2013-04-11 08:09:05 +08:00
//===-- Materializer.cpp ----------------------------------------*- C++ -*-===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
2013-04-13 02:10:34 +08:00
# include "lldb/Core/Log.h"
2013-04-16 04:51:24 +08:00
# include "lldb/Core/RegisterValue.h"
2013-04-13 02:10:34 +08:00
# include "lldb/Core/ValueObjectConstResult.h"
2013-04-13 05:40:34 +08:00
# include "lldb/Core/ValueObjectVariable.h"
2013-04-11 08:09:05 +08:00
# include "lldb/Expression/ClangExpressionVariable.h"
# include "lldb/Expression/Materializer.h"
# include "lldb/Symbol/ClangASTContext.h"
# include "lldb/Symbol/Symbol.h"
# include "lldb/Symbol/Type.h"
# include "lldb/Symbol/Variable.h"
# include "lldb/Target/ExecutionContext.h"
2013-04-16 04:51:24 +08:00
# include "lldb/Target/RegisterContext.h"
2013-04-13 05:40:34 +08:00
# include "lldb/Target/StackFrame.h"
This commit changes the way LLDB executes user
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
2013-04-19 06:06:33 +08:00
# include "lldb/Target/Target.h"
2013-04-11 08:09:05 +08:00
using namespace lldb_private ;
uint32_t
Materializer : : AddStructMember ( Entity & entity )
{
uint32_t size = entity . GetSize ( ) ;
uint32_t alignment = entity . GetAlignment ( ) ;
uint32_t ret ;
2013-04-12 05:16:36 +08:00
if ( m_current_offset = = 0 )
2013-04-11 08:09:05 +08:00
m_struct_alignment = alignment ;
if ( m_current_offset % alignment )
m_current_offset + = ( alignment - ( m_current_offset % alignment ) ) ;
ret = m_current_offset ;
m_current_offset + = size ;
return ret ;
}
void
Materializer : : Entity : : SetSizeAndAlignmentFromType ( ClangASTType & type )
{
m_size = type . GetTypeByteSize ( ) ;
uint32_t bit_alignment = type . GetTypeBitAlign ( ) ;
if ( bit_alignment % 8 )
{
bit_alignment + = 8 ;
bit_alignment & = ~ ( ( uint32_t ) 0x111u ) ;
}
m_alignment = bit_alignment / 8 ;
}
class EntityPersistentVariable : public Materializer : : Entity
{
public :
EntityPersistentVariable ( lldb : : ClangExpressionVariableSP & persistent_variable_sp ) :
Entity ( ) ,
m_persistent_variable_sp ( persistent_variable_sp )
{
2013-04-17 07:25:35 +08:00
// Hard-coding to maximum size of a pointer since persistent variables are materialized by reference
m_size = 8 ;
m_alignment = 8 ;
2013-04-11 08:09:05 +08:00
}
2013-04-13 02:10:34 +08:00
void MakeAllocation ( IRMemoryMap & map , Error & err )
{
Log * log ( lldb_private : : GetLogIfAllCategoriesSet ( LIBLLDB_LOG_EXPRESSIONS ) ) ;
// Allocate a spare memory area to store the persistent variable's contents.
Error allocate_error ;
lldb : : addr_t mem = map . Malloc ( m_persistent_variable_sp - > GetByteSize ( ) ,
8 ,
lldb : : ePermissionsReadable | lldb : : ePermissionsWritable ,
IRMemoryMap : : eAllocationPolicyMirror ,
allocate_error ) ;
if ( ! allocate_error . Success ( ) )
{
2013-05-17 08:55:28 +08:00
err . SetErrorStringWithFormat ( " couldn't allocate a memory area to store %s: %s " , m_persistent_variable_sp - > GetName ( ) . GetCString ( ) , allocate_error . AsCString ( ) ) ;
2013-04-13 02:10:34 +08:00
return ;
}
if ( log )
log - > Printf ( " Allocated %s (0x% " PRIx64 " ) sucessfully " , m_persistent_variable_sp - > GetName ( ) . GetCString ( ) , mem ) ;
// Put the location of the spare memory into the live data of the ValueObject.
m_persistent_variable_sp - > m_live_sp = ValueObjectConstResult : : Create ( map . GetBestExecutionContextScope ( ) ,
m_persistent_variable_sp - > GetTypeFromUser ( ) . GetASTContext ( ) ,
m_persistent_variable_sp - > GetTypeFromUser ( ) . GetOpaqueQualType ( ) ,
m_persistent_variable_sp - > GetName ( ) ,
mem ,
eAddressTypeLoad ,
m_persistent_variable_sp - > GetByteSize ( ) ) ;
// Clear the flag if the variable will never be deallocated.
if ( m_persistent_variable_sp - > m_flags & ClangExpressionVariable : : EVKeepInTarget )
2013-05-23 06:49:06 +08:00
{
Error leak_error ;
map . Leak ( mem , leak_error ) ;
2013-04-13 02:10:34 +08:00
m_persistent_variable_sp - > m_flags & = ~ ClangExpressionVariable : : EVNeedsAllocation ;
2013-05-23 06:49:06 +08:00
}
2013-04-13 02:10:34 +08:00
// Write the contents of the variable to the area.
Error write_error ;
map . WriteMemory ( mem ,
m_persistent_variable_sp - > GetValueBytes ( ) ,
m_persistent_variable_sp - > GetByteSize ( ) ,
write_error ) ;
if ( ! write_error . Success ( ) )
{
2013-05-17 08:55:28 +08:00
err . SetErrorStringWithFormat ( " couldn't write %s to the target: %s " , m_persistent_variable_sp - > GetName ( ) . AsCString ( ) ,
2013-04-13 02:10:34 +08:00
write_error . AsCString ( ) ) ;
return ;
}
}
void DestroyAllocation ( IRMemoryMap & map , Error & err )
{
Error deallocate_error ;
map . Free ( ( lldb : : addr_t ) m_persistent_variable_sp - > m_live_sp - > GetValue ( ) . GetScalar ( ) . ULongLong ( ) , deallocate_error ) ;
2013-04-20 10:40:45 +08:00
m_persistent_variable_sp - > m_live_sp . reset ( ) ;
2013-04-13 02:10:34 +08:00
if ( ! deallocate_error . Success ( ) )
{
2013-05-17 08:55:28 +08:00
err . SetErrorStringWithFormat ( " couldn't deallocate memory for %s: %s " , m_persistent_variable_sp - > GetName ( ) . GetCString ( ) , deallocate_error . AsCString ( ) ) ;
2013-04-13 02:10:34 +08:00
}
}
2013-04-17 07:25:35 +08:00
void Materialize ( lldb : : StackFrameSP & frame_sp , IRMemoryMap & map , lldb : : addr_t process_address , Error & err )
2013-04-11 08:09:05 +08:00
{
2013-04-13 02:10:34 +08:00
Log * log ( lldb_private : : GetLogIfAllCategoriesSet ( LIBLLDB_LOG_EXPRESSIONS ) ) ;
if ( log )
{
log - > Printf ( " EntityPersistentVariable::Materialize [process_address = 0x%llx, m_name = %s, m_flags = 0x%hx] " ,
( uint64_t ) process_address ,
m_persistent_variable_sp - > GetName ( ) . AsCString ( ) ,
m_persistent_variable_sp - > m_flags ) ;
}
if ( m_persistent_variable_sp - > m_flags & ClangExpressionVariable : : EVNeedsAllocation )
{
MakeAllocation ( map , err ) ;
2013-04-20 10:40:45 +08:00
m_persistent_variable_sp - > m_flags | = ClangExpressionVariable : : EVIsLLDBAllocated ;
2013-04-13 02:10:34 +08:00
if ( ! err . Success ( ) )
return ;
}
if ( ( m_persistent_variable_sp - > m_flags & ClangExpressionVariable : : EVIsProgramReference & & m_persistent_variable_sp - > m_live_sp ) | |
m_persistent_variable_sp - > m_flags & ClangExpressionVariable : : EVIsLLDBAllocated )
{
Error write_error ;
2013-04-20 10:40:45 +08:00
2013-04-13 02:10:34 +08:00
map . WriteScalarToMemory ( process_address + m_offset ,
m_persistent_variable_sp - > m_live_sp - > GetValue ( ) . GetScalar ( ) ,
2013-04-20 10:40:45 +08:00
map . GetAddressByteSize ( ) ,
2013-04-13 02:10:34 +08:00
write_error ) ;
if ( ! write_error . Success ( ) )
{
2013-05-17 08:55:28 +08:00
err . SetErrorStringWithFormat ( " couldn't write the location of %s to memory: %s " , m_persistent_variable_sp - > GetName ( ) . AsCString ( ) , write_error . AsCString ( ) ) ;
2013-04-13 02:10:34 +08:00
}
}
else
{
2013-05-17 08:55:28 +08:00
err . SetErrorStringWithFormat ( " no materialization happened for persistent variable %s " , m_persistent_variable_sp - > GetName ( ) . AsCString ( ) ) ;
2013-04-13 02:10:34 +08:00
return ;
}
2013-04-11 08:09:05 +08:00
}
2013-04-17 07:25:35 +08:00
void Dematerialize ( lldb : : StackFrameSP & frame_sp , IRMemoryMap & map , lldb : : addr_t process_address ,
lldb : : addr_t frame_top , lldb : : addr_t frame_bottom , Error & err )
2013-04-11 08:09:05 +08:00
{
2013-04-13 02:10:34 +08:00
Log * log ( lldb_private : : GetLogIfAllCategoriesSet ( LIBLLDB_LOG_EXPRESSIONS ) ) ;
if ( log )
{
log - > Printf ( " EntityPersistentVariable::Dematerialize [process_address = 0x%llx, m_name = %s, m_flags = 0x%hx] " ,
( uint64_t ) process_address ,
m_persistent_variable_sp - > GetName ( ) . AsCString ( ) ,
m_persistent_variable_sp - > m_flags ) ;
}
if ( ( m_persistent_variable_sp - > m_flags & ClangExpressionVariable : : EVIsLLDBAllocated ) | |
( m_persistent_variable_sp - > m_flags & ClangExpressionVariable : : EVIsProgramReference ) )
{
if ( m_persistent_variable_sp - > m_flags & ClangExpressionVariable : : EVIsProgramReference & &
! m_persistent_variable_sp - > m_live_sp )
{
// If the reference comes from the program, then the ClangExpressionVariable's
// live variable data hasn't been set up yet. Do this now.
2013-04-17 07:25:35 +08:00
lldb : : addr_t location ;
2013-04-13 02:10:34 +08:00
Error read_error ;
2013-04-17 07:25:35 +08:00
map . ReadPointerFromMemory ( & location , process_address + m_offset , read_error ) ;
2013-04-13 02:10:34 +08:00
if ( ! read_error . Success ( ) )
{
2013-05-17 08:55:28 +08:00
err . SetErrorStringWithFormat ( " couldn't read the address of program-allocated variable %s: %s " , m_persistent_variable_sp - > GetName ( ) . GetCString ( ) , read_error . AsCString ( ) ) ;
2013-04-13 02:10:34 +08:00
return ;
}
2013-04-17 07:25:35 +08:00
2013-04-13 02:10:34 +08:00
m_persistent_variable_sp - > m_live_sp = ValueObjectConstResult : : Create ( map . GetBestExecutionContextScope ( ) ,
m_persistent_variable_sp - > GetTypeFromUser ( ) . GetASTContext ( ) ,
m_persistent_variable_sp - > GetTypeFromUser ( ) . GetOpaqueQualType ( ) ,
m_persistent_variable_sp - > GetName ( ) ,
location ,
eAddressTypeLoad ,
m_persistent_variable_sp - > GetByteSize ( ) ) ;
if ( frame_top ! = LLDB_INVALID_ADDRESS & &
frame_bottom ! = LLDB_INVALID_ADDRESS & &
location > = frame_bottom & &
location < = frame_top )
{
// If the variable is resident in the stack frame created by the expression,
// then it cannot be relied upon to stay around. We treat it as needing
// reallocation.
m_persistent_variable_sp - > m_flags | = ClangExpressionVariable : : EVIsLLDBAllocated ;
m_persistent_variable_sp - > m_flags | = ClangExpressionVariable : : EVNeedsAllocation ;
m_persistent_variable_sp - > m_flags | = ClangExpressionVariable : : EVNeedsFreezeDry ;
m_persistent_variable_sp - > m_flags & = ~ ClangExpressionVariable : : EVIsProgramReference ;
}
}
lldb : : addr_t mem = m_persistent_variable_sp - > m_live_sp - > GetValue ( ) . GetScalar ( ) . ULongLong ( ) ;
if ( ! m_persistent_variable_sp - > m_live_sp )
{
2013-05-17 08:55:28 +08:00
err . SetErrorStringWithFormat ( " couldn't find the memory area used to store %s " , m_persistent_variable_sp - > GetName ( ) . GetCString ( ) ) ;
2013-04-13 02:10:34 +08:00
return ;
}
if ( m_persistent_variable_sp - > m_live_sp - > GetValue ( ) . GetValueAddressType ( ) ! = eAddressTypeLoad )
{
2013-05-17 08:55:28 +08:00
err . SetErrorStringWithFormat ( " the address of the memory area for %s is in an incorrect format " , m_persistent_variable_sp - > GetName ( ) . GetCString ( ) ) ;
2013-04-13 02:10:34 +08:00
return ;
}
if ( m_persistent_variable_sp - > m_flags & ClangExpressionVariable : : EVNeedsFreezeDry | |
m_persistent_variable_sp - > m_flags & ClangExpressionVariable : : EVKeepInTarget )
{
if ( log )
log - > Printf ( " Dematerializing %s from 0x% " PRIx64 " (size = %llu) " , m_persistent_variable_sp - > GetName ( ) . GetCString ( ) , ( uint64_t ) mem , ( unsigned long long ) m_persistent_variable_sp - > GetByteSize ( ) ) ;
// Read the contents of the spare memory area
m_persistent_variable_sp - > ValueUpdated ( ) ;
Error read_error ;
map . ReadMemory ( m_persistent_variable_sp - > GetValueBytes ( ) ,
mem ,
m_persistent_variable_sp - > GetByteSize ( ) ,
read_error ) ;
if ( ! read_error . Success ( ) )
{
2013-05-17 08:55:28 +08:00
err . SetErrorStringWithFormat ( " couldn't read the contents of %s from memory: %s " , m_persistent_variable_sp - > GetName ( ) . GetCString ( ) , read_error . AsCString ( ) ) ;
2013-04-13 02:10:34 +08:00
return ;
}
m_persistent_variable_sp - > m_flags & = ~ ClangExpressionVariable : : EVNeedsFreezeDry ;
}
}
else
{
2013-05-17 08:55:28 +08:00
err . SetErrorStringWithFormat ( " no dematerialization happened for persistent variable %s " , m_persistent_variable_sp - > GetName ( ) . AsCString ( ) ) ;
2013-04-13 02:10:34 +08:00
return ;
}
2013-04-20 10:40:45 +08:00
lldb : : ProcessSP process_sp = map . GetBestExecutionContextScope ( ) - > CalculateProcess ( ) ;
if ( ! process_sp | |
! process_sp - > CanJIT ( ) )
{
// Allocations are not persistent so persistent variables cannot stay materialized.
m_persistent_variable_sp - > m_flags | = ClangExpressionVariable : : EVNeedsAllocation ;
DestroyAllocation ( map , err ) ;
if ( ! err . Success ( ) )
return ;
}
else if ( m_persistent_variable_sp - > m_flags & ClangExpressionVariable : : EVNeedsAllocation & &
! ( m_persistent_variable_sp - > m_flags & ClangExpressionVariable : : EVKeepInTarget ) )
2013-04-13 02:10:34 +08:00
{
DestroyAllocation ( map , err ) ;
if ( ! err . Success ( ) )
return ;
}
2013-04-11 08:09:05 +08:00
}
2013-04-16 06:48:23 +08:00
void DumpToLog ( IRMemoryMap & map , lldb : : addr_t process_address , Log * log )
{
StreamString dump_stream ;
Error err ;
dump_stream . Printf ( " 0x%llx: EntityPersistentVariable (%s) \n " , ( unsigned long long ) process_address + m_offset , m_persistent_variable_sp - > GetName ( ) . AsCString ( ) ) ;
{
dump_stream . Printf ( " Pointer: \n " ) ;
DataBufferHeap data ( m_size , 0 ) ;
map . ReadMemory ( data . GetBytes ( ) , process_address + m_offset , m_size , err ) ;
if ( ! err . Success ( ) )
{
dump_stream . Printf ( " <could not be read> \n " ) ;
}
else
{
DataExtractor extractor ( data . GetBytes ( ) , data . GetByteSize ( ) , map . GetByteOrder ( ) , map . GetAddressByteSize ( ) ) ;
extractor . DumpHexBytes ( & dump_stream , data . GetBytes ( ) , data . GetByteSize ( ) , 16 , process_address + m_offset ) ;
2013-04-17 07:25:35 +08:00
dump_stream . PutChar ( ' \n ' ) ;
2013-04-16 06:48:23 +08:00
}
}
{
dump_stream . Printf ( " Target: \n " ) ;
lldb : : addr_t target_address ;
map . ReadPointerFromMemory ( & target_address , process_address + m_offset , err ) ;
if ( ! err . Success ( ) )
{
dump_stream . Printf ( " <could not be read> \n " ) ;
}
else
{
DataBufferHeap data ( m_persistent_variable_sp - > GetByteSize ( ) , 0 ) ;
map . ReadMemory ( data . GetBytes ( ) , target_address , m_persistent_variable_sp - > GetByteSize ( ) , err ) ;
if ( ! err . Success ( ) )
{
dump_stream . Printf ( " <could not be read> \n " ) ;
}
else
{
DataExtractor extractor ( data . GetBytes ( ) , data . GetByteSize ( ) , map . GetByteOrder ( ) , map . GetAddressByteSize ( ) ) ;
extractor . DumpHexBytes ( & dump_stream , data . GetBytes ( ) , data . GetByteSize ( ) , 16 , target_address ) ;
2013-04-17 07:25:35 +08:00
dump_stream . PutChar ( ' \n ' ) ;
2013-04-16 06:48:23 +08:00
}
}
}
log - > PutCString ( dump_stream . GetData ( ) ) ;
}
2013-04-17 07:25:35 +08:00
void Wipe ( IRMemoryMap & map , lldb : : addr_t process_address )
{
}
2013-04-11 08:09:05 +08:00
private :
lldb : : ClangExpressionVariableSP m_persistent_variable_sp ;
} ;
uint32_t
Materializer : : AddPersistentVariable ( lldb : : ClangExpressionVariableSP & persistent_variable_sp , Error & err )
{
EntityVector : : iterator iter = m_entities . insert ( m_entities . end ( ) , EntityUP ( ) ) ;
iter - > reset ( new EntityPersistentVariable ( persistent_variable_sp ) ) ;
uint32_t ret = AddStructMember ( * * iter ) ;
( * iter ) - > SetOffset ( ret ) ;
return ret ;
}
class EntityVariable : public Materializer : : Entity
{
public :
EntityVariable ( lldb : : VariableSP & variable_sp ) :
Entity ( ) ,
2013-04-13 05:40:34 +08:00
m_variable_sp ( variable_sp ) ,
m_is_reference ( false ) ,
2013-04-16 06:48:23 +08:00
m_temporary_allocation ( LLDB_INVALID_ADDRESS ) ,
m_temporary_allocation_size ( 0 )
2013-04-11 08:09:05 +08:00
{
2013-04-13 05:40:34 +08:00
// Hard-coding to maximum size of a pointer since all variables are materialized by reference
2013-04-12 05:16:36 +08:00
m_size = 8 ;
m_alignment = 8 ;
2013-04-13 05:40:34 +08:00
m_is_reference = ClangASTContext : : IsReferenceType ( m_variable_sp - > GetType ( ) - > GetClangForwardType ( ) ) ;
2013-04-11 08:09:05 +08:00
}
2013-04-17 07:25:35 +08:00
void Materialize ( lldb : : StackFrameSP & frame_sp , IRMemoryMap & map , lldb : : addr_t process_address , Error & err )
2013-04-11 08:09:05 +08:00
{
2013-04-13 05:40:34 +08:00
Log * log ( lldb_private : : GetLogIfAllCategoriesSet ( LIBLLDB_LOG_EXPRESSIONS ) ) ;
if ( log )
{
log - > Printf ( " EntityVariable::Materialize [process_address = 0x%llx, m_variable_sp = %s] " ,
( uint64_t ) process_address ,
m_variable_sp - > GetName ( ) . AsCString ( ) ) ;
}
2013-04-16 01:12:47 +08:00
ExecutionContextScope * scope = frame_sp . get ( ) ;
if ( ! scope )
scope = map . GetBestExecutionContextScope ( ) ;
lldb : : ValueObjectSP valobj_sp = ValueObjectVariable : : Create ( scope , m_variable_sp ) ;
2013-04-13 05:40:34 +08:00
if ( ! valobj_sp )
{
2013-05-17 08:55:28 +08:00
err . SetErrorStringWithFormat ( " couldn't get a value object for variable %s " , m_variable_sp - > GetName ( ) . AsCString ( ) ) ;
2013-04-13 05:40:34 +08:00
return ;
}
if ( m_is_reference )
{
DataExtractor valobj_extractor ;
valobj_sp - > GetData ( valobj_extractor ) ;
lldb : : offset_t offset = 0 ;
lldb : : addr_t reference_addr = valobj_extractor . GetAddress ( & offset ) ;
Error write_error ;
map . WritePointerToMemory ( process_address + m_offset , reference_addr , write_error ) ;
if ( ! write_error . Success ( ) )
{
2013-05-17 08:55:28 +08:00
err . SetErrorStringWithFormat ( " couldn't write the contents of reference variable %s to memory: %s " , m_variable_sp - > GetName ( ) . AsCString ( ) , write_error . AsCString ( ) ) ;
2013-04-13 05:40:34 +08:00
return ;
}
}
else
{
Error get_address_error ;
lldb : : ValueObjectSP addr_of_valobj_sp = valobj_sp - > AddressOf ( get_address_error ) ;
if ( get_address_error . Success ( ) )
{
DataExtractor valobj_extractor ;
addr_of_valobj_sp - > GetData ( valobj_extractor ) ;
lldb : : offset_t offset = 0 ;
lldb : : addr_t addr_of_valobj_addr = valobj_extractor . GetAddress ( & offset ) ;
Error write_error ;
map . WritePointerToMemory ( process_address + m_offset , addr_of_valobj_addr , write_error ) ;
if ( ! write_error . Success ( ) )
{
2013-05-17 08:55:28 +08:00
err . SetErrorStringWithFormat ( " couldn't write the address of variable %s to memory: %s " , m_variable_sp - > GetName ( ) . AsCString ( ) , write_error . AsCString ( ) ) ;
2013-04-13 05:40:34 +08:00
return ;
}
}
else
{
DataExtractor data ;
valobj_sp - > GetData ( data ) ;
This commit changes the way LLDB executes user
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
2013-04-19 06:06:33 +08:00
if ( m_temporary_allocation ! = LLDB_INVALID_ADDRESS )
2013-04-13 05:40:34 +08:00
{
2013-05-17 08:55:28 +08:00
err . SetErrorStringWithFormat ( " trying to create a temporary region for %s but one exists " , m_variable_sp - > GetName ( ) . AsCString ( ) ) ;
2013-04-13 05:40:34 +08:00
return ;
}
if ( data . GetByteSize ( ) ! = m_variable_sp - > GetType ( ) - > GetByteSize ( ) )
{
2013-05-17 08:55:28 +08:00
if ( data . GetByteSize ( ) = = 0 & & m_variable_sp - > LocationExpression ( ) . IsValid ( ) = = false )
{
err . SetErrorStringWithFormat ( " the variable '%s' has no location, it may have been optimized out " , m_variable_sp - > GetName ( ) . AsCString ( ) ) ;
}
else
{
err . SetErrorStringWithFormat ( " size of variable %s disagrees with the ValueObject's size " , m_variable_sp - > GetName ( ) . AsCString ( ) ) ;
}
2013-04-13 05:40:34 +08:00
return ;
}
size_t bit_align = ClangASTType : : GetTypeBitAlign ( m_variable_sp - > GetType ( ) - > GetClangAST ( ) , m_variable_sp - > GetType ( ) - > GetClangLayoutType ( ) ) ;
size_t byte_align = ( bit_align + 7 ) / 8 ;
Error alloc_error ;
m_temporary_allocation = map . Malloc ( data . GetByteSize ( ) , byte_align , lldb : : ePermissionsReadable | lldb : : ePermissionsWritable , IRMemoryMap : : eAllocationPolicyMirror , alloc_error ) ;
2013-04-16 06:48:23 +08:00
m_temporary_allocation_size = data . GetByteSize ( ) ;
2013-04-13 05:40:34 +08:00
if ( ! alloc_error . Success ( ) )
{
2013-05-17 08:55:28 +08:00
err . SetErrorStringWithFormat ( " couldn't allocate a temporary region for %s: %s " , m_variable_sp - > GetName ( ) . AsCString ( ) , alloc_error . AsCString ( ) ) ;
2013-04-13 05:40:34 +08:00
return ;
}
Error write_error ;
map . WriteMemory ( m_temporary_allocation , data . GetDataStart ( ) , data . GetByteSize ( ) , write_error ) ;
if ( ! write_error . Success ( ) )
{
2013-05-17 08:55:28 +08:00
err . SetErrorStringWithFormat ( " couldn't write to the temporary region for %s: %s " , m_variable_sp - > GetName ( ) . AsCString ( ) , write_error . AsCString ( ) ) ;
2013-04-13 05:40:34 +08:00
return ;
}
Error pointer_write_error ;
map . WritePointerToMemory ( process_address + m_offset , m_temporary_allocation , pointer_write_error ) ;
if ( ! pointer_write_error . Success ( ) )
{
2013-05-17 08:55:28 +08:00
err . SetErrorStringWithFormat ( " couldn't write the address of the temporary region for %s: %s " , m_variable_sp - > GetName ( ) . AsCString ( ) , pointer_write_error . AsCString ( ) ) ;
2013-04-13 05:40:34 +08:00
}
}
}
2013-04-11 08:09:05 +08:00
}
2013-04-17 07:25:35 +08:00
void Dematerialize ( lldb : : StackFrameSP & frame_sp , IRMemoryMap & map , lldb : : addr_t process_address ,
2013-04-20 03:47:32 +08:00
lldb : : addr_t frame_top , lldb : : addr_t frame_bottom , Error & err )
2013-04-11 08:09:05 +08:00
{
2013-04-13 05:40:34 +08:00
Log * log ( lldb_private : : GetLogIfAllCategoriesSet ( LIBLLDB_LOG_EXPRESSIONS ) ) ;
if ( log )
{
log - > Printf ( " EntityVariable::Dematerialize [process_address = 0x%llx, m_variable_sp = %s] " ,
( uint64_t ) process_address ,
m_variable_sp - > GetName ( ) . AsCString ( ) ) ;
}
if ( m_temporary_allocation ! = LLDB_INVALID_ADDRESS )
{
2013-04-16 01:12:47 +08:00
ExecutionContextScope * scope = frame_sp . get ( ) ;
if ( ! scope )
scope = map . GetBestExecutionContextScope ( ) ;
lldb : : ValueObjectSP valobj_sp = ValueObjectVariable : : Create ( scope , m_variable_sp ) ;
2013-04-13 05:40:34 +08:00
if ( ! valobj_sp )
{
2013-05-17 08:55:28 +08:00
err . SetErrorStringWithFormat ( " couldn't get a value object for variable %s " , m_variable_sp - > GetName ( ) . AsCString ( ) ) ;
2013-04-13 05:40:34 +08:00
return ;
}
2013-04-13 10:06:42 +08:00
lldb_private : : DataExtractor data ;
Error extract_error ;
map . GetMemoryData ( data , m_temporary_allocation , valobj_sp - > GetByteSize ( ) , extract_error ) ;
if ( ! extract_error . Success ( ) )
{
2013-05-17 08:55:28 +08:00
err . SetErrorStringWithFormat ( " couldn't get the data for variable %s " , m_variable_sp - > GetName ( ) . AsCString ( ) ) ;
2013-04-13 10:06:42 +08:00
return ;
}
Error set_error ;
valobj_sp - > SetData ( data , set_error ) ;
if ( ! set_error . Success ( ) )
{
2013-05-17 08:55:28 +08:00
err . SetErrorStringWithFormat ( " couldn't write the new contents of %s back into the variable " , m_variable_sp - > GetName ( ) . AsCString ( ) ) ;
2013-04-13 10:06:42 +08:00
return ;
}
2013-04-13 05:40:34 +08:00
Error free_error ;
map . Free ( m_temporary_allocation , free_error ) ;
if ( ! free_error . Success ( ) )
{
2013-05-17 08:55:28 +08:00
err . SetErrorStringWithFormat ( " couldn't free the temporary region for %s: %s " , m_variable_sp - > GetName ( ) . AsCString ( ) , free_error . AsCString ( ) ) ;
2013-04-13 05:40:34 +08:00
return ;
}
m_temporary_allocation = LLDB_INVALID_ADDRESS ;
2013-04-16 06:48:23 +08:00
m_temporary_allocation_size = 0 ;
2013-04-13 05:40:34 +08:00
}
2013-04-11 08:09:05 +08:00
}
2013-04-16 06:48:23 +08:00
void DumpToLog ( IRMemoryMap & map , lldb : : addr_t process_address , Log * log )
{
StreamString dump_stream ;
This commit changes the way LLDB executes user
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
2013-04-19 06:06:33 +08:00
dump_stream . Printf ( " 0x%llx: EntityVariable \n " , ( unsigned long long ) process_address + m_offset ) ;
2013-04-16 06:48:23 +08:00
This commit changes the way LLDB executes user
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
2013-04-19 06:06:33 +08:00
Error err ;
lldb : : addr_t ptr = LLDB_INVALID_ADDRESS ;
2013-04-16 06:48:23 +08:00
{
dump_stream . Printf ( " Pointer: \n " ) ;
DataBufferHeap data ( m_size , 0 ) ;
map . ReadMemory ( data . GetBytes ( ) , process_address + m_offset , m_size , err ) ;
if ( ! err . Success ( ) )
{
dump_stream . Printf ( " <could not be read> \n " ) ;
}
else
{
DataExtractor extractor ( data . GetBytes ( ) , data . GetByteSize ( ) , map . GetByteOrder ( ) , map . GetAddressByteSize ( ) ) ;
extractor . DumpHexBytes ( & dump_stream , data . GetBytes ( ) , data . GetByteSize ( ) , 16 , process_address + m_offset ) ;
2013-04-17 07:25:35 +08:00
This commit changes the way LLDB executes user
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
2013-04-19 06:06:33 +08:00
lldb : : offset_t offset ;
ptr = extractor . GetPointer ( & offset ) ;
2013-04-17 07:25:35 +08:00
dump_stream . PutChar ( ' \n ' ) ;
2013-04-16 06:48:23 +08:00
}
}
2013-04-17 07:25:35 +08:00
if ( m_temporary_allocation = = LLDB_INVALID_ADDRESS )
2013-04-16 06:48:23 +08:00
{
This commit changes the way LLDB executes user
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
2013-04-19 06:06:33 +08:00
dump_stream . Printf ( " Points to process memory: \n " ) ;
2013-04-16 06:48:23 +08:00
}
else
{
dump_stream . Printf ( " Temporary allocation: \n " ) ;
This commit changes the way LLDB executes user
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
2013-04-19 06:06:33 +08:00
}
if ( ptr = = LLDB_INVALID_ADDRESS )
{
dump_stream . Printf ( " <could not be be found> \n " ) ;
}
else
{
2013-04-17 07:25:35 +08:00
DataBufferHeap data ( m_temporary_allocation_size , 0 ) ;
map . ReadMemory ( data . GetBytes ( ) , m_temporary_allocation , m_temporary_allocation_size , err ) ;
if ( ! err . Success ( ) )
2013-04-16 06:48:23 +08:00
{
dump_stream . Printf ( " <could not be read> \n " ) ;
}
else
{
2013-04-17 07:25:35 +08:00
DataExtractor extractor ( data . GetBytes ( ) , data . GetByteSize ( ) , map . GetByteOrder ( ) , map . GetAddressByteSize ( ) ) ;
2013-04-16 06:48:23 +08:00
2013-04-17 07:25:35 +08:00
extractor . DumpHexBytes ( & dump_stream , data . GetBytes ( ) , data . GetByteSize ( ) , 16 , process_address + m_offset ) ;
2013-04-16 06:48:23 +08:00
2013-04-17 07:25:35 +08:00
dump_stream . PutChar ( ' \n ' ) ;
2013-04-16 06:48:23 +08:00
}
}
log - > PutCString ( dump_stream . GetData ( ) ) ;
}
2013-04-17 07:25:35 +08:00
void Wipe ( IRMemoryMap & map , lldb : : addr_t process_address )
{
if ( m_temporary_allocation ! = LLDB_INVALID_ADDRESS )
{
Error free_error ;
map . Free ( m_temporary_allocation , free_error ) ;
m_temporary_allocation = LLDB_INVALID_ADDRESS ;
m_temporary_allocation_size = 0 ;
}
}
2013-04-11 08:09:05 +08:00
private :
2013-04-13 05:40:34 +08:00
lldb : : VariableSP m_variable_sp ;
bool m_is_reference ;
lldb : : addr_t m_temporary_allocation ;
2013-04-16 06:48:23 +08:00
size_t m_temporary_allocation_size ;
2013-04-11 08:09:05 +08:00
} ;
uint32_t
Materializer : : AddVariable ( lldb : : VariableSP & variable_sp , Error & err )
{
EntityVector : : iterator iter = m_entities . insert ( m_entities . end ( ) , EntityUP ( ) ) ;
iter - > reset ( new EntityVariable ( variable_sp ) ) ;
uint32_t ret = AddStructMember ( * * iter ) ;
( * iter ) - > SetOffset ( ret ) ;
return ret ;
}
class EntityResultVariable : public Materializer : : Entity
{
public :
This commit changes the way LLDB executes user
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
2013-04-19 06:06:33 +08:00
EntityResultVariable ( const TypeFromUser & type , bool is_program_reference , bool keep_in_memory ) :
2013-04-11 08:09:05 +08:00
Entity ( ) ,
2013-04-13 02:10:34 +08:00
m_type ( type ) ,
This commit changes the way LLDB executes user
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
2013-04-19 06:06:33 +08:00
m_is_program_reference ( is_program_reference ) ,
m_keep_in_memory ( keep_in_memory ) ,
m_temporary_allocation ( LLDB_INVALID_ADDRESS ) ,
m_temporary_allocation_size ( 0 )
2013-04-11 08:09:05 +08:00
{
This commit changes the way LLDB executes user
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
2013-04-19 06:06:33 +08:00
// Hard-coding to maximum size of a pointer since all results are materialized by reference
m_size = 8 ;
m_alignment = 8 ;
2013-04-11 08:09:05 +08:00
}
2013-04-17 07:25:35 +08:00
void Materialize ( lldb : : StackFrameSP & frame_sp , IRMemoryMap & map , lldb : : addr_t process_address , Error & err )
2013-04-11 08:09:05 +08:00
{
This commit changes the way LLDB executes user
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
2013-04-19 06:06:33 +08:00
if ( ! m_is_program_reference )
{
if ( m_temporary_allocation ! = LLDB_INVALID_ADDRESS )
{
err . SetErrorString ( " Trying to create a temporary region for the result but one exists " ) ;
return ;
}
size_t byte_size = m_type . GetTypeByteSize ( ) ;
size_t bit_align = m_type . GetTypeBitAlign ( ) ;
size_t byte_align = ( bit_align + 7 ) / 8 ;
Error alloc_error ;
m_temporary_allocation = map . Malloc ( byte_size , byte_align , lldb : : ePermissionsReadable | lldb : : ePermissionsWritable , IRMemoryMap : : eAllocationPolicyMirror , alloc_error ) ;
m_temporary_allocation_size = byte_size ;
if ( ! alloc_error . Success ( ) )
{
2013-05-17 08:55:28 +08:00
err . SetErrorStringWithFormat ( " couldn't allocate a temporary region for the result: %s " , alloc_error . AsCString ( ) ) ;
This commit changes the way LLDB executes user
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
2013-04-19 06:06:33 +08:00
return ;
}
Error pointer_write_error ;
map . WritePointerToMemory ( process_address + m_offset , m_temporary_allocation , pointer_write_error ) ;
if ( ! pointer_write_error . Success ( ) )
{
2013-05-17 08:55:28 +08:00
err . SetErrorStringWithFormat ( " couldn't write the address of the temporary region for the result: %s " , pointer_write_error . AsCString ( ) ) ;
This commit changes the way LLDB executes user
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
2013-04-19 06:06:33 +08:00
}
}
2013-04-11 08:09:05 +08:00
}
2013-04-17 07:25:35 +08:00
void Dematerialize ( lldb : : StackFrameSP & frame_sp , IRMemoryMap & map , lldb : : addr_t process_address ,
This commit changes the way LLDB executes user
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
2013-04-19 06:06:33 +08:00
lldb : : addr_t frame_top , lldb : : addr_t frame_bottom , Error & err )
2013-04-11 08:09:05 +08:00
{
This commit changes the way LLDB executes user
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
2013-04-19 06:06:33 +08:00
err . SetErrorString ( " Tried to detmaterialize a result variable with the normal Dematerialize method " ) ;
}
void Dematerialize ( lldb : : ClangExpressionVariableSP & result_variable_sp ,
lldb : : StackFrameSP & frame_sp , IRMemoryMap & map , lldb : : addr_t process_address ,
lldb : : addr_t frame_top , lldb : : addr_t frame_bottom , Error & err )
{
err . Clear ( ) ;
ExecutionContextScope * exe_scope = map . GetBestExecutionContextScope ( ) ;
if ( ! exe_scope )
{
err . SetErrorString ( " Couldn't dematerialize a result variable: invalid execution context scope " ) ;
return ;
}
lldb : : addr_t address ;
Error read_error ;
map . ReadPointerFromMemory ( & address , process_address + m_offset , read_error ) ;
if ( ! read_error . Success ( ) )
{
err . SetErrorString ( " Couldn't dematerialize a result variable: couldn't read its address " ) ;
return ;
}
lldb : : TargetSP target_sp = exe_scope - > CalculateTarget ( ) ;
if ( ! target_sp )
{
err . SetErrorString ( " Couldn't dematerialize a result variable: no target " ) ;
return ;
}
ConstString name = target_sp - > GetPersistentVariables ( ) . GetNextPersistentVariableName ( ) ;
lldb : : ClangExpressionVariableSP ret ;
ret = target_sp - > GetPersistentVariables ( ) . CreateVariable ( exe_scope ,
name ,
m_type ,
map . GetByteOrder ( ) ,
map . GetAddressByteSize ( ) ) ;
if ( ! ret )
{
2013-05-17 08:55:28 +08:00
err . SetErrorStringWithFormat ( " couldn't dematerialize a result variable: failed to make persistent variable %s " , name . AsCString ( ) ) ;
This commit changes the way LLDB executes user
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
2013-04-19 06:06:33 +08:00
return ;
}
2013-04-20 10:40:45 +08:00
lldb : : ProcessSP process_sp = map . GetBestExecutionContextScope ( ) - > CalculateProcess ( ) ;
2013-04-27 10:19:33 +08:00
bool can_persist = ( m_is_program_reference & & process_sp & & process_sp - > CanJIT ( ) & & ! ( address > = frame_bottom & & address < frame_top ) ) ;
2013-04-20 10:40:45 +08:00
if ( can_persist & & m_keep_in_memory )
{
ret - > m_live_sp = ValueObjectConstResult : : Create ( exe_scope ,
m_type . GetASTContext ( ) ,
m_type . GetOpaqueQualType ( ) ,
name ,
address ,
eAddressTypeLoad ,
ret - > GetByteSize ( ) ) ;
}
This commit changes the way LLDB executes user
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
2013-04-19 06:06:33 +08:00
ret - > ValueUpdated ( ) ;
const size_t pvar_byte_size = ret - > GetByteSize ( ) ;
uint8_t * pvar_data = ret - > GetValueBytes ( ) ;
map . ReadMemory ( pvar_data , address , pvar_byte_size , read_error ) ;
if ( ! read_error . Success ( ) )
{
err . SetErrorString ( " Couldn't dematerialize a result variable: couldn't read its memory " ) ;
return ;
}
result_variable_sp = ret ;
2013-04-20 10:40:45 +08:00
if ( ! can_persist | | ! m_keep_in_memory )
This commit changes the way LLDB executes user
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
2013-04-19 06:06:33 +08:00
{
ret - > m_flags | = ClangExpressionVariable : : EVNeedsAllocation ;
2013-04-20 10:40:45 +08:00
if ( m_temporary_allocation ! = LLDB_INVALID_ADDRESS )
{
Error free_error ;
map . Free ( m_temporary_allocation , free_error ) ;
}
This commit changes the way LLDB executes user
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
2013-04-19 06:06:33 +08:00
}
else
{
ret - > m_flags | = ClangExpressionVariable : : EVIsLLDBAllocated ;
}
m_temporary_allocation = LLDB_INVALID_ADDRESS ;
m_temporary_allocation_size = 0 ;
2013-04-11 08:09:05 +08:00
}
2013-04-16 06:48:23 +08:00
void DumpToLog ( IRMemoryMap & map , lldb : : addr_t process_address , Log * log )
{
StreamString dump_stream ;
dump_stream . Printf ( " 0x%llx: EntityResultVariable \n " , ( unsigned long long ) process_address + m_offset ) ;
This commit changes the way LLDB executes user
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
2013-04-19 06:06:33 +08:00
Error err ;
lldb : : addr_t ptr = LLDB_INVALID_ADDRESS ;
{
dump_stream . Printf ( " Pointer: \n " ) ;
DataBufferHeap data ( m_size , 0 ) ;
map . ReadMemory ( data . GetBytes ( ) , process_address + m_offset , m_size , err ) ;
if ( ! err . Success ( ) )
{
dump_stream . Printf ( " <could not be read> \n " ) ;
}
else
{
DataExtractor extractor ( data . GetBytes ( ) , data . GetByteSize ( ) , map . GetByteOrder ( ) , map . GetAddressByteSize ( ) ) ;
extractor . DumpHexBytes ( & dump_stream , data . GetBytes ( ) , data . GetByteSize ( ) , 16 , process_address + m_offset ) ;
lldb : : offset_t offset ;
ptr = extractor . GetPointer ( & offset ) ;
dump_stream . PutChar ( ' \n ' ) ;
}
}
if ( m_temporary_allocation = = LLDB_INVALID_ADDRESS )
{
dump_stream . Printf ( " Points to process memory: \n " ) ;
}
else
{
dump_stream . Printf ( " Temporary allocation: \n " ) ;
}
if ( ptr = = LLDB_INVALID_ADDRESS )
{
dump_stream . Printf ( " <could not be be found> \n " ) ;
}
else
{
DataBufferHeap data ( m_temporary_allocation_size , 0 ) ;
map . ReadMemory ( data . GetBytes ( ) , m_temporary_allocation , m_temporary_allocation_size , err ) ;
if ( ! err . Success ( ) )
{
dump_stream . Printf ( " <could not be read> \n " ) ;
}
else
{
DataExtractor extractor ( data . GetBytes ( ) , data . GetByteSize ( ) , map . GetByteOrder ( ) , map . GetAddressByteSize ( ) ) ;
extractor . DumpHexBytes ( & dump_stream , data . GetBytes ( ) , data . GetByteSize ( ) , 16 , process_address + m_offset ) ;
dump_stream . PutChar ( ' \n ' ) ;
}
}
2013-04-16 06:48:23 +08:00
log - > PutCString ( dump_stream . GetData ( ) ) ;
}
2013-04-17 07:25:35 +08:00
void Wipe ( IRMemoryMap & map , lldb : : addr_t process_address )
{
This commit changes the way LLDB executes user
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
2013-04-19 06:06:33 +08:00
if ( ! m_keep_in_memory & & m_temporary_allocation ! = LLDB_INVALID_ADDRESS )
{
Error free_error ;
map . Free ( m_temporary_allocation , free_error ) ;
}
m_temporary_allocation = LLDB_INVALID_ADDRESS ;
m_temporary_allocation_size = 0 ;
2013-04-17 07:25:35 +08:00
}
2013-04-11 08:09:05 +08:00
private :
This commit changes the way LLDB executes user
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
2013-04-19 06:06:33 +08:00
TypeFromUser m_type ;
bool m_is_program_reference ;
2013-04-13 02:10:34 +08:00
bool m_keep_in_memory ;
This commit changes the way LLDB executes user
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
2013-04-19 06:06:33 +08:00
lldb : : addr_t m_temporary_allocation ;
size_t m_temporary_allocation_size ;
2013-04-11 08:09:05 +08:00
} ;
uint32_t
This commit changes the way LLDB executes user
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
2013-04-19 06:06:33 +08:00
Materializer : : AddResultVariable ( const TypeFromUser & type , bool is_program_reference , bool keep_in_memory , Error & err )
2013-04-11 08:09:05 +08:00
{
EntityVector : : iterator iter = m_entities . insert ( m_entities . end ( ) , EntityUP ( ) ) ;
This commit changes the way LLDB executes user
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
2013-04-19 06:06:33 +08:00
iter - > reset ( new EntityResultVariable ( type , is_program_reference , keep_in_memory ) ) ;
2013-04-11 08:09:05 +08:00
uint32_t ret = AddStructMember ( * * iter ) ;
( * iter ) - > SetOffset ( ret ) ;
This commit changes the way LLDB executes user
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
2013-04-19 06:06:33 +08:00
m_result_entity = iter - > get ( ) ;
2013-04-11 08:09:05 +08:00
return ret ;
}
class EntitySymbol : public Materializer : : Entity
{
public :
EntitySymbol ( const Symbol & symbol ) :
Entity ( ) ,
m_symbol ( symbol )
{
// Hard-coding to maximum size of a symbol
m_size = 8 ;
m_alignment = 8 ;
}
2013-04-17 07:25:35 +08:00
void Materialize ( lldb : : StackFrameSP & frame_sp , IRMemoryMap & map , lldb : : addr_t process_address , Error & err )
2013-04-11 08:09:05 +08:00
{
2013-04-16 04:51:24 +08:00
Log * log ( lldb_private : : GetLogIfAllCategoriesSet ( LIBLLDB_LOG_EXPRESSIONS ) ) ;
if ( log )
{
log - > Printf ( " EntitySymbol::Materialize [process_address = 0x%llx, m_symbol = %s] " ,
( uint64_t ) process_address ,
m_symbol . GetName ( ) . AsCString ( ) ) ;
}
2013-04-13 10:25:02 +08:00
Address & sym_address = m_symbol . GetAddress ( ) ;
ExecutionContextScope * exe_scope = map . GetBestExecutionContextScope ( ) ;
lldb : : TargetSP target_sp ;
if ( exe_scope )
target_sp = map . GetBestExecutionContextScope ( ) - > CalculateTarget ( ) ;
if ( ! target_sp )
{
2013-05-17 08:55:28 +08:00
err . SetErrorStringWithFormat ( " couldn't resolve symbol %s because there is no target " , m_symbol . GetName ( ) . AsCString ( ) ) ;
2013-04-13 10:25:02 +08:00
return ;
}
lldb : : addr_t resolved_address = sym_address . GetLoadAddress ( target_sp . get ( ) ) ;
if ( resolved_address = = LLDB_INVALID_ADDRESS )
resolved_address = sym_address . GetFileAddress ( ) ;
Error pointer_write_error ;
map . WritePointerToMemory ( process_address + m_offset , resolved_address , pointer_write_error ) ;
if ( ! pointer_write_error . Success ( ) )
{
2013-05-17 08:55:28 +08:00
err . SetErrorStringWithFormat ( " couldn't write the address of symbol %s: %s " , m_symbol . GetName ( ) . AsCString ( ) , pointer_write_error . AsCString ( ) ) ;
2013-04-16 04:51:24 +08:00
return ;
2013-04-13 10:25:02 +08:00
}
2013-04-11 08:09:05 +08:00
}
2013-04-17 07:25:35 +08:00
void Dematerialize ( lldb : : StackFrameSP & frame_sp , IRMemoryMap & map , lldb : : addr_t process_address ,
2013-04-13 02:10:34 +08:00
lldb : : addr_t frame_top , lldb : : addr_t frame_bottom , Error & err )
2013-04-11 08:09:05 +08:00
{
2013-04-16 04:51:24 +08:00
Log * log ( lldb_private : : GetLogIfAllCategoriesSet ( LIBLLDB_LOG_EXPRESSIONS ) ) ;
if ( log )
{
log - > Printf ( " EntitySymbol::Dematerialize [process_address = 0x%llx, m_symbol = %s] " ,
( uint64_t ) process_address ,
m_symbol . GetName ( ) . AsCString ( ) ) ;
}
2013-04-13 10:25:02 +08:00
// no work needs to be done
2013-04-11 08:09:05 +08:00
}
2013-04-16 06:48:23 +08:00
void DumpToLog ( IRMemoryMap & map , lldb : : addr_t process_address , Log * log )
{
StreamString dump_stream ;
Error err ;
dump_stream . Printf ( " 0x%llx: EntitySymbol (%s) \n " , ( unsigned long long ) process_address + m_offset , m_symbol . GetName ( ) . AsCString ( ) ) ;
{
dump_stream . Printf ( " Pointer: \n " ) ;
DataBufferHeap data ( m_size , 0 ) ;
map . ReadMemory ( data . GetBytes ( ) , process_address + m_offset , m_size , err ) ;
if ( ! err . Success ( ) )
{
dump_stream . Printf ( " <could not be read> \n " ) ;
}
else
{
DataExtractor extractor ( data . GetBytes ( ) , data . GetByteSize ( ) , map . GetByteOrder ( ) , map . GetAddressByteSize ( ) ) ;
extractor . DumpHexBytes ( & dump_stream , data . GetBytes ( ) , data . GetByteSize ( ) , 16 , process_address + m_offset ) ;
2013-04-17 07:25:35 +08:00
dump_stream . PutChar ( ' \n ' ) ;
2013-04-16 06:48:23 +08:00
}
}
log - > PutCString ( dump_stream . GetData ( ) ) ;
}
2013-04-17 07:25:35 +08:00
void Wipe ( IRMemoryMap & map , lldb : : addr_t process_address )
{
}
2013-04-11 08:09:05 +08:00
private :
Symbol m_symbol ;
} ;
uint32_t
Materializer : : AddSymbol ( const Symbol & symbol_sp , Error & err )
{
EntityVector : : iterator iter = m_entities . insert ( m_entities . end ( ) , EntityUP ( ) ) ;
iter - > reset ( new EntitySymbol ( symbol_sp ) ) ;
uint32_t ret = AddStructMember ( * * iter ) ;
( * iter ) - > SetOffset ( ret ) ;
return ret ;
}
class EntityRegister : public Materializer : : Entity
{
public :
EntityRegister ( const RegisterInfo & register_info ) :
Entity ( ) ,
m_register_info ( register_info )
{
// Hard-coding alignment conservatively
m_size = m_register_info . byte_size ;
m_alignment = m_register_info . byte_size ;
}
2013-04-17 07:25:35 +08:00
void Materialize ( lldb : : StackFrameSP & frame_sp , IRMemoryMap & map , lldb : : addr_t process_address , Error & err )
2013-04-11 08:09:05 +08:00
{
2013-04-16 04:51:24 +08:00
Log * log ( lldb_private : : GetLogIfAllCategoriesSet ( LIBLLDB_LOG_EXPRESSIONS ) ) ;
if ( log )
{
log - > Printf ( " EntityRegister::Materialize [process_address = 0x%llx, m_register_info = %s] " ,
( uint64_t ) process_address ,
m_register_info . name ) ;
}
RegisterValue reg_value ;
if ( ! frame_sp . get ( ) )
{
2013-05-17 08:55:28 +08:00
err . SetErrorStringWithFormat ( " couldn't materialize register %s without a stack frame " , m_register_info . name ) ;
2013-04-16 04:51:24 +08:00
return ;
}
lldb : : RegisterContextSP reg_context_sp = frame_sp - > GetRegisterContext ( ) ;
if ( ! reg_context_sp - > ReadRegister ( & m_register_info , reg_value ) )
{
2013-05-17 08:55:28 +08:00
err . SetErrorStringWithFormat ( " couldn't read the value of register %s " , m_register_info . name ) ;
2013-04-16 04:51:24 +08:00
return ;
}
DataExtractor register_data ;
if ( ! reg_value . GetData ( register_data ) )
{
2013-05-17 08:55:28 +08:00
err . SetErrorStringWithFormat ( " couldn't get the data for register %s " , m_register_info . name ) ;
2013-04-16 04:51:24 +08:00
return ;
}
if ( register_data . GetByteSize ( ) ! = m_register_info . byte_size )
{
2013-05-17 08:55:28 +08:00
err . SetErrorStringWithFormat ( " data for register %s had size %llu but we expected %llu " , m_register_info . name , ( unsigned long long ) register_data . GetByteSize ( ) , ( unsigned long long ) m_register_info . byte_size ) ;
2013-04-16 04:51:24 +08:00
return ;
}
Error write_error ;
map . WriteMemory ( process_address + m_offset , register_data . GetDataStart ( ) , register_data . GetByteSize ( ) , write_error ) ;
if ( ! write_error . Success ( ) )
{
2013-05-17 08:55:28 +08:00
err . SetErrorStringWithFormat ( " couldn't write the contents of register %s: %s " , m_register_info . name , write_error . AsCString ( ) ) ;
2013-04-16 04:51:24 +08:00
return ;
}
2013-04-11 08:09:05 +08:00
}
2013-04-17 07:25:35 +08:00
void Dematerialize ( lldb : : StackFrameSP & frame_sp , IRMemoryMap & map , lldb : : addr_t process_address ,
2013-04-13 02:10:34 +08:00
lldb : : addr_t frame_top , lldb : : addr_t frame_bottom , Error & err )
2013-04-11 08:09:05 +08:00
{
2013-04-16 04:51:24 +08:00
Log * log ( lldb_private : : GetLogIfAllCategoriesSet ( LIBLLDB_LOG_EXPRESSIONS ) ) ;
if ( log )
{
log - > Printf ( " EntityRegister::Dematerialize [process_address = 0x%llx, m_register_info = %s] " ,
( uint64_t ) process_address ,
m_register_info . name ) ;
}
Error extract_error ;
DataExtractor register_data ;
if ( ! frame_sp . get ( ) )
{
2013-05-17 08:55:28 +08:00
err . SetErrorStringWithFormat ( " couldn't dematerialize register %s without a stack frame " , m_register_info . name ) ;
2013-04-16 04:51:24 +08:00
return ;
}
lldb : : RegisterContextSP reg_context_sp = frame_sp - > GetRegisterContext ( ) ;
map . GetMemoryData ( register_data , process_address + m_offset , m_register_info . byte_size , extract_error ) ;
if ( ! extract_error . Success ( ) )
{
2013-05-17 08:55:28 +08:00
err . SetErrorStringWithFormat ( " couldn't get the data for register %s: %s " , m_register_info . name , extract_error . AsCString ( ) ) ;
2013-04-16 04:51:24 +08:00
return ;
}
RegisterValue register_value ( const_cast < uint8_t * > ( register_data . GetDataStart ( ) ) , register_data . GetByteSize ( ) , register_data . GetByteOrder ( ) ) ;
if ( ! reg_context_sp - > WriteRegister ( & m_register_info , register_value ) )
{
2013-05-17 08:55:28 +08:00
err . SetErrorStringWithFormat ( " couldn't write the value of register %s " , m_register_info . name ) ;
2013-04-16 04:51:24 +08:00
return ;
}
2013-04-11 08:09:05 +08:00
}
2013-04-16 06:48:23 +08:00
void DumpToLog ( IRMemoryMap & map , lldb : : addr_t process_address , Log * log )
{
StreamString dump_stream ;
Error err ;
dump_stream . Printf ( " 0x%llx: EntityRegister (%s) \n " , ( unsigned long long ) process_address + m_offset , m_register_info . name ) ;
{
dump_stream . Printf ( " Value: \n " ) ;
DataBufferHeap data ( m_size , 0 ) ;
map . ReadMemory ( data . GetBytes ( ) , process_address + m_offset , m_size , err ) ;
if ( ! err . Success ( ) )
{
dump_stream . Printf ( " <could not be read> \n " ) ;
}
else
{
DataExtractor extractor ( data . GetBytes ( ) , data . GetByteSize ( ) , map . GetByteOrder ( ) , map . GetAddressByteSize ( ) ) ;
extractor . DumpHexBytes ( & dump_stream , data . GetBytes ( ) , data . GetByteSize ( ) , 16 , process_address + m_offset ) ;
2013-04-17 07:25:35 +08:00
dump_stream . PutChar ( ' \n ' ) ;
2013-04-16 06:48:23 +08:00
}
}
log - > PutCString ( dump_stream . GetData ( ) ) ;
}
2013-04-17 07:25:35 +08:00
void Wipe ( IRMemoryMap & map , lldb : : addr_t process_address )
{
}
2013-04-11 08:09:05 +08:00
private :
RegisterInfo m_register_info ;
} ;
uint32_t
Materializer : : AddRegister ( const RegisterInfo & register_info , Error & err )
{
EntityVector : : iterator iter = m_entities . insert ( m_entities . end ( ) , EntityUP ( ) ) ;
iter - > reset ( new EntityRegister ( register_info ) ) ;
uint32_t ret = AddStructMember ( * * iter ) ;
( * iter ) - > SetOffset ( ret ) ;
return ret ;
}
Materializer : : Materializer ( ) :
2013-04-17 07:25:35 +08:00
m_dematerializer_wp ( ) ,
This commit changes the way LLDB executes user
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
2013-04-19 06:06:33 +08:00
m_result_entity ( NULL ) ,
2013-04-11 08:09:05 +08:00
m_current_offset ( 0 ) ,
m_struct_alignment ( 8 )
{
}
2013-04-17 07:25:35 +08:00
Materializer : : ~ Materializer ( )
{
DematerializerSP dematerializer_sp = m_dematerializer_wp . lock ( ) ;
if ( dematerializer_sp )
dematerializer_sp - > Wipe ( ) ;
}
2013-04-11 08:09:05 +08:00
2013-04-17 07:25:35 +08:00
Materializer : : DematerializerSP
This commit changes the way LLDB executes user
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
2013-04-19 06:06:33 +08:00
Materializer : : Materialize ( lldb : : StackFrameSP & frame_sp , IRMemoryMap & map , lldb : : addr_t process_address , Error & error )
2013-04-11 08:09:05 +08:00
{
2013-04-16 01:12:47 +08:00
ExecutionContextScope * exe_scope = frame_sp . get ( ) ;
2013-04-17 07:25:35 +08:00
2013-04-16 01:12:47 +08:00
if ( ! exe_scope )
exe_scope = map . GetBestExecutionContextScope ( ) ;
2013-04-17 07:25:35 +08:00
DematerializerSP dematerializer_sp = m_dematerializer_wp . lock ( ) ;
if ( dematerializer_sp )
{
error . SetErrorToGenericError ( ) ;
error . SetErrorString ( " Couldn't materialize: already materialized " ) ;
}
DematerializerSP ret ( new Dematerializer ( * this , frame_sp , map , process_address ) ) ;
2013-04-16 01:12:47 +08:00
if ( ! exe_scope )
{
error . SetErrorToGenericError ( ) ;
2013-04-17 07:25:35 +08:00
error . SetErrorString ( " Couldn't materialize: target doesn't exist " ) ;
2013-04-16 01:12:47 +08:00
}
2013-04-11 08:09:05 +08:00
for ( EntityUP & entity_up : m_entities )
{
entity_up - > Materialize ( frame_sp , map , process_address , error ) ;
if ( ! error . Success ( ) )
2013-04-17 07:25:35 +08:00
return DematerializerSP ( ) ;
2013-04-11 08:09:05 +08:00
}
2013-04-17 07:25:35 +08:00
if ( Log * log = lldb_private : : GetLogIfAllCategoriesSet ( LIBLLDB_LOG_EXPRESSIONS ) )
2013-04-16 06:48:23 +08:00
{
log - > Printf ( " Materializer::Materialize (frame_sp = %p, process_address = 0x%llx) materialized: " , frame_sp . get ( ) , process_address ) ;
for ( EntityUP & entity_up : m_entities )
entity_up - > DumpToLog ( map , process_address , log ) ;
}
2013-04-11 08:09:05 +08:00
2013-04-17 07:25:35 +08:00
m_dematerializer_wp = ret ;
return ret ;
2013-04-11 08:09:05 +08:00
}
void
2013-04-27 10:19:33 +08:00
Materializer : : Dematerializer : : Dematerialize ( Error & error , lldb : : ClangExpressionVariableSP & result_sp , lldb : : addr_t frame_bottom , lldb : : addr_t frame_top )
2013-04-11 08:09:05 +08:00
{
lldb : : StackFrameSP frame_sp = m_frame_wp . lock ( ) ;
2013-04-17 07:25:35 +08:00
ExecutionContextScope * exe_scope = m_map - > GetBestExecutionContextScope ( ) ;
if ( ! IsValid ( ) )
{
error . SetErrorToGenericError ( ) ;
error . SetErrorString ( " Couldn't dematerialize: invalid dematerializer " ) ;
}
2013-04-16 01:12:47 +08:00
if ( ! exe_scope )
2013-04-11 08:09:05 +08:00
{
error . SetErrorToGenericError ( ) ;
2013-04-16 01:12:47 +08:00
error . SetErrorString ( " Couldn't dematerialize: target is gone " ) ;
2013-04-11 08:09:05 +08:00
}
else
{
2013-04-16 06:48:23 +08:00
if ( Log * log = lldb_private : : GetLogIfAllCategoriesSet ( LIBLLDB_LOG_EXPRESSIONS ) )
{
log - > Printf ( " Materializer::Dematerialize (frame_sp = %p, process_address = 0x%llx) about to dematerialize: " , frame_sp . get ( ) , m_process_address ) ;
2013-04-17 07:25:35 +08:00
for ( EntityUP & entity_up : m_materializer - > m_entities )
entity_up - > DumpToLog ( * m_map , m_process_address , log ) ;
2013-04-16 06:48:23 +08:00
}
2013-04-17 07:25:35 +08:00
for ( EntityUP & entity_up : m_materializer - > m_entities )
2013-04-11 08:09:05 +08:00
{
This commit changes the way LLDB executes user
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
2013-04-19 06:06:33 +08:00
if ( entity_up . get ( ) = = m_materializer - > m_result_entity )
{
static_cast < EntityResultVariable * > ( m_materializer - > m_result_entity ) - > Dematerialize ( result_sp , frame_sp , * m_map , m_process_address , frame_top , frame_bottom , error ) ;
}
else
{
entity_up - > Dematerialize ( frame_sp , * m_map , m_process_address , frame_top , frame_bottom , error ) ;
}
2013-04-11 08:09:05 +08:00
if ( ! error . Success ( ) )
break ;
}
}
2013-04-17 07:25:35 +08:00
Wipe ( ) ;
}
void
Materializer : : Dematerializer : : Wipe ( )
{
if ( ! IsValid ( ) )
return ;
for ( EntityUP & entity_up : m_materializer - > m_entities )
{
entity_up - > Wipe ( * m_map , m_process_address ) ;
}
m_materializer = NULL ;
m_map = NULL ;
m_process_address = LLDB_INVALID_ADDRESS ;
2013-04-11 08:09:05 +08:00
}