expressions.
Previously, ClangUserExpression assumed that if
there was a constant result for an expression
then it could be determined during parsing. In
particular, the IRInterpreter ran while parser
state (in particular, ClangExpressionDeclMap)
was present. This approach is flawed, because
the IRInterpreter actually is capable of using
external variables, and hence the result might
be different each run. Until now, we papered
over this flaw by re-parsing the expression each
time we ran it.
I have rewritten the IRInterpreter to be
completely independent of the ClangExpressionDeclMap.
Instead of special-casing external variable lookup,
which ties the IRInterpreter closely to LLDB,
we now interpret the exact same IR that the JIT
would see. This IR assumes that materialization
has occurred; hence the recent implementation of the
Materializer, which does not require parser state
(in the form of ClangExpressionDeclMap) to be
present.
Materialization, interpretation, and dematerialization
are now all independent of parsing. This means that
in theory we can parse expressions once and run them
many times. I have three outstanding tasks before
shutting this down:
- First, I will ensure that all of this works with
core files. Core files have a Process but do not
allow allocating memory, which currently confuses
materialization.
- Second, I will make expression breakpoint
conditions remember their ClangUserExpression and
re-use it.
- Third, I will tear out all the redundant code
(for example, materialization logic in
ClangExpressionDeclMap) that is no longer used.
While implementing this fix, I also found a bug in
IRForTarget's handling of floating-point constants.
This should be fixed.
llvm-svn: 179801
Materializer for all expressions that need to
run in the target. This includes the following
changes:
- Removed a bunch of (de-)materialization code
from ClangExpressionDeclMap and assumed the
presence of a Materializer where we previously
had a fallback.
- Ensured that an IRMemoryMap is passed into
ClangExpressionDeclMap::Materialize().
- Fixed object ownership on LLVMContext; it is
now owned by the IRExecutionUnit, since the
Module and the ExecutionEngine both depend on
its existence.
- Fixed a few bugs in IRMemoryMap and the
Materializer that showed up during testing.
llvm-svn: 179649
from IRExecutionUnit into a superclass called
IRMemoryMap. IRMemoryMap handles all reading and
writing, ensuring that areas are kept track of and
memory is properly cached (and deleted).
Also fixed several cases where we would simply leak
binary data in the target process over time. Now
the expression objects explicitly own their
IRExecutionUnit and delete it when they go away. This
is why I had to modify ClangUserExpression,
ClangUtilityFunction, and ClangFunction.
As a side effect of this, I am removing the JIT
mutex for an IRMemoryMap. If it turns out that we
need this mutex, I'll add it in then, but right now
it's just adding complexity.
This is part of a more general project to make
expressions fully reusable. The next step is to
make materialization and dematerialization use
the IRMemoryMap API rather than writing and
reading directly from the process's memory.
This will allow the IR interpreter to use the
same data, but in the host's memory, without having
to use a different set of pointers.
llvm-svn: 178832
Note: although it is now possible to declare blocks
and call them inside the same expression, we do not
generate correct block descriptors so these blocks
cannot be passed to functions like dispatch_async.
<rdar://problem/12578656>
llvm-svn: 178509
LLDB is crashing when logging is enabled from lldb-perf-clang. This has to do with the global destructor chain as the process and its threads are being torn down.
All logging channels now make one and only one instance that is kept in a global pointer which is never freed. This guarantees that logging can correctly continue as the process tears itself down.
llvm-svn: 178191
and the JITted code are managed by a standalone
class that handles memory management itself.
I have removed RecordingMemoryManager and
ProcessDataAllocator, which filled similar roles
and had confusing ownership, with a common class
called IRExecutionUnit. The IRExecutionUnit
manages all allocations ever made for an expression
and frees them when it goes away. It also contains
the code generator and can vend the Module for an
expression to other clases.
The end goal here is to make the output of the
expression parser re-usable; that is, to avoid
re-parsing when re-parsing isn't necessary.
I've also cleaned up some code and used weak pointers
in more places. Please let me know if you see any
leaks; I checked myself as well but I might have
missed a case.
llvm-svn: 177364
Calculate "can branch" using the MC API's rather than our hand-rolled regex'es.
As extra credit, allow setting the disassembly flavor for x86 based architectures to intel or att.
<rdar://problem/11319574>
<rdar://problem/9329275>
llvm-svn: 176392
On x86-64 platforms, the small code model assumes that code will be loaded below the 2GB boundary. With the static relocation model, the fact that the expression code is initially loaded (in the LLDB debugger address space) above that boundary causes problems. Switching to the JITDefault code model causes the large code model to be used for 64-bit targets and small code model of 32-bit targets.
llvm-svn: 175828
Major fixed to allow reading files that are over 4GB. The main problems were that the DataExtractor was using 32 bit offsets as a data cursor, and since we mmap all of our object files we could run into cases where if we had a very large core file that was over 4GB, we were running into the 4GB boundary.
So I defined a new "lldb::offset_t" which should be used for all file offsets.
After making this change, I enabled warnings for data loss and for enexpected implicit conversions temporarily and found a ton of things that I fixed.
Any functions that take an index internally, should use "size_t" for any indexes and also should return "size_t" for any sizes of collections.
llvm-svn: 173463
migration in r171366.
I don't know anything about lldb, but a force run of the build bot indicated it
would need this patch. I'll try to watch the build bot to get it green.
llvm-svn: 171374
- add new header lldb-python.h to be included before other system headers
- short term fix (eventually python dependencies must be cleaned up)
Patch by Matt Kopec!
llvm-svn: 169341
top-of-tree. Removed all local patches and llvm.zip.
The intent is that fron now on top-of-tree will
always build against LLVM/Clang top-of-tree, and
that problems building will be resolved as they
occur. Stable release branches of LLDB can be
constructed as needed and linked to specific release
branches of LLVM/Clang.
llvm-svn: 164563
keep a shared pointer to their disassembler. This
is important for the LLVM-C disassembler because
it needs to lock its parent in order to disassemble
itself.
This means that every interface that returned a
Disassembler* needs to return a DisassemblerSP, so
that the instructions and any external owners share
the same reference count on the object. I changed
all clients to use this shared pointer, which also
plugged a few leaks.
<rdar://problem/12002822>
llvm-svn: 161123
- On iOS, we select the "apcs-gnu" ABI to match
what libraries expect.
- Literals are now allocated at their preferred
alignment, eliminating many alignment crashes.
llvm-svn: 158236
various other syntactic sugar work. Lambdas do
not due to some problems relocating code containing
lambdas. Rvalue references work when returned from
expressions, but need more testing.
llvm-svn: 156948
Fixed the DisassemblerLLVMC disassembler to parse more efficiently instead of parsing opcodes over and over. The InstructionLLVMC class now only reads the opcode in the InstructionLLVMC::Decode function. This can be done very efficiently for ARM and architectures that have fixed opcode sizes. For x64 it still calls the disassembler to get the byte size.
Moved the lldb_private::Instruction::Dump(...) function up into the lldb_private::Instruction class and it now uses the function that gets the mnemonic, operandes and comments so that all disassembly is using the same code.
Added StreamString::FillLastLineToColumn() to allow filling a line up to a column with a character (which is used by the lldb_private::Instruction::Dump(...) function).
Modified the Opcode::GetData() fucntion to "do the right thing" for thumb instructions.
llvm-svn: 156532
This takes two important changes:
- Calling blocks is now supported. You need to
cast their return values, but that works fine.
- We now can correctly run JIT-compiled
expressions that use floating-point numbers.
Also, we have taken a fix that allows us to
ignore access control in Objective-C as in C++.
llvm-svn: 152286
allocations by section. We install these sections
in the target process and inform the JIT of their
new locations.
Also removed some unused variable warnings.
llvm-svn: 151789
I started work on being able to add symbol files after a debug session
had started with a new "target symfile add" command and quickly ran into
problems with stale Address objects in breakpoint locations that had
lldb_private::Section pointers into modules that had been removed or
replaced. This also let to grabbing stale modules from those sections.
So I needed to thread harded the Address, Section and related objects.
To do this I modified the ModuleChild class to now require a ModuleSP
on initialization so that a weak reference can created. I also changed
all places that were handing out "Section *" to have them hand out SectionSP.
All ObjectFile, SymbolFile and SymbolVendors were inheriting from ModuleChild
so all of the find plug-in, static creation function and constructors now
require ModuleSP references instead of Module *.
Address objects now have weak references to their sections which can
safely go stale when a module gets destructed.
This checkin doesn't complete the "target symfile add" command, but it
does get us a lot clioser to being able to do such things without a high
risk of crashing or memory corruption.
llvm-svn: 151336
the lldb_private::StackFrame objects hold onto a weak pointer to the thread
object. The lldb_private::StackFrame objects the the most volatile objects
we have as when we are doing single stepping, frames can often get lost or
thrown away, only to be re-created as another object that still refers to the
same frame. We have another bug tracking that. But we need to be able to
have frames no longer be able to get the thread when they are not part of
a thread anymore, and this is the first step (this fix makes that possible
but doesn't implement it yet).
Also changed lldb_private::ExecutionContextScope to return shared pointers to
all objects in the execution context to further thread harden the internals.
llvm-svn: 150871
LLVM/Clang. This brings in several fixes, including:
- Improvements in the Just-In-Time compiler's
allocation of memory: the JIT now allocates
memory in chunks of sections, improving its
ability to generate relocations. I have
revamped the RecordingMemoryManager to reflect
these changes, as well as to get the memory
allocation and data copying out fo the
ClangExpressionParser code. Jim Grosbach wrote
the updates to the JIT on the LLVM side.
- A new ExternalASTSource interface to allow LLDB to
report accurate structure layout information to
Clang. Previously we could only report the sizes
of fields, not their offsets. This meant that if
data structures included field alignment
directives, we could not communicate the necessary
alignment to Clang and accesses to the data would
fail. Now we can (and I have update the relevant
test case). Thanks to Doug Gregor for implementing
the Clang side of this fix.
- The way Objective-C interfaces are completed by
Clang has been made consistent with RecordDecls;
with help from Doug Gregor and Greg Clayton I have
ensured that this still works.
- I have eliminated all local LLVM and Clang patches,
committing the ones that are still relevant to LLVM
and Clang as needed.
I have tested the changes extensively locally, but
please let me know if they cause any trouble for you.
llvm-svn: 149775
an error along with its boolean result. The
expression parser reports this error if the
interpreter fails and the expression could not be
run in the target.
llvm-svn: 148870
parser has hitherto been an implementation waiting
for a use. I have now tied the '-o' option for
the expression command -- which indicates that the
result is an Objective-C object and needs to be
printed -- to the ExpressionParser, which
communicates the desired type to Clang.
Now, if the result of an expression is determined
by an Objective-C method call for which there is
no type information, that result is implicitly
cast to id if and only if the -o option is passed
to the expression command. (Otherwise if there
is no explicit cast Clang will issue an error.
This behavior is identical to what happened before
r146756.)
Also added a testcase for -o enabled and disabled.
llvm-svn: 147099
enhancements. With these enhancements, the return values
of Objective-C methods with unknown return types can be
implicitly cast to id for the purpose of making method
calls.
So what would have required this:
(int)[(id)[ClassWithNoDebugInfo methodReturningObject] methodReturningInt]
can now be written as:
(int)[[ClassWithNoDebugInfo methodReturningObject] methodReturningInt]
llvm-svn: 145567
which will in the future allow expressions to be
compiled as C, C++, and Objective-C instead of the
current default Objective-C++. This feature requires
some additional support from Clang -- specifically, it
requires reference types in the parser regardless of
language -- so it is not yet exposed to the user.
llvm-svn: 144042
C++ vtables, fixing a record layout problem in the
expression parser.
Also fixed various problems with the generation
and unpacking of llvm.zip given our new better
handling of multiple architectures in the LLVM
build.
(And added a log message that will hopefully catch
record layout problems in the future.)
llvm-svn: 143741
Sean Callanan