. This is extremely hairy, because in general we need to have loaded
both the template and the pattern before we can determine whether either should
be merged, so we temporarily violate the rule that all merging happens before
reading a decl ends, but *only* in the case where a template's pattern is being
loaded while loading the template itself.
In order to accomodate this for class templates, delay loading the injected
class name type for the pattern of the template until after we've loaded the
template itself, if we happen to load the template first.
llvm-svn: 207063
This paves the way to making OnDiskHashTable work with hashes that are
not 32 bits wide and to making OnDiskHashTable work very large hash
tables. The LLVM change to use these types is upcoming.
llvm-svn: 206640
To differentiate between two modules with the same name, we will
consider the path the module map file that they are defined by* part of
the ‘key’ for looking up the precompiled module (pcm file).
Specifically, this patch renames the precompiled module (pcm) files from
cache-path/<module hash>/Foo.pcm
to
cache-path/<module hash>/Foo-<hash of module map path>.pcm
In addition, I’ve taught the ASTReader to re-resolve the names of
imported modules during module loading so that if the header search
context changes between when a module was originally built and when it
is loaded we can rebuild it if necessary. For example, if module A
imports module B
first time:
clang -I /path/to/A -I /path/to/B ...
second time:
clang -I /path/to/A -I /different/path/to/B ...
will now rebuild A as expected.
* in the case of inferred modules, we use the module map file that
allowed the inference, not the __inferred_module.map file, since the
inferred file path is the same for every inferred module.
llvm-svn: 206201
Currently the on disk hash table's key_iterator and data_iterator make
the assumption that the table data starts exactly four bytes after the
base of the table. This happens to be true for all of the tables we
currently iterate over, but not for all of the OnDiskHashTables we
currently use. For example, key_ and data_iterator would iterate over
meaningless data if they were used on the hash tables in PTHLexer.
We make the API safer by breaking this into two types. One doesn't
have the iterators, and the other must be told where the payload
starts.
llvm-svn: 206189
Committed this by accident before it was done last time.
Original message:
Rather than rolling our own functions to read little endian data
from a buffer, we can use the support in llvm's Endian.h.
No functional change.
llvm-svn: 205062
Committed this by accident before it was done last time.
Original message:
Rather than rolling our own functions to write little endian data
to an ostream, we can use the support in llvm's EndianStream.h.
No functional change.
llvm-svn: 205061
Rather than rolling our own functions to read little endian data from
a buffer, we can use the support in llvm's Endian.h.
No functional change.
llvm-svn: 205045
Rather than rolling our own functions to write little endian data to
an ostream, we can use the support in llvm's EndianStream.h.
No functional change.
llvm-svn: 205044
correctly order comments in SourceManager::isBeforeInTranslationUnit() order
Unfortunately, this is not as simple as it was implemented previously, and
actually requires doing a merge sort.
llvm-svn: 204936
an out-of-date external decls list). This happens if we declare some names,
force the lookup table for the decl context to be built, import a module that
adds more decls for the name, then write out our module without looking up the
name.
llvm-svn: 204694
specialization from a module. (This can also happen for function template
specializations in PCHs if they're instantiated eagerly, because they're
constexpr or have a deduced return type.)
llvm-svn: 204547
at which that PCH imported each visible submodule of the module. Such locations
are needed when synthesizing macro directives resulting from the import.
llvm-svn: 204417
block as decl and type emission. This allows decl updates include statements
and expressions. No functionality change (but the generated PCM files are
incompatible with earlier versions of Clang).
llvm-svn: 204385
What's going on in the test case (without the patch applied) is this:
When the header is parsed, decltype(B()) is canonicalized to decltype(Y()),
because that was the first parsed equivalent decltype expression. Hence, the
TemplateSpecializationType for Id<decltype(B())> ends up with
SubstTemplateTypeParmType(T, decltype(Y())) as the AliasedType member.
When the PCH file is included and the AST reader reads Id<decltype(B())>, it
sees decltype(B()) before decltype(Y()). So, this time decltype(B()) ends up
being the canonical type for both decltypes, which leads to an assert violation
when the reader calls getSubstTemplateTypeParmType with the non-canonical
decltype(Y()) as the replacement type.
Reviewers: rsmith
Reviewed By: rsmith
CC: cfe-commits, aemerson
Differential Revision: http://llvm-reviews.chandlerc.com/D3073
llvm-svn: 204005
This is because the PCH is tied to the module files, if one of the module files changes or gets removed
the build system should re-build the PCH file.
rdar://16321245
llvm-svn: 203885
Drive-by fixing some incorrect types where a for loop would be improperly using ObjCInterfaceDecl::protocol_iterator. No functional changes in these cases.
llvm-svn: 203842
When building an AST file, we don't want to output HeaderFileInfo
structures for files that are not actually used as headers in the
current context. This can lead to assuming that unrelated files have
include counts of 0, defeating multiple-include prevention.
This is accomplished by adding an IsValid bit to the HFI.
llvm-svn: 203813
Add module dependencies to the dependency files created by -MD/-MMD/etc.
by attaching an ASTReaderListener that will call into the dependency
file generator when a module input file is seen in the serialized AST.
llvm-svn: 203208
submodule macro overriding within the same top-level module (necessary for the
testcase to be remotely reasonable). Incidentally reduces the number of libc++
testsuite regressions with modules enabled from 7 to 6.
llvm-svn: 203063
predicate. The wrapper used by SetVector was erroneously requiring an
adaptable predicate. It has been fixed and we really don't want to
require an indirect call for every predicate evaluation.
llvm-svn: 202744
it, importers of B should not see the macro. This is complicated by the fact
that A's macro could also be visible through a different path. The rules (as
hashed out on cfe-commits) are included as a documentation update in this
change.
With this, the number of regressions in libc++'s testsuite when modules are
enabled drops from 47 to 7. Those remaining 7 are also macro-related, and are
due to remaining bugs in this change (in particular, the handling of submodules
is imperfect).
llvm-svn: 202560
This does;
- clang_tablegen() adds each tblgen'd target to global property CLANG_TABLEGEN_TARGETS as list.
- List of targets is added to LLVM_COMMON_DEPENDS.
- all clang libraries and targets depend on generated headers.
You might wonder this would be regression, but in fact, this is little loss.
- Almost all of clang libraries depend on tblgen'd files and clang-tblgen.
- clang-tblgen may cause short stall-out but doesn't cause unconditional rebuild.
- Each library's dependencies to tblgen'd files might vary along headers' structure.
It made hard to track and update *really optimal* dependencies.
Each dependency to intrinsics_gen and ClangSACheckers is left as DEPENDS.
llvm-svn: 201842
Previously reverted in r201755 due to causing an assertion failure.
I've removed the offending assertion, and taught the CompilerInstance to
create a default virtual file system inside createFileManager. In the
future, we should be able to reach into the CompilerInvocation to
customize this behaviour without breaking clients that don't care.
llvm-svn: 201818
the build
When Clang loads the module, it verifies the user source files that the module
was built from. If any file was changed, the module is rebuilt. There are two
problems with this:
1. correctness: we don't verify system files (there are too many of them, and
stat'ing all of them would take a lot of time);
2. performance: the same module file is verified again and again during a
single build.
This change allows the build system to optimize source file verification. The
idea is based on the fact that while the project is being built, the source
files don't change. This allows us to verify the module only once during a
single build session. The build system passes a flag,
-fbuild-session-timestamp=, to inform Clang of the time when the build started.
The build system also requests to enable this feature by passing
-fmodules-validate-once-per-build-session. If these flags are not passed, the
behavior is not changed. When Clang verifies the module the first time, it
writes out a timestamp file. Then, when Clang loads the module the second
time, it finds a timestamp file, so it can compare the verification timestamp
of the module with the time when the build started. If the verification
timestamp is too old, the module is verified again, and the timestamp file is
updated.
llvm-svn: 201224
This triggered a miscompilation of code using Boost's function_template.hpp
when it was included inside a PCH file. A local static within
that header would be treated as local extern, resulting in the wrong
mangling. This only occurred during PCH deserialization.
Fixes <rdar://problem/15975816> and <rdar://problem/15926311>.
llvm-svn: 201130
We don't stat the system headers to check for stalenes during regular
PCH loading for performance reasons. When explicitly saying
-verify-pch, we want to check all the dependencies - user or system.
llvm-svn: 200979
This option will:
- load the given pch file
- verify it is not out of date by stat'ing dependencies, and
- return 0 on success and non-zero on error
llvm-svn: 200884
Add the ImportDecl to the set of interesting delcarations that are
deserialized eagerly when an AST file is loaded (rather than lazily like
most decls). This is required to get auto linking to work when there is
no explicit import in the main file. Also resolve a FIXME to rename
'ExternalDefinitions', since that is only one of the things that need eager
deserialization. The new name is 'EagerlyDeserializedDecls'. The corresponding
AST bitcode is also renamed.
llvm-svn: 200505
A return type is the declared or deduced part of the function type specified in
the declaration.
A result type is the (potentially adjusted) type of the value of an expression
that calls the function.
Rule of thumb:
* Declarations have return types and parameters.
* Expressions have result types and arguments.
llvm-svn: 200082
Fix a perennial source of confusion in the clang type system: Declarations and
function prototypes have parameters to which arguments are supplied, so calling
these 'arguments' was a stretch even in C mode, let alone C++ where default
arguments, templates and overloading make the distinction important to get
right.
Readability win across the board, especially in the casting, ADL and
overloading implementations which make a lot more sense at a glance now.
Will keep an eye on the builders and update dependent projects shortly.
No functional change.
llvm-svn: 199686
Show the top-level pch file as the culprit, rather than the immediate
dependency when a pch file imports a pcm from a module. To clarify the
relationship, the pch import stack is printed as notes. The old behaviour was
misleading when a pch imported a pcm (from a module), since removing the pcm
would not fix the problem, whereas rebuilding the pch would.
llvm-svn: 199446
This makes the C++ ABI depend entirely on the target: MS ABI for -win32 triples,
Itanium otherwise. It's no longer possible to do weird combinations.
To be able to run a test with a specific ABI without constraining it to a
specific triple, new substitutions are added to lit: %itanium_abi_triple and
%ms_abi_triple can be used to get the current target triple adjusted to the
desired ABI. For example, if the test suite is running with the i686-pc-win32
target, %itanium_abi_triple will expand to i686-pc-mingw32.
Differential Revision: http://llvm-reviews.chandlerc.com/D2545
llvm-svn: 199250
- Remove the additions to ObjCMethodDecl & ObjCIVarDecl that were getting de/serialized and consolidate
all functionality for the checking for this warning in Sema::DiagnoseUnusedBackingIvarInAccessor
- Don't check immediately after the method body is finished, check when the @implementation is finished.
This is so we can see if the ivar was referenced by any other method, even if the method was defined after the accessor.
- Don't silence the warning if any method is called from the accessor silence it if the accessor delegates to another method via self.
rdar://15727325
llvm-svn: 198432
Remove UnaryTypeTraitExpr and switch all remaining type trait related handling
over to TypeTraitExpr.
The UTT/BTT/TT enum prefix and evaluation code is retained pending further
cleanup.
This is part of the ongoing work to unify type traits following the removal of
BinaryTypeTraitExpr in r197273.
llvm-svn: 198271
This caused some crazy crashes involving std::unordered_map being
deserialized from a PCH file and then template instantiation requiring
an explicit instantiation location; unfortunately I don't really know
how to come up with a minimal test case.
llvm-svn: 197764
We started by trying to deserialize decltype(func-param) in a trailing return
type, which causes the function parameter decl to be deserialized, which pulls
in the function decl, which pulls the function type, which pulls the same
decltype() in the return type, and then we crashed.
llvm-svn: 197644
There's nothing special about type traits accepting two arguments.
This commit eliminates BinaryTypeTraitExpr and switches all related handling
over to TypeTraitExpr.
Also fixes a CodeGen failure with variadic type traits appearing in a
non-constant expression.
The BTT/TT prefix and evaluation code is retained as-is for now but will soon
be further cleaned up.
This is part of the ongoing work to unify type traits.
llvm-svn: 197273
more than one such initializer in a union, make mem-initializers override
default initializers for other union members, handle anonymous unions with
anonymous struct members better. Fix a couple of semi-related bugs exposed by
the tests for same.
llvm-svn: 196892
Summary:
In general, this type node can be used to represent any type adjustment
that occurs implicitly without losing type sugar. The immediate use of
this is to adjust the calling conventions of member function pointer
types without breaking template instantiation.
Fixes PR17996.
Reviewers: rsmith
Differential Revision: http://llvm-reviews.chandlerc.com/D2332
llvm-svn: 196451
designated initializers of an interface.
If the interface declaration does not have methods marked as designated
initializers then the interface inherits the designated initializers of
its super class.
llvm-svn: 196315
This change fixes Richard's testcase for r193815. Now we include non-explicit
submodules into the list of exports.
The test failed previously because:
- recursive_visibility_a1.inner is not imported (only recursive_visibility_a1 is),
- thus the 'inner' submodule is not showing up in any of the import lists,
- and because of this getExportedModules() is not returning the
correct module set -- it only considers modules that are imported.
The fix is to make Module::getExportedModules() include non-explicit submodules
into the list of exports.
llvm-svn: 194018
would be deleted are still declared, but are ignored by overload resolution.
Also, don't delete such members if a subobject has no corresponding move
operation and a non-trivial copy. This causes us to implicitly declare move
operations in more cases, but risks move-assigning virtual bases multiple
times in some circumstances (a warning for that is to follow).
llvm-svn: 193969
requires ! feature
The purpose of this is to allow (for instance) the module map for /usr/include
to exclude <tgmath.h> and <complex.h> when building in C++ (these headers are
instead provided by the C++ standard library in this case, and the glibc C
<tgmath.h> header would otherwise try to include <complex.h>, resulting in a
module cycle).
llvm-svn: 193549
A prior commit of this patch was reverted because it was within the blamelist's purview of a failing test. The failure of that test has been addressed here: http://lists.cs.uiuc.edu/pipermail/cfe-commits/Week-of-Mon-20131021/091546.html. Therefore I am recommitting this patch (all tests pass on windows, except for the usual modules & index suspects that never pass on my box).
Some background: Both Doug and Richard had asked me in Chicago to remove the circular reference in CXXRecordDecl to LambdaExpr by factoring out and storing the needed information from LambdaExpr directly into CXXRecordDecl.
In addition, I have added an IsGenericLambda flag - this makes life a little easier when we implement capturing, and are Sema-analyzing the body of a lambda (and the calloperator hasn't been wired to the closure class yet). Any inner lambdas can have potential captures that could require walking up the scope chain and checking if any generic lambdas are capture-ready. This 'bit' makes some of that checking easier.
No change in functionality.
This patch was approved by Doug with minor modifications (comments were cleaned up, and all data members were converted from bool/enum to unsigned, as requested):
http://llvm-reviews.chandlerc.com/D1856
Thanks!
llvm-svn: 193246
They were causing CodeGenCXX/mangle-exprs.cpp to fail.
Revert "Remove the circular reference to LambdaExpr in CXXRecordDecl."
Revert "Again: Teach TreeTransform and family how to transform generic lambdas nested within templates and themselves."
llvm-svn: 193226
Both Doug and Richard had asked me to remove the circular reference in CXXRecordDecl to LambdaExpr by factoring out and storing the needed information from LambdaExpr directly into CXXRecordDecl.
No change in functionality.
In addition, I have added an IsGenericLambda flag - this makes life a little easier when we implement capturing, and are Sema-analyzing the body of a lambda (and the calloperator hasn't been wired to the closure class yet). Any inner lambdas can have potential captures that could require walking up the scope chain and checking if any generic lambdas are capture-ready. This 'bit' makes some of that checking easier.
This patch was approved by Doug with minor modifications (comments were cleaned up, and all data members were converted from bool/enum to unsigned, as requested):
http://llvm-reviews.chandlerc.com/D1856
Thanks!
llvm-svn: 193223
modules.
With this fixed, I no longer see any test regressions in the libc++ test suite
when enabling a single-module module.map for libc++ (other than issues with my
system headers).
llvm-svn: 193219
* NamedDecl and CXXMethodDecl were missing getMostRecentDecl.
* The const version can just forward to the non const.
* getMostRecentDecl can use cast instead of cast_or_null.
This then removes some casts from the callers.
llvm-svn: 193039
If we have multiple definitions of the same entity from different modules, we
nominate the first definition which we see as being the canonical definition.
If we load a declaration from a different definition and we can't find a
corresponding declaration in the canonical definition, issue a diagnostic.
This is insufficient to prevent things from going horribly wrong in all cases
-- we might be in the middle of emitting IR for a function when we trigger some
deserialization and discover that it refers to an incoherent piece of the AST,
by which point it's probably too late to bail out -- but we'll at least produce
a diagnostic.
llvm-svn: 192950
This change doesn't go all the way to making fields redeclarable; instead, it
makes them 'mergeable', which means we can find the canonical declaration, but
not much else (and for a declaration that's not from a module, the canonical
declaration is always that declaration).
llvm-svn: 192092
Specifically, the following features are not included in this commit:
- any sort of capturing within generic lambdas
- generic lambdas within template functions and nested
within other generic lambdas
- conversion operator for captureless lambdas
- ensuring all visitors are generic lambda aware
(Although I have gotten some useful feedback on my patches of the above and will be incorporating that as I submit those patches for commit)
As an example of what compiles through this commit:
template <class F1, class F2>
struct overload : F1, F2 {
using F1::operator();
using F2::operator();
overload(F1 f1, F2 f2) : F1(f1), F2(f2) { }
};
auto Recursive = [](auto Self, auto h, auto ... rest) {
return 1 + Self(Self, rest...);
};
auto Base = [](auto Self, auto h) {
return 1;
};
overload<decltype(Base), decltype(Recursive)> O(Base, Recursive);
int num_params = O(O, 5, 3, "abc", 3.14, 'a');
Please see attached tests for more examples.
This patch has been reviewed by Doug and Richard. Minor changes (non-functionality affecting) have been made since both of them formally looked at it, but the changes involve removal of supernumerary return type deduction changes (since they are now redundant, with richard having committed a recent patch to address return type deduction for C++11 lambdas using C++14 semantics).
Some implementation notes:
- Add a new Declarator context => LambdaExprParameterContext to
clang::Declarator to allow the use of 'auto' in declaring generic
lambda parameters
- Add various helpers to CXXRecordDecl to facilitate identifying
and querying a closure class
- LambdaScopeInfo (which maintains the current lambda's Sema state)
was augmented to house the current depth of the template being
parsed (id est the Parser calls Sema::RecordParsingTemplateParameterDepth)
so that SemaType.cpp::ConvertDeclSpecToType may use it to immediately
generate a template-parameter-type when 'auto' is parsed in a generic
lambda parameter context. (i.e we do NOT use AutoType deduced to
a template parameter type - Richard seemed ok with this approach).
We encode that this template type was generated from an auto by simply
adding $auto to the name which can be used for better diagnostics if needed.
- SemaLambda.h was added to hold some common lambda utility
functions (this file is likely to grow ...)
- Teach Sema::ActOnStartOfFunctionDef to check whether it
is being called to instantiate a generic lambda's call
operator, and if so, push an appropriately prepared
LambdaScopeInfo object on the stack.
- various tests were added - but much more will be needed.
There is obviously more work to be done, and both Richard (weakly) and Doug (strongly)
have requested that LambdaExpr be removed form the CXXRecordDecl LambdaDefinitionaData
in a future patch which is forthcoming.
A greatful thanks to all reviewers including Eli Friedman, James Dennett,
and especially the two gracious wizards (Richard Smith and Doug Gregor)
who spent hours providing feedback (in person in Chicago and on the mailing lists).
And yet I am certain that I have allowed unidentified bugs to creep in; bugs, that I will do my best to slay, once identified!
Thanks!
llvm-svn: 191453
Review: http://llvm-reviews.chandlerc.com/D1546.
I have picked up this patch form Lawrence
(http://llvm-reviews.chandlerc.com/D1063) and did a few changes.
From the original change description (updated as appropriate):
This patch adds a check that ensures that modules only use modules they
have so declared. To this end, it adds a statement on intended module
use to the module.map grammar:
use module-id
A module can then only use headers from other modules if it 'uses' them.
This enforcement is off by default, but may be turned on with the new
option -fmodules-decluse.
When enforcing the module semantics, we also need to consider a source
file part of a module. This is achieved with a compiler option
-fmodule-name=<module-id>.
The compiler at present only applies restrictions to the module directly
being built.
llvm-svn: 191283
1. Fixed constructor of shared clause.
2. Some macros for clauses processing are replaced by private template methods.
3. Additional checks in sema analysis of OpenMP clauses.
llvm-svn: 191265
variable from being the function to being the enclosing namespace scope (in
C++) or the TU (in C). This allows us to fix a selection of related issues
where we would build incorrect redeclaration chains for such declarations, and
fail to notice type mismatches.
Such declarations are put into a new IdentifierNamespace, IDNS_LocalExtern,
which is only found when searching scopes, and not found when searching
DeclContexts. Such a declaration is only made visible in its DeclContext if
there are no non-LocalExtern declarations.
llvm-svn: 191064
LLVM supports applying conversion instructions to vectors of the same number of
elements (fptrunc, fptosi, etc.) but there had been no way for a Clang user to
cause such instructions to be generated when using builtin vector types.
C-style casting on vectors is already defined in terms of bitcasts, and so
cannot be used for these conversions as well (without leading to a very
confusing set of semantics). As a result, this adds a __builtin_convertvector
intrinsic (patterned after the OpenCL __builtin_astype intrinsic). This is
intended to aid the creation of vector intrinsic headers that create generic IR
instead of target-dependent intrinsics (in other words, this is a generic
_mm_cvtepi32_ps). As noted in the documentation, the action of
__builtin_convertvector is defined in terms of the action of a C-style cast on
each vector element.
llvm-svn: 190915
Summary:
When selecting a mangling for an anonymous tag type:
- We should first try it's typedef'd name.
- If that doesn't work, we should mangle in the name of the declarator
that specified it as a declaration specifier.
- If that doesn't work, fall back to a static mangling of
<unnamed-type>.
This should make our anonymous type mangling compatible.
This partially fixes PR16994; we would need to have an implementation of
scope numbering to get it right (a separate issue).
Reviewers: rnk, rsmith, rjmccall, cdavis5x
CC: cfe-commits
Differential Revision: http://llvm-reviews.chandlerc.com/D1540
llvm-svn: 190892
it is an implicit instantiation of a class template specialization), pick the
first-loaded definition to be the canonical definition, and merge all other
definitions into it.
This is still rather incomplete -- we need to extend every form of declaration
that can appear within a CXXRecordDecl to be redeclarable if it came from an
AST file (this includes fields, enumerators, ...).
llvm-svn: 190315
When an AST file is built based on another AST file, it can use a decl from
the fist file, and therefore mark the "isUsed" bit. We need to note this in
the AST file so that the bit is set correctly when the second AST file is
loaded.
This patch introduces the distinction between setIsUsed() and markUsed() so
that we don't call into the ASTMutationListener callback when it wouldn't
be appropriate.
Fixes PR16635.
llvm-svn: 190016
these in eagerly if we're not actually processing a translation unit. The added
laziness here also avoids us loading in parts of a CXXRecordDecl earlier than an
upcoming class template specialization merging patch would like.
Ideally, we should mark the vtable as used when we see a definition for the key
function, rather than having a separate pass over dynamic classes at the end of
the TU. The existing approach is pretty bad for PCH/modules, since it forcibly
loads the declarations of all key functions in all imported modules, whether or
not those key functions are defined.
llvm-svn: 189627
This was only used to ensure that the traversal order was the same as the
insertion order, but that guarantee was already being provided by the use
of a FoldingSetVector.
llvm-svn: 189075
Specifically, the following features are not included in this commit:
- any sort of capturing within generic lambdas
- nested lambdas
- conversion operator for captureless lambdas
- ensuring all visitors are generic lambda aware
As an example of what compiles:
template <class F1, class F2>
struct overload : F1, F2 {
using F1::operator();
using F2::operator();
overload(F1 f1, F2 f2) : F1(f1), F2(f2) { }
};
auto Recursive = [](auto Self, auto h, auto ... rest) {
return 1 + Self(Self, rest...);
};
auto Base = [](auto Self, auto h) {
return 1;
};
overload<decltype(Base), decltype(Recursive)> O(Base, Recursive);
int num_params = O(O, 5, 3, "abc", 3.14, 'a');
Please see attached tests for more examples.
Some implementation notes:
- Add a new Declarator context => LambdaExprParameterContext to
clang::Declarator to allow the use of 'auto' in declaring generic
lambda parameters
- Augment AutoType's constructor (similar to how variadic
template-type-parameters ala TemplateTypeParmDecl are implemented) to
accept an IsParameterPack to encode a generic lambda parameter pack.
- Add various helpers to CXXRecordDecl to facilitate identifying
and querying a closure class
- LambdaScopeInfo (which maintains the current lambda's Sema state)
was augmented to house the current depth of the template being
parsed (id est the Parser calls Sema::RecordParsingTemplateParameterDepth)
so that Sema::ActOnLambdaAutoParameter may use it to create the
appropriate list of corresponding TemplateTypeParmDecl for each
auto parameter identified within the generic lambda (also stored
within the current LambdaScopeInfo). Additionally,
a TemplateParameterList data-member was added to hold the invented
TemplateParameterList AST node which will be much more useful
once we teach TreeTransform how to transform generic lambdas.
- SemaLambda.h was added to hold some common lambda utility
functions (this file is likely to grow ...)
- Teach Sema::ActOnStartOfFunctionDef to check whether it
is being called to instantiate a generic lambda's call
operator, and if so, push an appropriately prepared
LambdaScopeInfo object on the stack.
- Teach Sema::ActOnStartOfLambdaDefinition to set the
return type of a lambda without a trailing return type
to 'auto' in C++1y mode, and teach the return type
deduction machinery in SemaStmt.cpp to process either
C++11 and C++14 lambda's correctly depending on the flag.
- various tests were added - but much more will be needed.
A greatful thanks to all reviewers including Eli Friedman,
James Dennett and the ever illuminating Richard Smith. And
yet I am certain that I have allowed unidentified bugs to creep in;
bugs, that I will do my best to slay, once identified!
Thanks!
llvm-svn: 188977
In addition to storing more useful information in the AST, this
fixes a semantic check in template instantiation which checks whether
the l-paren location is valid.
Fixes PR16903.
llvm-svn: 188495
When a local extern declaration redeclares some other entity, the type of that
entity is merged with the prior type if the prior declaration is visible (in C)
or is declared in the same scope (in C++).
- Make LookupRedeclarationWithLinkage actually work in C++, use it in the right
set of cases, and make it track whether it found a shadowed declaration.
- Track whether we found a declaration in the same scope (for C++) including
across serialization and template instantiation.
llvm-svn: 188307
Summary:
Source-centric tools need access to the location of a C++11
lambda expression's capture-default ('&' or '=') when it's present.
It's possible for them to find it by re-lexing and re-implementing
rules that Clang's parser has already applied, but the cost of storing
the SourceLocation and making it available to them is 32 bits per
LambdaExpr (a small delta, proportionally), and the simplification in
client code is significant.
Reviewers: rsmith
Reviewed By: rsmith
CC: cfe-commits, klimek, revane
Differential Revision: http://llvm-reviews.chandlerc.com/D1192
llvm-svn: 188121
The problem was that an enum without closing semicolon could be associated as a forward enum
in an erroneous declaration, leading to the identifier being associated with the enum decl but
without a declaration actually referencing it.
This resulted in not having it serialized before serializing the identifier that is associated with.
Also prevent the ASTUnit from querying the serialized DeclID for an invalid top-level decl; it may not
have been serialized.
rdar://14539667
llvm-svn: 187914
This field is just IsDefaulted && !IsDeleted; in all places it's used,
a simple check for isDefaulted() is superior anyway, and we were forgetting
to set it in a few cases.
Also eliminate CXXDestructorDecl::IsImplicitlyDefined, for the same reasons.
No intended functionality change.
llvm-svn: 187891
in one module but is only declared as a friend in another module, keep it
visible in the result of the merge.
This is incomplete on two axes:
1) Our handling of local extern declarations is basically broken (we put them
in the wrong decl context, and don't find them in redeclaration lookup, unless
they've previously been declared), and this results in them making friends
visible after a merge.
2) Eventually we'll need to mark that this has happened, and more carefully
check whether a declaration should be visible if it was only visible in some
of the modules in which it was declared. Fortunately it's rare for the
identifier namespace of a declaration to change along its redeclaration chain.
llvm-svn: 187639
The headers in the compiler's own resource include directory are
system headers, which means we don't stat() them eagerly when loading
a module. Use module.map as a proxy for these headers and the compiler
itself. Fixes <rdar://problem/13856838>.
llvm-svn: 186870
No functionality change.
In Sema helper functions:
* renamed isTypeName as HasTypenameKeyword
In UsingDecl:
* renamed get/setUsingLocation to get/setUsingLoc
* renamed is/setTypeName as has/setTypename
llvm-svn: 186816
This is the same way GenericSelectionExpr works, and it's generally a
more consistent approach.
A large part of this patch is devoted to caching the value of the condition
of a ChooseExpr; it's needed to avoid threading an ASTContext into
IgnoreParens().
Fixes <rdar://problem/14438917>.
llvm-svn: 186738
decls. That can reenter deserialization and explode horribly by trying to merge
a declaration that we've not got very far through deserializing yet.
llvm-svn: 186236
numbers as we deserialize class template partial specializations. We can't
assume that the old sequence numbers will work.
The sequence numbers are still deterministic, but are now a lot less
predictable for class template partial specializations in modules/PCH.
llvm-svn: 184811
The goal of this sugar node is to be able to look at an arbitrary
FunctionType and tell if any of the parameters were decayed from an
array or function type. Ultimately this is necessary to implement
Microsoft's C++ name mangling scheme, which mangles decayed arrays
differently from normal pointers.
Reviewers: rsmith
Differential Revision: http://llvm-reviews.chandlerc.com/D1014
llvm-svn: 184763
Introduce CXXStdInitializerListExpr node, representing the implicit
construction of a std::initializer_list<T> object from its underlying array.
The AST representation of such an expression goes from an InitListExpr with a
flag set, to a CXXStdInitializerListExpr containing a MaterializeTemporaryExpr
containing an InitListExpr (possibly wrapped in a CXXBindTemporaryExpr).
This more detailed representation has several advantages, the most important of
which is that the new MaterializeTemporaryExpr allows us to directly model
lifetime extension of the underlying temporary array. Using that, this patch
*drastically* simplifies the IR generation of this construct, provides IR
generation support for nested global initializer_list objects, fixes several
bugs where the destructors for the underlying array would accidentally not get
invoked, and provides constant expression evaluation support for
std::initializer_list objects.
llvm-svn: 183872
A while ago we allowed libclang to build a PCH that had compiler errors; this was to retain the performance
afforded by a PCH even if the user's code is in an intermediate state.
Extend this for the precompiled preamble as well.
rdar://14109828
llvm-svn: 183717
handle temporaries which have been lifetime-extended to static storage duration
within constant expressions. This correctly handles nested lifetime extension
(through reference members of aggregates in aggregate initializers) but
non-constant-expression emission hasn't yet been updated to do the same.
llvm-svn: 183283
In a certain code-path we were not deserializing an anonymous field initializer correctly,
leading to a crash when trying to IRGen it.
This is a simpler version of a patch by Yunzhong Gao!
llvm-svn: 182974
John noticed that the fix for pr15930 (r181981) didn't handle indirect
uses of local types. For example, a pointer to local struct, or a
function that returns it.
One way to implement this would be to recursively look for local
types. This would look a lot like the linkage computation itself for
types.
To avoid code duplication and utilize the existing linkage cache, this
patch just makes the computation of "type with no linkage but
externally visible because it is from an inline function" part of the
linkage computation itself.
llvm-svn: 182711
specialization with modules enabled. Just don't merge them at all for now;
we'll revisit this when support for template merging is added.
In passing, make Decl::dump() a little safer to use with PCH/modules, by making
it not deserialize any additional declarations. From a debugger you can call
decls_begin() or similar first if you want to dump all child decls.
llvm-svn: 182544
a FieldDecl from it, and propagate both into the closure type and the
LambdaExpr.
You can't do much useful with them yet -- you can't use them within the body
of the lambda, because we don't have a representation for "the this of the
lambda, not the this of the enclosing context". We also don't have support or a
representation for a nested capture of an init-capture yet, which was intended
to work despite not being allowed by the current standard wording.
llvm-svn: 181985
This patch renames getLinkage to getLinkageInternal. Only code that
needs to handle UniqueExternalLinkage specially should call this.
Linkage, as defined in the c++ standard, is provided by
getFormalLinkage. It maps UniqueExternalLinkage to ExternalLinkage.
Most places in the compiler actually want isExternallyVisible, which
handles UniqueExternalLinkage as internal.
llvm-svn: 181677
Sometimes people hack on their system headers. In such cases, they'll
need to delete their module cache, but may not know where it is. Add a
note to show them where it is.
llvm-svn: 181638
After r180934 we may initiate module map parsing for modules not related to the module what we are building,
make sure we ignore the header file info of headers from such modules.
First part of rdar://13840148
llvm-svn: 181489
Add serialization for captured statements and captured decls. Also add
a const_capture_iterator to CapturedStmt.
Test contributed by Wei Pan
Differential Revision: http://llvm-reviews.chandlerc.com/D727
llvm-svn: 181048
the actual parser and support arbitrary id-expressions.
We're actually basically set up to do arbitrary expressions here
if we wanted to.
Assembly operands permit things like A::x to be written regardless
of language mode, which forces us to embellish the evaluation
context logic somewhat. The logic here under template instantiation
is incorrect; we need to preserve the fact that an expression was
unevaluated. Of course, template instantiation in general is fishy
here because we have no way of delaying semantic analysis in the
MC parser. It's all just fishy.
I've also fixed the serialization of MS asm statements.
This commit depends on an LLVM commit.
llvm-svn: 180976
are now two distinct canonical 'AutoType's: one is the undeduced 'auto'
placeholder type, and the other is a deduced-but-dependent type. All
deduced-to-a-non-dependent-type cases are still non-canonical.
llvm-svn: 180789
-Make sure that a deserialized external decl gets added to the TU scope.
-When associating an identifier with a set of decls, use the most recent local ones,
if they exist, otherwise associating decls from modules (that came after a local one)
will lead to an incomplete reconstructed re-declaration chain.
rdar://13712705
llvm-svn: 180634
Add a CXXDefaultInitExpr, analogous to CXXDefaultArgExpr, and use it both in
CXXCtorInitializers and in InitListExprs to represent a default initializer.
There's an additional complication here: because the default initializer can
refer to the initialized object via its 'this' pointer, we need to make sure
that 'this' points to the right thing within the evaluation.
llvm-svn: 179958
Typo correction for an unqualified name needs to walk through all of the identifier tables of all modules.
When we have a global index, just walk its identifier table only.
rdar://13425732
llvm-svn: 179730
This is done by extending ObjCMethodList (which is only used by the global method pool) to have 2 extra bits of information.
We will later take advantage of this info in global method pool for the overridden methods calculation.
llvm-svn: 179652
Add CapturedDecl to be the DeclContext for CapturedStmt, and perform semantic
analysis. Currently captures all variables by reference.
TODO: templates
Author: Ben Langmuir <ben.langmuir@intel.com>
Differential Revision: http://llvm-reviews.chandlerc.com/D433
llvm-svn: 179618
don't serialize a lookup map for the translation unit outside C++ mode, so we
can't tell when lookup within the TU needs to look within modules. Only apply
the fix outside C++ mode, and only to the translation unit.
llvm-svn: 178706
http://lab.llvm.org:8011/builders/clang-x86_64-darwin10-gdb went back green
before it processed the reverted 178663, so it could not have been the culprit.
Revert "Revert 178663."
This reverts commit 4f8a3eb2ce5d4ba422483439e20c8cbb4d953a41.
llvm-svn: 178682
Syntactically means the function macro parameter names do not need to use the same
identifiers in order for the definitions to be considered identical.
Syntactic equivalence is a microsoft extension for macro redefinitions and we'll also
use this kind of comparison to check for ambiguous macros coming from modules.
rdar://13562254
llvm-svn: 178671
For variables and functions clang used to store two storage classes. The one
"as written" in the code and a patched one, which, for example, propagates
static to the following decls.
This apparently is from the days clang lacked linkage computation. It is now
redundant and this patch removes it.
llvm-svn: 178663
This option can be useful for end users who want to know why they
ended up with a ton of different variants of the "std" module in their
module cache. This problem should go away over time, as we reduce the
need for module variants, but it will never go away entirely.
llvm-svn: 178148
the system macro uses a not identical definition compared to a macro from the clang headers.
For example (these come from different modules):
\#define LONG_MAX __LONG_MAX__ (clang's limits.h)
\#define LONG_MAX 0x7fffffffffffffffL (system's limits.h)
in which case don't mark them ambiguous to avoid the "ambiguous macro expansion" warning.
llvm-svn: 178109
For each macro directive (define, undefine, visibility) have a separate object that gets chained
to the macro directive history. This has several benefits:
-No need to mutate a MacroDirective when there is a undefine/visibility directive. Stuff like
PPMutationListener become unnecessary.
-No need to keep extra source locations for the undef/visibility locations for the define directive object
(which is the majority of the directives)
-Much easier to hide/unhide a section in the macro directive history.
-Easier to track the effects of the directives across different submodules.
llvm-svn: 178037
-Serialize the macro directives history into its own section
-Get rid of the macro updates section
-When de/serializing an identifier from a module, associate only one macro per
submodule that defined+exported it.
llvm-svn: 177761
The refactoring in r177367 introduced a serious performance bug where
the "lazy" resolution of module file names in the global module index
to actual module file entries in the module manager would perform
repeated negative stats(). The new interaction requires the module
manager to inform the global module index when a module file has been
loaded, eliminating the extraneous stat()s and a bunch of bookkeeping
on both sides.
llvm-svn: 177750
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).
llvm-svn: 177466
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.
llvm-svn: 177367
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.
llvm-svn: 177194
In a module-enabled Cocoa PCH file, we spend a lot of time stat'ing the headers
in order to associate the FileEntries with their modules and support implicit
module import.
Use a more lazy scheme by enhancing HeaderInfoTable to store extra info about
the module that a header belongs to, and associate it with its module only when
there is a request for loading the header info for a particular file.
Part of rdar://13391765
llvm-svn: 176976
This allows resolving top-header filenames of modules to FileEntries when
we need them, not eagerly.
Note that that this breaks ABI for libclang functions
clang_Module_getTopLevelHeader / clang_Module_getNumTopLevelHeaders
but this is fine because they are experimental and not widely used yet.
llvm-svn: 176975
It passes to the visitor, that the caller provides, CXCursor_InclusionDirective cursors for
all the include directives in a particular file.
llvm-svn: 176682
Stat'ing all the headers from the PCH to make sure they are up-to-date takes significant time.
In a particular source file (whose PCH file included Cocoa.h) from total -fsyntax-only time
12% was just stat calls. Change pre-validation to only check non-system headers.
There are some notable disadvantages:
-If a system header, that is not include-guarded, changes after the PCH was created, we will not
find it in the header info table and we will #import it, effectively #importing it twice, thus
we will emit some error due to a multiple definition and after that the "header was modified" error will likely
be emitted, for example something like:
NSDictionary.h:12:1: error: duplicate interface definition for class 'NSDictionary'
@interface NSDictionary : NSObject <NSCopying, NSMutableCopying, NSSecureCoding, NSFastEnumeration>
^
NSDictionary.h:12:12: note: previous definition is here
@interface NSDictionary : NSObject <NSCopying, NSMutableCopying, NSSecureCoding, NSFastEnumeration>
^
fatal error: file 'NSDictionary.h' has been modified since the precompiled header was built
Though we get the "header was modified" error, this is a bit confusing.
-Theoretically it is possible that such a system header will cause no errors but it will just cause an
unfortunate semantic change, though I find this rather unlikely.
The advantages:
-Reduces compilation time when using a huge PCH like the Cocoa ones
-System headers change very infrequent and when they do, users/build systems should be able to know that
re-building from scratch is needed.
Addresses rdar://13056262
llvm-svn: 176567
Previously the hash would be the filename portion of the path, which could be
different for a filename with different case or a symbolic link with a different
name completely.
This did not actually create any issue so far because by validating all headers
in the PCH we created uniqued FileEntries based on inodes, so an #include of
a symbolic link (refering to a file from the PCH) would end up with a FileEntry
with filename same as the one recorded in the PCH.
llvm-svn: 176566
llvm::sys::fs::equivalent() does 2 stat calls every time it's called. Use FileManager::getFile() to take advantage
of the stat caching that FileManager is providing.
llvm-svn: 176450
Previously we would return null for an out-of-date file. This inhibited ASTReader::ReadSLocEntry
from creating a FileID to recover gracefully in such a case.
llvm-svn: 176332
its index in the preprocessed entities vector.
This is because the order of the entities in the vector can change in some (uncommon) cases.
llvm-svn: 175907
Introduce a new AST Decl node "EmptyDecl" to model empty-declaration. Have attributes from attribute-declaration appertain
to the EmptyDecl node by creating the AST representations of these attributes and attach them to the EmptyDecl node so these
attributes can be sema checked just as attributes attached to "normal" declarations.
llvm-svn: 175900
Add an ability to specify custom documentation block comment commands via a new
class CommentOptions. The intention is that this class will hold future
customizations for comment parsing, including defining documentation comments
with specific numbers of parameters, etc.
CommentOptions instance is a member of LangOptions.
CommentOptions is controlled by a new command-line parameter
-fcomment-block-commands=Foo,Bar,Baz.
llvm-svn: 175892
for the data specific to a macro definition (e.g. what the tokens are), and
MacroDirective class which encapsulates the changes to the "macro namespace"
(e.g. the location where the macro name became active, the location where it was undefined, etc.)
(A MacroDirective always points to a MacroInfo object.)
Usually a macro definition (MacroInfo) is where a macro name becomes active (MacroDirective) but
splitting the concepts allows us to better model the effect of modules to the macro namespace
(also as a bonus it allows better modeling of push_macro/pop_macro #pragmas).
Modules can have their own macro history, separate from the local (current translation unit)
macro history; MacroDirectives will be used to model the macro history (changes to macro namespace).
For example, if "@import A;" imports macro FOO, there will be a new local MacroDirective created
to indicate that "FOO" became active at the import location. Module "A" itself will contain another
MacroDirective in its macro history (at the point of the definition of FOO) and both MacroDirectives
will point to the same MacroInfo object.
Introducing the separation of macro concepts is the first part towards better modeling of module macros.
llvm-svn: 175585
This commit introduces a set of related changes to ensure that the
declaration that shows up in the identifier chain after deserializing
declarations with a given identifier is, in fact, the most recent
declaration. The primary change involves waiting until after we
deserialize and wire up redeclaration chains before updating the
identifier chains. There is a minor optimization in here to avoid
recursively deserializing names as part of looking to see whether
top-level declarations for a given name exist.
A related change that became suddenly more urgent is to property
record a merged declaration when an entity first declared in the
current translation unit is later deserialized from a module (that had
not been loaded at the time of the original declaration). Since we key
off the canonical declaration (which is parsed, not from an AST file)
for emitted redeclarations, we simply record this as a merged
declaration during AST writing and let the readers merge them.
Re-fixes <rdar://problem/13189985>, presumably for good this time.
llvm-svn: 175447
until recursive loading is finished.
Otherwise we may end up with a template trying to deserialize a template
parameter that is in the process of getting loaded.
rdar://13135282
llvm-svn: 175329
the linkage of functions and variables while merging declarations from modules,
and we don't necessarily have enough of the rest of the AST loaded at that
point to allow us to compute linkage, so serialize it instead.
llvm-svn: 174943
These two related tweaks to keep the information associated with a
given identifier correct when the identifier has been given some
top-level information (say, a top-level declaration) and more
information is then loaded from a module. The first ensures that an
identifier that was "interesting" before being loaded from an AST is
considered to be different from its on-disk counterpart. Otherwise, we
lose such changes when writing the current translation unit as a
module.
Second, teach the code that injects AST-loaded names into the
identifier chain for name lookup to keep the most recent declaration,
so that we don't end up confusing our declaration chains by having a
different declaration in there.
llvm-svn: 174895
name lookup has been performed in that context (this probably only happens in
C++).
1) Whenever we add names to a context, set a flag on it, and if we perform
lookup and discover that the context has had a lookup table built but has the
flag set, update all entries in the lookup table with additional names from
the external source.
2) When marking a DeclContext as having external visible decls, mark the
context in which lookup is performed, not the one we are adding. These won't
be the same if we're adding another copy of a pre-existing namespace.
llvm-svn: 174577
if it found any decls, rather than returning a list of found decls. This
removes a returning-ArrayRef-to-deleted-storage bug from
MultiplexExternalSemaSource (in code not exercised by any of the clang
binaries), reduces the work required in the found-no-decls case with PCH, and
importantly removes the need for DeclContext::lookup to be reentrant.
No functionality change intended!
llvm-svn: 174576
This can happen when one abuses precompiled headers by passing more -D
options when using a precompiled hedaer than when it was built. This
is intentionally permitted by precompiled headers (and is exploited by
some build environments), but causes problems for modules.
First part of <rdar://problem/13165109>, detecting when something when
horribly wrong.
llvm-svn: 174554
Essentially, a module file on disk could change size between the time
we stat() it and the time we open it, and we need to be robust against
such a problem.
llvm-svn: 174529
Different modules may have different views of the various "special"
types in the AST, such as the redefinition type for "id". Merge those
types rather than only considering the redefinition types for the
first AST file loaded.
llvm-svn: 174234
- The only group where it makes sense for the "ExternC" bit is System, so this
simplifies having to have the extra isCXXAware (or ImplicitExternC, depending
on what code you talk to) bit caried around.
llvm-svn: 173859
ModuleManager::visit() by keeping a free list of the two data
structures used to store state (a preallocated stack and a visitation
number vector). Improves -fsyntax-only performance for my modules test
case by 2.8%. Modules has pulled ahead by almost 10% with the global
module index.
llvm-svn: 173692
Title: [PR9027] volatile struct bug: member is not loaded at -O;
This is caused by last flag passed to @llvm.memcpy being false,
not honoring that aggregate has at least one 'volatile' data member
(even though aggregate itself has not been qualified as 'volatile'.
As a result, optimization optimizes away the memcpy altogether.
Patch review by John MaCall (I still need to fix up a test though).
llvm-svn: 173535
index, optimizing the operation that skips lookup in modules where we
know the identifier will not be found. This makes the global module
index optimization actually useful, providing an 8.5% speedup over
modules without the global module index for -fsyntax-only.
llvm-svn: 173529
never key functions. We did not implement that rule for the
iOS ABI, which was driven by what was implemented in gcc-4.2.
However, implement it now for other ARM-based platforms.
llvm-svn: 173515
and limiting ourselves to two memory allocations. 10% speedup in
-fsyntax-only time for modules.
With this change, we can actually see some performance different from
the global module index, but it's still about 1%.
llvm-svn: 173512
AST reader.
The global module index tracks all of the identifiers known to a set
of module files. Lookup of those identifiers looks first in the global
module index, which returns the set of module files in which that
identifier can be found. The AST reader only needs to look into those
module files and any module files not known to the global index (e.g.,
because they were (re)built after the global index), reducing the
number of on-disk hash tables to visit. For an example source I'm
looking at, we go from 237844 total identifier lookups into on-disk
hash tables down to 126817.
Unfortunately, this does not translate into a performance advantage.
At best, it's a wash once the global module index has been built, but
that's ignore the cost of building the global module index (which
is itself fairly large). Profiles show that the global module index
code is far less efficient than it should be; optimizing it might give
enough of an advantage to justify its continued inclusion.
llvm-svn: 173405
The global module index is a "global" index for all of the module
files within a particular subdirectory in the module cache, which
keeps track of all of the "interesting" identifiers and selectors
known in each of the module files. One can perform a fast lookup in
the index to determine which module files will have more information
about entities with a particular name/selector. This information can
help eliminate redundant lookups into module files (a serious
performance problem) and help with creating auto-import/auto-include
Fix-Its.
The global module index is created or updated at the end of a
translation unit that has triggered a (re)build of a module by
scraping all of the .pcm files out of the module cache subdirectory,
so it catches everything. As with module rebuilds, we use the file
system's atomicity to synchronize.
llvm-svn: 173301
identifiers into two parts: the part that involves dealing with the
key (which can be re-used) and the ASTReader-specific part that
creates the IdentifierInfos. While I'm at it, StringRef'ify this code,
which was using pair<const char*, unsigned>. No functionality change.
llvm-svn: 173283
This change also makes the serialisation store the required semantics,
fixing an issue where PPC128 was always assumed when re-reading a
128-bit value.
llvm-svn: 173139
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.
llvm-svn: 173066
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.
llvm-svn: 173056
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.
llvm-svn: 173050
Richard Smith