Before:
aaaaaaaaaaaaaaaaaaaaaaaa(aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa()
.aaaaaaaaaaaaaaaaa());
After:
aaaaaaaaaaaaaaaaaaaaaaaaa(
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa().aaaaaaaaaaaaaaaaa());
Not sure which of the formattings above is better, but we should not pick
one by accident.
llvm-svn: 175165
This gives a clearer separation of the context, e.g. in GMOCK
statements.
Before:
EXPECT_CALL(SomeObject,
SomeFunction(Parameter)).WillRepeatedly(Return(SomeValue));
After:
EXPECT_CALL(SomeObject, SomeFunction(Parameter))
.WillRepeatedly(Return(SomeValue));
Minor format cleanups.
llvm-svn: 175162
So far, clang-format has always assumed the whitespace belonging to the
subsequent token. This has the negative side-effect that when
clang-format formats a line, it does not remove its trailing whitespace,
as it belongs to the next token.
Thus, this patch fixes most of llvm.org/PR15062.
We are not zapping a file's trailing whitespace so far, as this does not
belong to any token we see during formatting. We need to fix this in a
subsequent patch.
llvm-svn: 175152
The formatter can now format:
void aaaaaaaaaaaaaaaaaa(int level,
double *min_x,
double *max_x,
double *min_y,
double *max_y,
double *min_z,
double *max_z, ) {
}
Although this is invalid code, it frequently happens during development and
clang-format should be nicer :-).
llvm-svn: 175151
This fixes llvm.org/PR15179.
Before:
class ColorChooserMac : public content::ColorChooser,
public content::WebContentsObserver {
};
After:
class ColorChooserMac : public content::ColorChooser,
public content::WebContentsObserver {
};
llvm-svn: 175147
The code generation stuff is going to set attributes on the functions it
generates. To do that it needs the target options. Pass them through.
llvm-svn: 175141
I added hasCLanguageLinkage while fixing some language linkage bugs some
time ago so that I wouldn't have to check all users of isExternC. It turned
out to be a much longer detour than expected, but this patch finally
merges the two again. The isExternC function now implements just the
standard notion of having C language linkage.
llvm-svn: 175119
some cases where functions with no language linkage were being treated as having
C language linkage. In particular, don't warn in
extern "C" {
static NonPod foo();
}
Since getLanguageLinkage checks the language linkage, the linkage computation
cannot use the language linkage. Break the loop by checking just the context
in the linkage computation.
llvm-svn: 175117
instantiation in order to permit devirtualization later in codegen, skip over
pure functions since those can't be devirtualization targets.
llvm-svn: 175116
in the course of property synthesis deterministic (ordered
by their type size), instead of having hashtable order
(as it is currently). // rdar://13192366
llvm-svn: 175100
This has so far been disabled for Google style, but should be done
before breaking at nested name specifiers or in template parameters.
Before (in Google style):
template <typename T>
aaaaaaaa::aaaaa::aaaaaa<T, aaaaaaaaaaaaaaaaaaaaaaaaa> aaaaaaaaaaaaaaaaaaaaaaaa<
T>::aaaaaaa() {}
After:
template <typename T>
aaaaaaaa::aaaaa::aaaaaa<T, aaaaaaaaaaaaaaaaaaaaaaaaa>
aaaaaaaaaaaaaaaaaaaaaaaa<T>::aaaaaaa() {}
llvm-svn: 175074
- clear ownership: the SpecificBumpPtrAllocator owns all StateNodes
- this allows us to simplify the memoization data structure into a
std::set (FIXME: figure out whether we want to use a hash based
data structure).
- introduces StateNode as recursive data structure, instead of using
Edge and the Seen-map combined to drill through the graph
- using a count to stabilize the penalty instead of relying on the
container
- pulled out a method to forward-apply states in the end
This leads to a ~40% runtime decrease on Nico's benchmark.
Main FiXME is that the parameter lists of some function get too long.
I'd vote for either pulling the Queue etc into the Formatter proper,
or creating an inner class just for the search algorithm.
llvm-svn: 175051
Previously, we were handling only simple integer constants for globals and
the smattering of implicitly-valued expressions handled by Environment for
default arguments. Now, we can use any integer constant expression that
Clang can evaluate, in addition to everything we handled before.
PR15094 / <rdar://problem/12830437>
llvm-svn: 175026
Still the formatting can be improved, but at least we don't assert any
more. This happened when trying to format lib/Sema/SemaType.cpp.
llvm-svn: 175003
MSVC accepts this:
class A {
A::A();
};
Clang accepts regular member functions with extra qualification as an MS
extension, but not constructors. This changes the parser to defer rejecting
qualified constructors so that the same Sema logic can apply to constructors as
regular member functions. This also improves the error message when MS
extensions are disabled (in my opinion). Before it was:
/Users/jason/Desktop/test.cpp:2:8: error: expected member name or ';' after declaration specifiers
A::A();
~~~~ ^
1 error generated.
After:
/Users/jason/Desktop/test.cpp:2:6: error: extra qualification on member 'A'
A::A();
~~~^
1 error generated.
Patch by Jason Haslam.
llvm-svn: 174980
This redoes how '*' and '&' are classified as pointer / reference markers when
followed by ')', '>', or ','.
Previously, determineStarAmpUsage() marked a single '*' and '&' followed by
')', '>', or ',' as pointer or reference marker. Now, all '*'s and '&'s
preceding ')', '>', or ',' are marked as pointer / reference markers. Fixes
PR14884.
Since only the last '*' in 'int ***' was marked as pointer before (the rest
were unary operators, which don't reach spaceRequiredBetween()),
spaceRequiredBetween() now had to be thought about handing multiple '*'s in
sequence.
Before:
return sizeof(int * *);
Type **A = static_cast<Type * *>(P);
Now:
return sizeof(int**);
Type **A = static_cast<Type **>(P);
While here, also make all methods of AnnotatingParser except parseLine()
private.
Review URL: http://llvm-reviews.chandlerc.com/D384
llvm-svn: 174975
MarkMemberReferenced instead of marking functions referenced directly. An audit
of callers to MarkFunctionReferenced and DiagnoseUseOfDecl also caused a few
other changes:
* don't mark functions odr-used when considering them for an initialization
sequence. Do mark them referenced though.
* the function nominated by the cleanup attribute should be diagnosed.
* operator new/delete should be diagnosed when building a 'new' expression.
llvm-svn: 174951
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
move-constructors and move-assignment operators, use memcpy to copy adjacent
POD members.
Previously, classes with one or more Non-POD members would fall back on
element-wise copies for all members, including POD members. This often
generated a lot of IR. Without padding metadata, it wasn't often possible
for the LLVM optimizers to turn the element-wise copies into a memcpy.
This code hasn't yet received any serious tuning. I didn't see any serious
regressions on a self-hosted clang build, or any of the nightly tests, but
I think it's important to get this out in the wild to get more testing.
Insights, feedback and comments welcome.
Many thanks to David Blaikie, Richard Smith, and especially John McCall for
their help and feedback on this work.
llvm-svn: 174919
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
Before:
for (id foo in[self getStuffFor : bla]) {
}
Now:
for (id foo in [self getStuffFor:bla]) {
}
"in" is treated as loop keyword if the line starts with "for", and as a
regular identifier else. To check for "in", its IdentifierInfo is handed
through a few layers.
llvm-svn: 174889
In google style, trailing comments are separated by two spaces. This
patch fixes the counting of these spaces and prevents clang-format from
creating a line with 81 columns.
llvm-svn: 174879
The more general code for formatting ObjC method exprs does this and more,
it's no longer necessary to special-case this. No behavior change.
llvm-svn: 174843
Before:
@{
foo:
bar
}
;
Now:
@{ foo : bar };
parseBracedList() already does the right thing from an UnwrappedLineParser
perspective, so check for "@{" in all loops that process constructs that can
contain expressions and call parseBracedList() if found.
llvm-svn: 174840
Apple's kernel engineers have been expecting this behavior even though
we've never implemented it before, as far as I can tell. In recent months,
clang has gotten better at using vector instructions to optimize memcpy-like
operations, and that has exposed problems when vector/floating-point
instructions are used in kexts that don't support that. This behavior also
matches what Apple's GCC did for PowerPC targets.
llvm-svn: 174838
For x86 targets, we've been using the -msoft-float option to control passing
the no-implicit-float option to cc1. Since the -mno-implicit-float option is
now accepted by the driver, this just makes it work for x86 the same as it
does for ARM targets.
llvm-svn: 174836
the "nonatomic" attribute in property redeclaration
in class extension. Also, improved on diagnostics in
this area while at it. // rdar://13156292
llvm-svn: 174821
visible.
The basic problem here is that a given translation unit can use
forward declarations to form pointers to a given type, say,
class X;
X *x;
and then import a module that includes a definition of X:
import XDef;
We will then fail when attempting to access a member of X, e.g.,
x->method()
because the AST reader did not know to look for a default of a class
named X within the new module.
This implementation is a bit of a C-centric hack, because the only
definitions that can have this property are enums, structs, unions,
Objective-C classes, and Objective-C protocols, and all of those are
either visible at the top-level or can't be defined later. Hence, we
can use the out-of-date-ness of the name and the identifier-update
mechanism to force the update.
In C++, we will not be so lucky, and will need a more advanced
solution, because the definitions could be in namespaces defined in
two different modules, e.g.,
// module 1
namespace N { struct X; }
// module 2
namespace N { struct X { /* ... */ }; }
One possible implementation here is for C++ to extend the information
associated with each identifier table to include the declaration IDs
of any definitions associated with that name, regardless of
context. We would have to eagerly load those definitions.
llvm-svn: 174794
Add warnings under -Wc++11-compat, -Wc++98-compat, and -Wc99-compat when a
particular UCN is incompatible with a different standard, and -Wunicode when
a UCN refers to a surrogate character in C++03.
llvm-svn: 174788
The missing definition check should be in the same category as the
missing ivar validation - in this case, the intent is to invalidate in
the given class, as described in the declaration, but the implementation
does not perform the invalidation. Whereas the MissingInvalidationMethod
checker checks the cases where the method intention is not to
invalidate. The second checker has potential to have a much higher false
positive rate.
llvm-svn: 174787
The new annotation allows having methods that only partially invalidate
IVars and might not be called from the invalidation methods directly
(instead, are guaranteed to be called before the invalidation occurs).
The checker is going to trust the programmer to call the partial
invalidation method before the invalidator.This is common in cases when
partial object tear down happens before the death of the object.
llvm-svn: 174779
"auto-synthesized may not work correctly with 'nib' loader"
when 'readonly' property is redeclared 'readwrite' in class
extension. // rdar://13123861
llvm-svn: 174775
This may not always be valid, but we were previously just
emitting them raw.
While here, s/isprint/isPrintable/ (using the new CharInfo).
llvm-svn: 174766
Rewriting the same predicates over and over again is bad for code size and
code maintainence. Using the functions in <ctype.h> is generally unsafe
unless they are specified to be locale-independent (i.e. only isdigit and
isxdigit).
The next commit will try to clean up uses of <ctype.h> functions within Clang.
llvm-svn: 174765
With this patch, the formatter introduces 'fake' parenthesis according
to the operator precedence of binary operators.
Before:
return aaaa & AAAAAAAAAAAAAAAAAAAAAAAAAAAAA || bbbb &
BBBBBBBBBBBBBBBBBBBBBBBBBBBBB || cccc & CCCCCCCCCCCCCCCCCCCCCCCCCC ||
dddd & DDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDD;
f(aaaaaaaaaaaaaaaaaaaa && aaaaaaaaaaaaaaaaaaaa, aaaaaaaaaaaaaaaaaaaa &&
aaaaaaaaaaaaaaaaaaaa, aaaaaaaaaaaaaaaaaaaa && aaaaaaaaaaaaaaaaaaaa);
After:
return aaaa & AAAAAAAAAAAAAAAAAAAAAAAAAAAAA ||
bbbb & BBBBBBBBBBBBBBBBBBBBBBBBBBBBB ||
cccc & CCCCCCCCCCCCCCCCCCCCCCCCCC ||
dddd & DDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDD;
f(aaaaaaaaaaaaaaaaaaaa && aaaaaaaaaaaaaaaaaaaa,
aaaaaaaaaaaaaaaaaaaa && aaaaaaaaaaaaaaaaaaaa,
aaaaaaaaaaaaaaaaaaaa && aaaaaaaaaaaaaaaaaaaa);
Future improvements:
- Get rid of some of the hacky ways to nicely format certain constructs.
- Merge this parser and the AnnotatingParser as we now have several parsers
that analyze (), [], etc.
llvm-svn: 174714
overloads of a name by claiming that there are no lookup results for that name
in modules while loading the names from the module. Lookups in deserialization
really don't want to find names which they themselves are in the process of
introducing. This also has the pleasant side-effect of automatically caching
PCH lookups which found no names.
The runtime here is still quadratic in the number of overloads, but the
constant is lower.
llvm-svn: 174685
The malloc checker will now catch the case when a previously malloc'ed
region is freed, but the pointer passed to free does not point to the
start of the allocated memory. For example:
int *p1 = malloc(sizeof(int));
p1++;
free(p1); // warn
From the "memory.LeakPtrValChanged enhancement to unix.Malloc" entry
in the list of potential checkers.
A patch by Branden Archer!
llvm-svn: 174678
The checkPointerEscape callback previously did not specify how a
pointer escaped. This change includes an enum which describes the
different ways a pointer may escape. This enum is passed to the
checkPointerEscape callback when a pointer escapes. If the escape
is due to a function call, the call is passed. This changes
previous behavior where the call is passed as NULL if the escape
was due to indirectly invalidating the region the pointer referenced.
A patch by Branden Archer!
llvm-svn: 174677
This patch makes sure that we do not reinitialize static globals when
the function is called more than once along a path. The motivation is
code with initialization patterns that rely on 2 static variables, where
one of them has an initializer while the other does not. Currently, we
reset the static variables with initializers on every visit to the
function along a path.
llvm-svn: 174676
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
Summary:
-Wimplicit-fallthrough: fixed two cases where "fallthrough annotation in unreachable code" was issued incorrectly:
1. In actual unreachable code, but not immediately on a fall-through execution
path "fallthrough annotation does not directly precede switch label" is better;
2. After default: in a switch with covered enum cases. Actually, these shouldn't
be treated as unreachable code for our purpose.
Reviewers: rsmith
Reviewed By: rsmith
CC: cfe-commits
Differential Revision: http://llvm-reviews.chandlerc.com/D374
llvm-svn: 174575
The use of this flag enables a modules optimization where a given set
of macros can be labeled as "ignored" by the modules
system. Definitions of those macros will be completely ignored when
building the module hash and will be stripped when actually building
modules. The overall effect is that this flag can be used to
drastically reduce the number of
Eventually, we'll want modules to tell us what set of macros they
respond to (the "configuration macros"), and anything not in that set
will be excluded. However, that requires a lot of per-module
information that must be accurate, whereas this option can be used
more readily.
Fixes the rest of <rdar://problem/13165109>.
llvm-svn: 174560
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
We now correctly format:
// Written as a macro, it is reformatted from:
#define foo(a) \
do { \
/* Initialize num to zero. */ \
int num = 10; \
/* This line ensures a is never zero. */ \
int i = a == 0 ? 1 : a; \
i = num / i; /* This division is OK. */ \
return i; \
} while (false)
llvm-svn: 174517
With this patch, clang-format can analyze the input file for two
properties:
1. Is "int *a" or "int* a" more common.
2. Are non-C++03 constructs used, e.g. A<A<A>>.
With Google-style, clang-format will now use the more common style for
(1) and format C++03 compatible, unless it finds C++11 constructs in the
input.
llvm-svn: 174504
Properly handle annotation contexts while calculating extra information
for each token. This enable nested ObjC calls and thus solves (most of)
llvm.org/PR15164. E.g., we can now format:
[contentsContainer replaceSubview:[subviews objectAtIndex:0]
with:contentsNativeView];
Also fix a problem with the formatting of types in casts as this was
trivial now.
llvm-svn: 174498
1. let determineStarAmp() check of unary operators before checking for
"is next '['". That check was added in r173150, and the test from that
revision passes either way.
2. change determineStarAmp() to categorize '*' and '&' after '=' as unary
operator.
3. don't let parseSquare() overwrite the type of a '*' or '&' before the start
of an objc message expression if has the role of unary operator.
llvm-svn: 174489
This is a "quick fix".
The underlining issue is that when a const pointer to a struct is passed
into a function, we do not invalidate the pointer fields. This results
in false positives that are common in C++ (since copy constructors are
prevalent). (Silences two llvm false positives.)
llvm-svn: 174468
This is to ensure that GlobalOpt in LLVM does not attempt to look through a
selector reference to a method var name at compile time.
I also added a test/updated old tests that need to recognize the new keyword.
rdar://12580965.
llvm-svn: 174461
This is a more natural order of evaluation, and it is very important
for visualization in the static analyzer. Within Xcode, the arrows
will not jump from right to left, which looks very visually jarring.
It also provides a more natural location for dataflow-based diagnostics.
Along the way, we found a case in the analyzer diagnostics where we
needed to indicate that a variable was "captured" by a block.
-fsyntax-only timings on sqlite3.c show no visible performance change,
although this is just one test case.
Fixes <rdar://problem/13016513>
llvm-svn: 174447
since only one of them is allowed in command-line, process them separately.
Otherwise, if more than one is specified in the command-line, one is processed normally
and the others are going to be treated and included as header files.
Related to radar://13140508
llvm-svn: 174385
We can now format stuff like:
- (void)doSomethingWith:(GTMFoo *)theFoo
rect:(NSRect)theRect
interval:(float)theInterval {
[myObject doFooWith:arg1 //
name:arg2
error:arg3];
}
This seems to fix everything mentioned in llvm.org/PR14939.
llvm-svn: 174364
In preprocessor definitions, we would not parse all the tokens and thus
not annotate them anymore. This led to a wrong formatting of comments
in google style:
#endif // HEADER_GUARD -- requires two spaces
llvm-svn: 174361
If there are string literals on either side of a '<<', chances are
high that they represent logically separate concepts. Otherwise,
the author could just have just a single literal (possible split
over multiple lines).
So, we can now nicely format things like:
cout << "somepacket = {\n"
<< " val a = " << ValueA << "\n"
<< " val b = " << ValueB << "\n"
<< "}";
llvm-svn: 174310
This combines several changes:
* Calculation token type (e.g. for * and &) in the AnnotatingParser.
* Calculate the scope binding strength in the AnnotatingParser.
* Let <> and [] scopes bind stronger than () and {} scopes.
* Add minimal debugging output.
llvm-svn: 174307
This is an improvement of r173630, that handles the following case:
struct VirualDestrClass
{
VirualDestrClass(int arg);
virtual ~VirualDestrClass();
};
struct ConstrWithCleanupsClass
{
ConstrWithCleanupsClass(const VirualDestrClass& cplx = VirualDestrClass(42));
};
ConstrWithCleanupsClass cwcNoArg;
That was printed as:
ConstrWithCleanupsClass cwcNoArg();
llvm-svn: 174296
This can yield dramatic speedups of dynamic_cast for simple inheritance trees,
at least with libsupc++. Neither libcxxabi nor libcxxrt make use of this
hint currently, it was never implemented because clang didn't support it.
There was some concern about the number of class hierarchy walks this change
introduces. If it turns out to be an issue we can add caching either at the cast
pair level or even deeper, but we also do a lot of walks in Sema so this
codepath is probably fairly optimized already.
llvm-svn: 174293
...again. The problem has not been fixed and our internal buildbot is still
getting hangs.
This reverts r174212, originally applied in r173951, then reverted in r174069.
Will not re-apply until the entire project analyzes successfully on my
local machine.
llvm-svn: 174265
designator" diagnostic with more correct and more human-friendly "cannot take
address of rvalue of type 'T'".
For the case of & &T::f, provide a custom diagnostic, rather than unhelpfully
saying "cannot take address of rvalue of type '<overloaded function type>'".
For the case of &array_temporary, treat it just like a class temporary
(including allowing it as an extension); the existing diagnostic wording
for the class temporary case works fine.
llvm-svn: 174262
Inlining these functions is essential for correctness. We often have
cases where we do not inline calls. For example, the shallow mode and
when reanalyzing previously inlined ObjC methods as top level.
llvm-svn: 174245
says, but that's a defect (to be filed). "Cls::purevfn()" is still an odr use.
Also fixes a bug that caused us to not mark the function referenced just
because we didn't want to mark it odr used.
llvm-svn: 174242
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
This change introduces a 'kind' attribute for the <Para> tag, that captures the
kind of the parent block command.
For example:
\todo Meow.
used to be just <Para>Meow.</Para>, but now it is
<Para kind="todo">Meow.</Para>
llvm-svn: 174216
This allows us to keep from chaining LazyCompoundVals in cases like this:
CGRect r = CGRectMake(0, 0, 640, 480);
CGRect r2 = r;
CGRect r3 = r2;
Previously we only made this optimization if the struct did not begin with
an aggregate member, to make sure that we weren't picking up an LCV for
the first field of the struct. But since LazyCompoundVals are typed, we can
make that inference directly by comparing types.
This is a pure optimization; the test changes are to guard against possible
future regressions.
llvm-svn: 174211
r173593 made us a little too eager to associate all code at the end of a
function with the user-written 'return' line. This caused problems with
breakpoints as they'd be set in exception handling code preceeding the
actual non-exception return handling code, leading to the breakpoint never
being hit in non-exceptional execution.
This change restores the pre-r173593 exception handling line information where
the cleanup code is associated with the '}' not the return line.
llvm-svn: 174206
Prior to the patch, Clang does not properly promote types when a complex
integer operand is combined with an integer via a binary operator, or when
one is assigned to the other in either order. This patch detects when
promotion is needed (and permissible) and generates the necessary code.
The test assmes no target has the same size operands for "char" and
"long long," and that no target performs arithmetic on char operands without
extending them to a larger format first. If there are any targets for
which this is not the case, they should be XFAILed.
llvm-svn: 174181
First, this implements a match() method on MatchFinder; this allows us
to get rid of the findAll implementation, as findAll is really a special
case of recursive matchers on match.
Instead of findAll, provide a convenience function match() that lets
users iterate easily over the results instead of needing to implement
callbacks.
llvm-svn: 174172
In order to end up with good solutions, clang-format needs to try
"all" combinations of line breaks, evaluate them and select the
best one. Before, we have done this using a DFS with memoization
and cut-off conditions. However, this approach is very limited
as shown by the huge static initializer in the attachment of
llvm.org/PR14959.
Instead, this new implementation uses a variant of Dijkstra's
algorithm to do a prioritized BFS over the solution space.
Some numbers:
lib/Format/TokenAnnotator.cpp: 1.5s -> 0.15s
Attachment of PR14959: 10min+ (didn't finish) -> 10s
No functional changes intended.
llvm-svn: 174166