So the issue here was that we have lldb_private::FileSpec that by default was
always resolving a path when using the:
FileSpec::FileSpec (const char *path);
and in the:
void FileSpec::SetFile(const char *pathname, bool resolve = true);
This isn't what we want in many many cases. One example is you have "/tmp" on
your file system which is really "/private/tmp". You compile code in that
directory and end up with debug info that mentions "/tmp/file.c". Then you
type:
(lldb) breakpoint set --file file.c --line 5
If your current working directory is "/tmp", then "file.c" would be turned
into "/private/tmp/file.c" which won't match anything in the debug info.
Also, it should have been just a FileSpec with no directory and a filename
of "file.c" which could (and should) potentially match any instances of "file.c"
in the debug info.
So I removed the constructor that just takes a path:
FileSpec::FileSpec (const char *path); // REMOVED
You must now use the other constructor that has a "bool resolve" parameter that you must always supply:
FileSpec::FileSpec (const char *path, bool resolve);
I also removed the default parameter to SetFile():
void FileSpec::SetFile(const char *pathname, bool resolve);
And fixed all of the code to use the right settings.
llvm-svn: 116944
We now spawn a thread to accept a unix socket connection from the inferior
when it spawns in the terminal, then we launch the process, then we get
the pid back through the unix socket, and then wait for it to SIGSTOP.
darwin-debug now clears the terminal screen and prints out the program and
arguments that are about to be launched.
llvm-svn: 116841
but something is still killing our inferior.
Fixed an issue with darwin-debug where it wasn't passing all needed arguments
to the inferior.
Fixed a race condition with the attach to named process code.
llvm-svn: 116697
static bool
Host::GetLLDBPath (lldb::PathType path_type, FileSpec &file_spec);
This will fill in "file_spec" with an appropriate path that is appropriate
for the current Host OS. MacOSX will return paths within the LLDB.framework,
and other unixes will return the paths they want. The current PathType
enums are:
typedef enum PathType
{
ePathTypeLLDBShlibDir, // The directory where the lldb.so (unix) or LLDB mach-o file in LLDB.framework (MacOSX) exists
ePathTypeSupportExecutableDir, // Find LLDB support executable directory (debugserver, etc)
ePathTypeHeaderDir, // Find LLDB header file directory
ePathTypePythonDir // Find Python modules (PYTHONPATH) directory
} PathType;
All places that were finding executables are and python paths are now updated
to use this Host call.
Added another new host call to launch the inferior in a terminal. This ability
will be very host specific and doesn't need to be supported on all systems.
MacOSX currently will create a new .command file and tell Terminal.app to open
the .command file. It also uses the new "darwin-debug" app which is a small
app that uses posix to exec (no fork) and stop at the entry point of the
program. The GDB remote plug-in is almost able launch a process and attach to
it, it currently will spawn the process, but it won't attach to it just yet.
This will let LLDB not have to share the terminal with another process and a
new terminal window will pop up when you launch. This won't get hooked up
until we work out all of the kinks. The new Host function is:
static lldb::pid_t
Host::LaunchInNewTerminal (
const char **argv, // argv[0] is executable
const char **envp,
const ArchSpec *arch_spec,
bool stop_at_entry,
bool disable_aslr);
Cleaned up FileSpec::GetPath to not use strncpy() as it was always zero
filling the entire path buffer.
Fixed an issue with the dynamic checker function where I missed a '$' prefix
that should have been added.
llvm-svn: 116690
function. It will inspect NAME and do the following:
- if the name contains '(' or starts with "-[" or "+[" then a full name search
will happen to match full function names with args (C++ demangled names) or
full objective C method prototypes.
- if the name contains "::" and no '(', then it is assumed to be a qualified
function name that is in a namespace or class. For "foo::bar::baz" we will
search for any functions with the basename or method name of "baz", then
filter the results to only those that contain "foo::bar::baz". This allows
setting breakpoint on C++ functions and methods without having to fully
qualify all of the types that would appear in C++ mangled names.
- if the name contains ":" (not "::"), then NAME is assumed to be an ObjC
selector.
_ otherwise, we assume just a plain function basename.
Now that "--name" is our "auto" mode, I introduced the new "--basename" option
("breakpoint set --basename NAME") to allow for function names that aren't
methods or selectors, just basenames. This can also be used to ignore C++
namespaces and class hierarchies for class methods.
Fixed clang enumeration promotion types to be correct.
llvm-svn: 116293
being chopped up correctly). The DWARF plug-in also keeps a map of the ObjC
class names to selectors for easy parsing of all class selectors when we parse
the class type.
llvm-svn: 116290
results. The clang opaque type for the expression result will be added to the
Target's ASTContext, and the bytes will be stored in a DataBuffer inside
the new object. The class is named: ValueObjectConstResult
Now after an expression is evaluated, we can get a ValueObjectSP back that
contains a ValueObjectConstResult object.
Relocated the value object dumping code into a static function within
the ValueObject class instead of being in the CommandObjectFrame.cpp file
which is what contained the code to dump variables ("frame variables").
llvm-svn: 115578
Added the start of Host specific launch services, though it currently isn't
hookup up to anything. We want to be able to launch a process and use the
native launch services to launch an app like it would be launched by the
user double clicking on the app. We also eventually want to be able to run
a command line app in a newly spawned terminal to avoid terminal sharing.
Fixed an issue with the new DWARF forward type declaration stuff. A crasher
was found that was happening when trying to properly expand the forward
declarations.
llvm-svn: 115213
adding methods to C++ and objective C classes. In order to make methods, we
need the function prototype which means we need the arguments. Parsing these
could cause a circular reference that caused an assertion.
Added a new typedef for the clang opaque types which are just void pointers:
lldb::clang_type_t. This appears in lldb-types.h.
This was fixed by enabling struct, union, class, and enum types to only get
a forward declaration when we make the clang opaque qual type for these
types. When they need to actually be resolved, lldb_private::Type will call
a new function in the SymbolFile protocol to resolve a clang type when it is
not fully defined (clang::TagDecl::getDefinition() returns NULL). This allows
us to be a lot more lazy when parsing clang types and keeps down the amount
of data that gets parsed into the ASTContext for each module.
Getting the clang type from a "lldb_private::Type" object now takes a boolean
that indicates if a forward declaration is ok:
clang_type_t lldb_private::Type::GetClangType (bool forward_decl_is_ok);
So function prototypes that define parameters that are "const T&" can now just
parse the forward declaration for type 'T' and we avoid circular references in
the type system.
llvm-svn: 115012
interface in ClangASTContext. Also added two bool returning functions that
indicated if an opaque clang qual type is a CXX class type, and if it is an
ObjC class type.
Objective C classes now will get their methods added lazily as they are
encountered. The reason for this is currently, unlike C++, the
DW_TAG_structure_type and owns the ivars, doesn't not also contain the
member functions. This means when we parse the objective C class interface
we either need to find all functions whose names start with "+[CLASS_NAME"
or "-[CLASS_NAME" and add them all to the class, or when we parse each objective
C function, we slowly add it to the class interface definition. Since objective
C's class doesn't change internal bits according to whether it has certain types
of member functions (like C++ does if it has virtual functions, or if it has
user ctors/dtors), I currently chose to lazily populate the class when each
functions is parsed. Another issue we run into with ObjC method declarations
is the "self" and "_cmd" implicit args are not marked as artificial in the
DWARF (DW_AT_artifical), so we currently have to look for the parameters by
name if we are trying to omit artificial function args if the language of the
compile unit is ObjC or ObjC++.
llvm-svn: 114722
into python-extensions.swig, which gets included into lldb.swig, and
adds them back into the classes when swig generates it's C++ file. This
keeps the Python stuff out of the general API classes.
Also fixed a small bug in the copy constructor for SBSymbolContext.
llvm-svn: 114602
whether a given register number is treated as volatile
or not for a given architecture/platform.
approx 450 lines of boilerplate, 50 lines of actual code. :)
llvm-svn: 114537
accessed by the objects that own the settings. The previous approach wasn't
very usable and made for a lot of unnecessary code just to access variables
that were already owned by the objects.
While I fixed those things, I saw that CommandObject objects should really
have a reference to their command interpreter so they can access the terminal
with if they want to output usaage. Fixed up all CommandObjects to take
an interpreter and cleaned up the API to not need the interpreter to be
passed in.
Fixed the disassemble command to output the usage if no options are passed
down and arguments are passed (all disassebmle variants take options, there
are no "args only").
llvm-svn: 114252
Added a "bool show_fullpaths" to many more objects that were
previously always dumping full paths.
Fixed a few places where the DWARF was not indexed when we
we needed it to be when making queries. Also fixed an issue
where the DWARF in .o files wasn't searching all .o files
for the types.
Fixed an issue with the output from "image lookup --type <TYPENAME>"
where the name and byte size might not be resolved and might not
display. We now call the accessors so we end up seeing all of the
type info.
llvm-svn: 113951
all types in all compile units. I added a new kind of accelerator table to
the DWARF that allows us to index the DWARF compile units and DIEs in a way
that doesn't require the data to stay loaded. Currently when indexing the
DWARF we check if the compile unit had parsed its DIEs and if it hasn't we
index the data and free all of the DIEs so we can reparse later when we need
to after using one of our complete accelerator tables to determine we need
to reparse some DWARF. If the DIEs had already been parsed we leave them
loaded. The new accelerator table uses the "const char *" pointers from our
ConstString class as the keys, and NameToDIE::Info as the value. This info
contains the compile unit index and the DIE index which means we are pointed
right to the DIE we need unlike the other DWARF accelerator tables that often
just point us to the compile unit we would find our answer in.
llvm-svn: 113933
Added the ability to specify a preference for mangled or demangled to Mangled::GetName.
Changed one place where mangled was prefered in GetName.
The Dynamic loader should look up the target of a stub by mangled name if it exists.
llvm-svn: 113869
union, or class that contained an enumeration type. When I was creating
the clang enumeration decl, I wasn't calling "EnumDecl::setIntegerType (QualType)"
which means that if the enum decl was ever asked to figure out it's bit width
(getTypeInfo()) it would crash. We didn't run into this with enum types that
weren't inside classes because the DWARF already told us how big the type was
and when we printed an enum we would never need to calculate the size, we
would use the pre-cached byte size we got from the DWARF. When the enum was
in a struct/union/class and we tried to layout the struct, the layout code
would attempt to get the type info and segfault.
llvm-svn: 113729
we cached remapping information using the old nlist index to the
new symbol index, yet we tried to lookup the symbol stubs that
were for symbols that had been remapped by ID instead of using
the new symbol index. This is now fixed and the mach-o symbol tables
are fixed.
Use the delta between two vector entries to determine the stride
in case any padding is inserted by compilers for bsearch calls
on symbol tables when finding symbols by their original ID.
llvm-svn: 113719
up a seciton offset address (SBAddress) within a module that returns a
symbol context (SBSymbolContext). Also added a SBSymbolContextList in
preparation for adding find/lookup APIs that can return multiple results.
Added a lookup example code that shows how to do address lookups.
llvm-svn: 113599
The Unwind and RegisterContext subclasses still need
to be finished; none of this code is used by lldb at
this point (unless you call into it by hand).
The ObjectFile class now has an UnwindTable object.
The UnwindTable object has a series of FuncUnwinders
objects (Function Unwinders) -- one for each function
in that ObjectFile we've backtraced through during this
debug session.
The FuncUnwinders object has a few different UnwindPlans.
UnwindPlans are a generic way of describing how to find
the canonical address of a given function's stack frame
(the CFA idea from DWARF/eh_frame) and how to restore the
caller frame's register values, if they have been saved
by this function.
UnwindPlans are created from different sources. One source is the
eh_frame exception handling information generated by the compiler
for unwinding an exception throw. Another source is an assembly
language inspection class (UnwindAssemblyProfiler, uses the Plugin
architecture) which looks at the instructions in the funciton
prologue and describes the stack movements/register saves that are
done.
Two additional types of UnwindPlans that are worth noting are
the "fast" stack UnwindPlan which is useful for making a first
pass over a thread's stack, determining how many stack frames there
are and retrieving the pc and CFA values for each frame (enough
to create StackFrameIDs). Only a minimal set of registers is
recovered during a fast stack walk.
The final UnwindPlan is an architectural default unwind plan.
These are provided by the ArchDefaultUnwindPlan class (which uses
the plugin architecture). When no symbol/function address range can
be found for a given pc value -- when we have no eh_frame information
and when we don't have a start address so we can't examine the assembly
language instrucitons -- we have to make a best guess about how to
unwind. That's when we use the architectural default UnwindPlan.
On x86_64, this would be to assume that rbp is used as a stack pointer
and we can use that to find the caller's frame pointer and pc value.
It's a last-ditch best guess about how to unwind out of a frame.
There are heuristics about when to use one UnwindPlan versues the other --
this will all happen in the still-begin-written UnwindLLDB subclass of
Unwind which runs the UnwindPlans.
llvm-svn: 113581
Renamed the "dispatchqaddr" setting that was coming back for stop reply packets
to be named "qaddr" so that gdb doesn't thing it is a register number. gdb
was checking the first character and assuming "di" was a hex register number
because 'd' is a hex digit. It has been shortened so gdb can safely ignore it.
llvm-svn: 113475
new change will omit unneeded symbol table entries and coalesce
function entries (N_FUN stab entries) with their linker code
symbol (N_SECT symbols) into only the function symbol to avoid
duplicate symbol table entries. It will also coalesce N_STSYM and
the data linker symbol into just one static data symbol.
llvm-svn: 113363
parent, sibling and first child block, and access to the
inline function information.
Added an accessor the StackFrame:
Block * lldb_private::StackFrame::GetFrameBlock();
LLDB represents inline functions as lexical blocks that have
inlined function information in them. The function above allows
us to easily get the top most lexical block that defines a stack
frame. When there are no inline functions in function, the block
returned ends up being the top most block for the function. When
the PC is in an inlined funciton for a frame, this will return the
first parent block that has inlined function information. The
other accessor: StackFrame::GetBlock() will return the deepest block
that matches the frame's PC value. Since most debuggers want to display
all variables in the current frame, the Block returned by
StackFrame::GetFrameBlock can be used to retrieve all variables for
the current frame.
Fixed the lldb_private::Block::DumpStopContext(...) to properly
display inline frames a block should display all of its inlined
functions. Prior to this fix, one of the call sites was being skipped.
This is a separate code path from the current default where inlined
functions get their own frames.
Fixed an issue where a block would always grab variables for any
child inline function blocks.
llvm-svn: 113195
handles user settable internal variables (the equivalent of set/show
variables in gdb). In addition to the basic infrastructure (most of
which is defined in UserSettingsController.{h,cpp}, there are examples
of two classes that have been set up to contain user settable
variables (the Debugger and Process classes). The 'settings' command
has been modified to be a command-subcommand structure, and the 'set',
'show' and 'append' commands have been moved into this sub-commabnd
structure. The old StateVariable class has been completely replaced
by this, and the state variable dictionary has been removed from the
Command Interpreter. Places that formerly accessed the state variable
mechanism have been modified to access the variables in this new
structure instead (checking the term-width; getting/checking the
prompt; etc.)
Variables are attached to classes; there are two basic "flavors" of
variables that can be set: "global" variables (static/class-wide), and
"instance" variables (one per instance of the class). The whole thing
has been set up so that any global or instance variable can be set at
any time (e.g. on start up, in your .lldbinit file), whether or not
any instances actually exist (there's a whole pending and default
values mechanism to help deal with that).
llvm-svn: 113041
function statics, file globals and static variables) that a frame contains.
The StackFrame objects can give out ValueObjects instances for
each variable which allows us to track when a variable changes and doesn't
depend on variable names when getting value objects.
StackFrame::GetVariableList now takes a boolean to indicate if we want to
get the frame compile unit globals and static variables.
The value objects in the stack frames can now correctly track when they have
been modified. There are a few more tweaks needed to complete this work. The
biggest issue is when stepping creates partial stacks (just frame zero usually)
and causes previous stack frames not to match up with the current stack frames
because the previous frames only has frame zero. We don't really want to
require that all previous frames be complete since stepping often must check
stack frames to complete their jobs. I will fix this issue tomorrow.
llvm-svn: 112800
expressions. Values used by the expression are
checked by validation functions which cause the
program to crash if the values are unsafe.
Major changes:
- Added IRDynamicChecks.[ch], which contains the
core code related to this feature
- Modified CommandObjectExpression to install the
validator functions into the target process.
- Added an accessor to Process that gets/sets the
helper functions
llvm-svn: 112690
debugger to insert self-contained functions for use by
expressions (mainly for error-checking).
In order to support detecting whether a crash occurred
in one of these helpers -- currently our preferred way
of reporting that an error-check failed -- added a bit
of support for getting the extent of a JITted function
in addition to just its base.
llvm-svn: 112324
The goal is to separate the parser's data from the data
belonging to the parser's clients. This allows clients
to use the parser to obtain (for example) a JIT compiled
function or some DWARF code, and then discard the parser
state.
Previously, parser state was held in ClangExpression and
used liberally by ClangFunction, which inherited from
ClangExpression. The main effects of this refactoring
are:
- reducing ClangExpression to an abstract class that
declares methods that any client must expose to the
expression parser,
- moving the code specific to implementing the "expr"
command from ClangExpression and
CommandObjectExpression into ClangUserExpression,
a new class,
- moving the common parser interaction code from
ClangExpression into ClangExpressionParser, a new
class, and
- making ClangFunction rely only on
ClangExpressionParser and not depend on the
internal implementation of ClangExpression.
Side effects include:
- the compiler interaction code has been factored
out of ClangFunction and is now in an AST pass
(ASTStructExtractor),
- the header file for ClangFunction is now fully
documented,
- several bugs that only popped up when Clang was
deallocated (which never happened, since the
lifetime of the compiler was essentially infinite)
are now fixed, and
- the developer-only "call" command has been
disabled.
I have tested the expr command and the Objective-C
step-into code, which use ClangUserExpression and
ClangFunction, respectively, and verified that they
work. Please let me know if you encounter bugs or
poor documentation.
llvm-svn: 112249
Greg Clayton