This patch completely defeated the "passing in a prestat'd size
to MemoryBuffer" optimization, leading to an extra fstat call for
every buffer opened, in order to find out if the datestamp and size
of the file on disk matches what is in the stat cache.
I fully admit that I don't completely understand what is going on here:
why punish code when a stat cache isn't in use? what is the point of a
stat cache if you have to turn around and stat stuff to validate it?
To resolve both these issues, just drop the modtime check and check the
file size, which is the important thing anyway. This should also resolve
PR6812, because presumably windows is stable when it comes to file sizes.
If the modtime is actually important, we should get it and keep it on the
first stat.
This eliminates 833 fstat syscalls when processing Cocoa.h, speeding up
system time on -Eonly Cocoa.h from 0.041 to 0.038s.
Chandler Carruth [Mon, 22 Nov 2010 08:06:31 +0000 (08:06 +0000)]
Undo part of my previous commit to mm_malloc.h, going back to the use of
stdlib.h. There were numerous problems with forward declaring 'malloc' and
'free', but the most important is that these are reserved by POSIX and may be
implemented via a function-like macro.
As suggested by Dale Johannesen, I'm instead guarding the only include of this
in our builtin headers with __STDC_HOSTED__, and I've removed the include of
the header from the test suite. I'll discuss with folks whether we want to have
a hosted section of the test suite or not, and add it (and perhaps other tests)
back there if that's the direction.
Francois Pichet [Sun, 21 Nov 2010 06:08:52 +0000 (06:08 +0000)]
Major anonymous union/struct redesign.
A new AST node is introduced:
def IndirectField : DDecl<Value>;
IndirectFields are injected into the anonymous's parent scope and chain back to
the original field. Name lookup for anonymous entities now result in an
IndirectFieldDecl instead of a FieldDecl.
There is no functionality change, the code generated should be the same.
Revert r119838 "Don't warn for empty 'if' body if there is a macro that expands to nothing"
and use a better and more general approach, where NullStmt has a flag to indicate whether it was preceded by an empty macro.
Several PPCallbacks take an SourceLocation + IdentifierInfo, rather
than a Token that holds the same information all in one easy-to-use
package. There's no technical reason to prefer the former -- the
information comes from a Token originally -- and it's clumsier to use,
so I've changed the code to use tokens everywhere.
Douglas Gregor [Fri, 19 Nov 2010 17:10:50 +0000 (17:10 +0000)]
When parsing something that looks like an ill-formed
protocol-qualifier list without a leading type (e.g., <#blah#>), don't
complain about it being an archaic protocol-qualifier list unless it
actually parses as one.
Bob Wilson [Thu, 18 Nov 2010 21:51:10 +0000 (21:51 +0000)]
Use a signed vector type for the shift amount operand of unsigned shifts.
Neon shifts allow a mix of positive and negative shift amounts to shift
left or right, respectively, so the shift amount vector should always be
signed. PR8482 (Radar 8603521).
Remove the hack where, to get the return status, we had special case for VerifyDiagnosticsClient
and just check the number of errors from the DiagnosticClient.
-Move the stuff of Diagnostic related to creating/querying diagnostic IDs into a new DiagnosticIDs class.
-DiagnosticIDs can be shared among multiple Diagnostics for multiple translation units.
-The rest of the state in Diagnostic object is considered related and tied to one translation unit.
-Have Diagnostic point to the SourceManager that is related with. Diagnostic can now accept just a
SourceLocation instead of a FullSourceLoc.
-Reflect the changes to various interfaces.
In some situations, TemplateArgumentLoc wasn't setting TypeSourceLoc (see
http://llvm.org/bugs/show_bug.cgi?id=8558). This patch fixes it. Thanks to
rjmccall for all the coaching!
John McCall [Thu, 18 Nov 2010 06:31:45 +0000 (06:31 +0000)]
Calculate the value kind of an expression when it's created and
store it on the expression node. Also store an "object kind",
which distinguishes ordinary "addressed" l-values (like
variable references and pointer dereferences) and bitfield,
@property, and vector-component l-values.
Currently we're not using these for much, but I aim to switch
pretty much everything calculating l-valueness over to them.
For now they shouldn't necessarily be trusted.
Marcin Swiderski [Thu, 18 Nov 2010 06:29:23 +0000 (06:29 +0000)]
Added method for handling CXXOperatorCallExpr differently from CallExpr if CXXOperatorCallExpr represents method call.
Also fixed returning ExpolodedNodeSet from VisitCXXMethodCallExpr.
Introduce option -Wargument-larger-than[=N] which warns about function definitions if they take by-value
or return by-value any POD that is larger than some threshold (default is 64 bytes).
Douglas Gregor [Wed, 17 Nov 2010 21:36:08 +0000 (21:36 +0000)]
Do not produce "purely informative" code completion results for
Objective-C message sends, which have only whitespace in their
TypedText chunk. Such results have no purpose.
Don't warn for parentheses for the '&&' inside '||' for cases like:
assert(a || b && "bad");
since this is safe. This way we avoid a big source of such warnings which in this case are practically useless.
Note that we don't handle *all* cases where precedence wouldn't matter because of constants since
this is a bit costly to check, and IMO clarifying precedence with parentheses is good for
readability in general.
Douglas Gregor [Wed, 17 Nov 2010 17:15:08 +0000 (17:15 +0000)]
When libclang is walking a member access expression, don't walk into
an implicit "this"; it causes clang_getCursor() to find the implicit
"this" expression (which isn't written in the source!) rather than the
actual member.
Douglas Gregor [Wed, 17 Nov 2010 17:14:07 +0000 (17:14 +0000)]
When comparing the source range of a declaration against the region of
interest (e.g., as used by clang_getCursor()), count the
decl-specifier-seq as part of the source range, as we do for
clang_annotateTokens(). Makes clang_getCursor() work properly for the
result types of functions, for example.
Chris Lattner [Wed, 17 Nov 2010 08:25:26 +0000 (08:25 +0000)]
When forming the !srcloc mdnode for an inline asm, add the SourceLocations
of all the lines of the inline asm. With the refactoring and enhancement
of the backend, we can now reports errors on the correct source line when
an asm contains multiple lines of text. For something like this:
void foo() {
asm("push %rax\n"
".code32\n");
}
we used to get this: (note that the line 4 in t.c isn't helpful)
t.c:4:7: error: warning: ignoring directive for now
asm("push %rax\n"
^
<inline asm>:2:1: note: instantiated into assembly here
.code32
^
now we get:
t.c:5:8: error: warning: ignoring directive for now
".code32\n"
^
<inline asm>:2:1: note: instantiated into assembly here
.code32
^
Note that we're pointing to line 5 properly now. This implements
rdar://7839391 - inline asm errors should point to the right line in the asm
and makes the error message in PR8595 much less confusing.
Chris Lattner [Wed, 17 Nov 2010 07:37:15 +0000 (07:37 +0000)]
a metric ton of refactoring later, Sema::getLocationOfStringLiteralByte
no longer depends on Preprocessor, so we can move it out of Sema into
a nice new StringLiteral::getLocationOfByte method that can be used by
any AST client.
projects / clang / log