Variable Formatting#

LLDB has a data formatters subsystem that allows users to define custom display options for their variables.

Usually, when you type frame variable or run some expression LLDB will automatically choose the way to display your results on a per-type basis, as in the following example:

(lldb) frame variable
(uint8_t) x = 'a'
(intptr_t) y = 124752287

Note: frame variable without additional arguments prints the list of variables of the current frame.

However, in certain cases, you may want to associate a different style to the display for certain datatypes. To do so, you need to give hints to the debugger as to how variables should be displayed. The LLDB type command allows you to do just that.

Using it you can change your visualization to look like this:

(lldb) frame variable
(uint8_t) x = chr='a' dec=65 hex=0x41
(intptr_t) y = 0x76f919f

In addition, some data structures can encode their data in a way that is not easily readable to the user, in which case a data formatter can be used to show the data in a human readable way. For example, without a formatter, printing a std::deque<int> with the elements {2, 3, 4, 5, 6} would result in something like:

(lldb) frame variable a_deque
(std::deque<Foo, std::allocator<int> >) $0 = {
   std::_Deque_base<Foo, std::allocator<int> > = {
      _M_impl = {
         _M_map = 0x000000000062ceb0
         _M_map_size = 8
         _M_start = {
            _M_cur = 0x000000000062cf00
            _M_first = 0x000000000062cf00
            _M_last = 0x000000000062d2f4
            _M_node = 0x000000000062cec8
         }
         _M_finish = {
            _M_cur = 0x000000000062d300
            _M_first = 0x000000000062d300
            _M_last = 0x000000000062d6f4
            _M_node = 0x000000000062ced0
         }
      }
   }
}

which is very hard to make sense of.

Note: frame variable <var> prints out the variable <var> in the current frame.

On the other hand, a proper formatter is able to produce the following output:

(lldb) frame variable a_deque
(std::deque<Foo, std::allocator<int> >) $0 = size=5 {
   [0] = 2
   [1] = 3
   [2] = 4
   [3] = 5
   [4] = 6
}

which is what the user would expect from a good debugger.

Note: you can also use v <var> instead of frame variable <var>.

It’s worth mentioning that the size=5 string is produced by a summary provider and the list of children is produced by a synthetic child provider. More information about these providers is available later in this document.

There are several features related to data visualization: formats, summaries, filters, synthetic children.

To reflect this, the type command has five subcommands:

type format
type summary
type filter
type synthetic
type category

These commands are meant to bind printing options to types. When variables are printed, LLDB will first check if custom printing options have been associated to a variable’s type and, if so, use them instead of picking the default choices.

Each of the commands (except type category) has four subcommands available:

  • add: associates a new printing option to one or more types

  • delete: deletes an existing association

  • list: provides a listing of all associations

  • clear: deletes all associations

Type Format#

Type formats enable you to quickly override the default format for displaying primitive types (the usual basic C/C++/ObjC types: int, float, char, …).

If for some reason you want all int variables in your program to print out as hex, you can add a format to the int type.

This is done by typing

(lldb) type format add --format hex int

at the LLDB command line.

The --format (which you can shorten to -f) option accepts a format name. Then, you provide one or more types to which you want the new format applied.

A frequent scenario is that your program has a typedef for a numeric type that you know represents something that must be printed in a certain way. Again, you can add a format just to that typedef by using type format add with the name alias.

But things can quickly get hierarchical. Let’s say you have a situation like the following:

typedef int A;
typedef A B;
typedef B C;
typedef C D;

and you want to show all A’s as hex, all C’s as byte arrays and leave the defaults untouched for other types (albeit its contrived look, the example is far from unrealistic in large software systems).

If you simply type

(lldb) type format add -f hex A
(lldb) type format add -f uint8_t[] C

values of type B will be shown as hex and values of type D as byte arrays, as in:

(lldb) frame variable -T
(A) a = 0x00000001
(B) b = 0x00000002
(C) c = {0x03 0x00 0x00 0x00}
(D) d = {0x04 0x00 0x00 0x00}

This is because by default LLDB cascades formats through typedef chains. In order to avoid that you can use the option -C no to prevent cascading, thus making the two commands required to achieve your goal:

(lldb) type format add -C no -f hex A
(lldb) type format add -C no -f uint8_t[] C

which provides the desired output:

(lldb) frame variable -T
(A) a = 0x00000001
(B) b = 2
(C) c = {0x03 0x00 0x00 0x00}
(D) d = 4

Note, that qualifiers such as const and volatile will be stripped when matching types for example:

(lldb) frame var x y z
(int) x = 1
(const int) y = 2
(volatile int) z = 4
(lldb) type format add -f hex int
(lldb) frame var x y z
(int) x = 0x00000001
(const int) y = 0x00000002
(volatile int) z = 0x00000004

Two additional options that you will want to look at are –skip-pointers (-p) and –skip-references (-r). These two options prevent LLDB from applying a format for type T to values of type T* and T& respectively.

(lldb) type format add -f float32[] int
(lldb) frame variable pointer *pointer -T
(int *) pointer = {1.46991e-39 1.4013e-45}
(int) *pointer = {1.53302e-42}
(lldb) type format add -f float32[] int -p
(lldb) frame variable pointer *pointer -T
(int *) pointer = 0x0000000100100180
(int) *pointer = {1.53302e-42}

While they can be applied to pointers and references, formats will make no attempt to dereference the pointer and extract the value before applying the format, which means you are effectively formatting the address stored in the pointer rather than the pointee value. For this reason, you may want to use the -p option when defining formats.

If you need to delete a custom format simply type type format delete followed by the name of the type to which the format applies.Even if you defined the same format for multiple types on the same command, type format delete will only remove the format for the type name passed as argument.

To delete ALL formats, use type format clear. To see all the formats defined, use type format list.

If all you need to do, however, is display one variable in a custom format, while leaving the others of the same type untouched, you can simply type:

(lldb) frame variable counter -f hex

This has the effect of displaying the value of counter as an hexadecimal number, and will keep showing it this way until you either pick a different format or till you let your program run again.

Finally, this is a list of formatting options available out of which you can pick:

Format name

Abbreviation

Description

default

the default LLDB algorithm is used to pick a format

boolean

B

show this as a true/false boolean, using the customary rule that 0 is false and everything else is true

binary

b

show this as a sequence of bits

bytes

y

show the bytes one after the other

bytes with ASCII

Y

show the bytes, but try to display them as ASCII characters as well

character

c

show the bytes as ASCII characters

printable character

C

show the bytes as printable ASCII characters

complex float

F

interpret this value as the real and imaginary part of a complex floating-point number

c-string

s

show this as a 0-terminated C string

decimal

d

show this as a signed integer number (this does not perform a cast, it simply shows the bytes as an integer with sign)

enumeration

E

show this as an enumeration, printing the value’s name if available or the integer value otherwise

hex

x

show this as in hexadecimal notation (this does not perform a cast, it simply shows the bytes as hex)

float

f

show this as a floating-point number (this does not perform a cast, it simply interprets the bytes as an IEEE754 floating-point value)

octal

o

show this in octal notation

OSType

O

show this as a MacOS OSType

unicode16

U

show this as UTF-16 characters

unicode32

show this as UTF-32 characters

unsigned decimal

u

show this as an unsigned integer number (this does not perform a cast, it simply shows the bytes as unsigned integer)

pointer

p

show this as a native pointer (unless this is really a pointer, the resulting address will probably be invalid)

char[]

show this as an array of characters

int8_t[], uint8_t[] int16_t[], uint16_t[] int32_t[], uint32_t[] int64_t[], uint64_t[] uint128_t[]

show this as an array of the corresponding integer type

float32[], float64[]

show this as an array of the corresponding

floating-point type

complex integer

I

interpret this value as the real and imaginary part of a complex integer number

character array

a

show this as a character array

address

A

show this as an address target (symbol/file/line + offset), possibly also the string this address is pointing to

hex float

show this as hexadecimal floating point

instruction

i

show this as an disassembled opcode

void

v

don’t show anything

Type Summary#

Type formats work by showing a different kind of display for the value of a variable. However, they only work for basic types. When you want to display a class or struct in a custom format, you cannot do that using formats.

A different feature, type summaries, works by extracting information from classes, structures, … (aggregate types) and arranging it in a user-defined format, as in the following example:

before adding a summary…

(lldb) frame variable -T one
(i_am_cool) one = {
   (int) x = 3
   (float) y = 3.14159
   (char) z = 'E'
}

after adding a summary…

(lldb) frame variable one
(i_am_cool) one = int = 3, float = 3.14159, char = 69

There are two ways to use type summaries: the first one is to bind a summary string to the type; the second is to write a Python script that returns the string to be used as summary. Both options are enabled by the type summary add command.

The command to obtain the output shown in the example is:

(lldb) type summary add --summary-string "int = ${var.x}, float = ${var.y}, char = ${var.z%u}" i_am_cool

Initially, we will focus on summary strings, and then describe the Python binding mechanism.

Summary Strings#

Summary strings are written using a simple control language, exemplified by the snippet above. A summary string contains a sequence of tokens that are processed by LLDB to generate the summary.

Summary strings can contain plain text, control characters and special variables that have access to information about the current object and the overall program state.

Plain text is any sequence of characters that doesn’t contain a {, }, $, or \ character, which are the syntax control characters.

The special variables are found in between a “${” prefix, and end with a “}” suffix. Variables can be a simple name or they can refer to complex objects that have subitems themselves. In other words, a variable looks like ${object} or ${object.child.otherchild}. A variable can also be prefixed or suffixed with other symbols meant to change the way its value is handled. An example is ${*var.int_pointer[0-3]}.

Basically, the syntax is the same one described Frame and Thread Formatting plus additional symbols specific for summary strings. The main of them is ${var, which is used refer to the variable that a summary is being created for.

The simplest thing you can do is grab a member variable of a class or structure by typing its expression path. In the previous example, the expression path for the field float y is simply .y. Thus, to ask the summary string to display y you would type ${var.y}.

If you have code like the following:

struct A {
   int x;
   int y;
};
struct B {
   A x;
   A y;
   int *z;
};

the expression path for the y member of the x member of an object of type B would be .x.y and you would type ${var.x.y} to display it in a summary string for type B.

By default, a summary defined for type T, also works for types T* and T& (you can disable this behavior if desired). For this reason, expression paths do not differentiate between . and ->, and the above expression path .x.y would be just as good if you were displaying a B*, or even if the actual definition of B were:

struct B {
   A *x;
   A y;
   int *z;
};

This is unlike the behavior of frame variable which, on the contrary, will enforce the distinction. As hinted above, the rationale for this choice is that waiving this distinction enables you to write a summary string once for type T and use it for both T and T* instances. As a summary string is mostly about extracting nested members’ information, a pointer to an object is just as good as the object itself for the purpose.

If you need to access the value of the integer pointed to by B::z, you cannot simply say ${var.z} because that symbol refers to the pointer z. In order to dereference it and get the pointed value, you should say ${*var.z}. The ${*var tells LLDB to get the object that the expression paths leads to, and then dereference it. In this example is it equivalent to *(bObject.z) in C/C++ syntax. Because . and -> operators can both be used, there is no need to have dereferences in the middle of an expression path (e.g. you do not need to type ${*(var.x).x}) to read A::x as contained in *(B::x). To achieve that effect you can simply write ${var.x->x}, or even ${var.x.x}. The * operator only binds to the result of the whole expression path, rather than piecewise, and there is no way to use parentheses to change that behavior.

Of course, a summary string can contain more than one ${var specifier, and can use ${var and ${*var specifiers together.

Formatting Summary Elements#

An expression path can include formatting codes. Much like the type formats discussed previously, you can also customize the way variables are displayed in summary strings, regardless of the format they have applied to their types. To do that, you can use %format inside an expression path, as in ${var.x->x%u}, which would display the value of x as an unsigned integer.

You can also use some other special format markers, not available for formats themselves, but which carry a special meaning when used in this context:

Symbol

Description

Symbol

Description

%S

Use this object’s summary (the default for aggregate types)

%V

Use this object’s value (the default for non-aggregate types)

%@

Use a language-runtime specific description (for C++ this does nothing,

for Objective-C it calls the NSPrintForDebugger API)

%L

Use this object’s location (memory address, register name, …)

%#

Use the count of the children of this object

%T

Use this object’s datatype name

%N

Print the variable’s basename

%>

Print the expression path for this item

Since lldb 3.7.0, you can also specify ${script.var:pythonFuncName}.

It is expected that the function name you use specifies a function whose signature is the same as a Python summary function. The return string from the function will be placed verbatim in the output.

You cannot use element access, or formatting symbols, in combination with this syntax. For example the following:

${script.var.element[0]:myFunctionName%@}

is not valid and will cause the summary to fail to evaluate.

Element Inlining#

Option –inline-children (-c) to type summary add tells LLDB not to look for a summary string, but instead to just print a listing of all the object’s children on one line.

As an example, given a type pair:

(lldb) frame variable --show-types a_pair
(pair) a_pair = {
   (int) first = 1;
   (int) second = 2;
}

If one types the following commands:

(lldb) type summary add --inline-children pair

the output becomes:

(lldb) frame variable a_pair
(pair) a_pair = (first=1, second=2)

Of course, one can obtain the same effect by typing

(lldb) type summary add pair --summary-string "(first=${var.first}, second=${var.second})"

While the final result is the same, using –inline-children can often save time. If one does not need to see the names of the variables, but just their values, the option –omit-names (-O, uppercase letter o), can be combined with –inline-children to obtain:

(lldb) frame variable a_pair
(pair) a_pair = (1, 2)

which is of course the same as typing

(lldb) type summary add pair --summary-string "(${var.first}, ${var.second})"

Bitfields And Array Syntax#

Sometimes, a basic type’s value actually represents several different values packed together in a bitfield.

With the classical view, there is no way to look at them. Hexadecimal display can help, but if the bits actually span nibble boundaries, the help is limited.

Binary view would show it all without ambiguity, but is often too detailed and hard to read for real-life scenarios.

To cope with the issue, LLDB supports native bitfield formatting in summary strings. If your expression paths leads to a so-called scalar type (the usual int, float, char, double, short, long, long long, double, long double and unsigned variants), you can ask LLDB to only grab some bits out of the value and display them in any format you like. If you only need one bit you can use the [n], just like indexing an array. To extract multiple bits, you can use a slice-like syntax: [n-m], e.g.

(lldb) frame variable float_point
(float) float_point = -3.14159
(lldb) type summary add --summary-string "Sign: ${var[31]%B} Exponent: ${var[30-23]%x} Mantissa: ${var[0-22]%u}" float
(lldb) frame variable float_point
(float) float_point = -3.14159 Sign: true Exponent: 0x00000080 Mantissa: 4788184

In this example, LLDB shows the internal representation of a float variable by extracting bitfields out of a float object.

When typing a range, the extremes n and m are always included, and the order of the indices is irrelevant.

LLDB also allows to use a similar syntax to display array members inside a summary string. For instance, you may want to display all arrays of a given type using a more compact notation than the default, and then just delve into individual array members that prove interesting to your debugging task. You can tell LLDB to format arrays in special ways, possibly independent of the way the array members’ datatype is formatted. e.g.

(lldb) frame variable sarray
(Simple [3]) sarray = {
   [0] = {
      x = 1
      y = 2
      z = '\x03'
   }
   [1] = {
      x = 4
      y = 5
      z = '\x06'
   }
   [2] = {
      x = 7
      y = 8
      z = '\t'
   }
}

(lldb) type summary add --summary-string "${var[].x}" "Simple [3]"

(lldb) frame variable sarray
(Simple [3]) sarray = [1,4,7]

The [] symbol amounts to: if var is an array and I know its size, apply this summary string to every element of the array. Here, we are asking LLDB to display .x for every element of the array, and in fact this is what happens. If you find some of those integers anomalous, you can then inspect that one item in greater detail, without the array format getting in the way:

(lldb) frame variable sarray[1]
(Simple) sarray[1] = {
   x = 4
   y = 5
   z = '\x06'
}

You can also ask LLDB to only print a subset of the array range by using the same syntax used to extract bit for bitfields:

(lldb) type summary add --summary-string "${var[1-2].x}" "Simple [3]"

(lldb) frame variable sarray
(Simple [3]) sarray = [4,7]

If you are dealing with a pointer that you know is an array, you can use this syntax to display the elements contained in the pointed array instead of just the pointer value. However, because pointers have no notion of their size, the empty brackets [] operator does not work, and you must explicitly provide higher and lower bounds.

In general, LLDB needs the square brackets operator [] in order to handle arrays and pointers correctly, and for pointers it also needs a range. However, a few special cases are defined to make your life easier:

you can print a 0-terminated string (C-string) using the %s format, omitting square brackets, as in:

(lldb) type summary add --summary-string "${var%s}" "char *"

This syntax works for char* as well as for char[] because LLDB can rely on the final 0 terminator to know when the string has ended.

LLDB has default summary strings for char* and char[] that use this special case. On debugger startup, the following are defined automatically:

(lldb) type summary add --summary-string "${var%s}" "char *"
(lldb) type summary add --summary-string "${var%s}" -x "char \[[0-9]+]"

any of the array formats (int8_t[], float32{}, …), and the y, Y and a formats work to print an array of a non-aggregate type, even if square brackets are omitted.

(lldb) type summary add --summary-string "${var%int32_t[]}" "int [10]"

This feature, however, is not enabled for pointers because there is no way for LLDB to detect the end of the pointed data.

This also does not work for other formats (e.g. boolean), and you must specify the square brackets operator to get the expected output.

Python Scripting#

Most of the times, summary strings prove good enough for the job of summarizing the contents of a variable. However, as soon as you need to do more than picking some values and rearranging them for display, summary strings stop being an effective tool. This is because summary strings lack the power to actually perform any kind of computation on the value of variables.

To solve this issue, you can bind some Python scripting code as a summary for your datatype, and that script has the ability to both extract children variables as the summary strings do and to perform active computation on the extracted values. As a small example, let’s say we have a Rectangle class:

class Rectangle
{
private:
   int height;
   int width;
public:
   Rectangle() : height(3), width(5) {}
   Rectangle(int H) : height(H), width(H*2-1) {}
   Rectangle(int H, int W) : height(H), width(W) {}
   int GetHeight() { return height; }
   int GetWidth() { return width; }
};

Summary strings are effective to reduce the screen real estate used by the default viewing mode, but are not effective if we want to display the area and perimeter of Rectangle objects

To obtain this, we can simply attach a small Python script to the Rectangle class, as shown in this example:

(lldb) type summary add -P Rectangle
Enter your Python command(s). Type 'DONE' to end.
def function (valobj,internal_dict,options):
   height_val = valobj.GetChildMemberWithName('height')
   width_val = valobj.GetChildMemberWithName('width')
   height = height_val.GetValueAsUnsigned(0)
   width = width_val.GetValueAsUnsigned(0)
   area = height*width
   perimeter = 2*(height + width)
   return 'Area: ' + str(area) + ', Perimeter: ' + str(perimeter)
   DONE
(lldb) frame variable
(Rectangle) r1 = Area: 20, Perimeter: 18
(Rectangle) r2 = Area: 72, Perimeter: 36
(Rectangle) r3 = Area: 16, Perimeter: 16

In order to write effective summary scripts, you need to know the LLDB public API, which is the way Python code can access the LLDB object model. For further details on the API you should look at the LLDB API reference documentation.

As a brief introduction, your script is encapsulated into a function that is passed two parameters: valobj and internal_dict.

internal_dict is an internal support parameter used by LLDB and you should not touch it.

valobj is the object encapsulating the actual variable being displayed, and its type is SBValue. Out of the many possible operations on an SBValue, the basic one is retrieve the children objects it contains (essentially, the fields of the object wrapped by it), by calling GetChildMemberWithName(), passing it the child’s name as a string.

If the variable has a value, you can ask for it, and return it as a string using GetValue(), or as a signed/unsigned number using GetValueAsSigned(), GetValueAsUnsigned(). It is also possible to retrieve an SBData object by calling GetData() and then read the object’s contents out of the SBData.

If you need to delve into several levels of hierarchy, as you can do with summary strings, you can use the method GetValueForExpressionPath(), passing it an expression path just like those you could use for summary strings (one of the differences is that dereferencing a pointer does not occur by prefixing the path with a *`, but by calling the Dereference() method on the returned SBValue). If you need to access array slices, you cannot do that (yet) via this method call, and you must use GetChildAtIndex() querying it for the array items one by one. Also, handling custom formats is something you have to deal with on your own.

options Python summary formatters can optionally define this third argument, which is an object of type lldb.SBTypeSummaryOptions, allowing for a few customizations of the result. The decision to adopt or not this third argument - and the meaning of options thereof - is up to the individual formatter’s writer.

Other than interactively typing a Python script there are two other ways for you to input a Python script as a summary:

  • using the –python-script option to type summary add and typing the script code as an option argument; as in:

(lldb) type summary add --python-script "height = valobj.GetChildMemberWithName('height').GetValueAsUnsigned(0);width = valobj.GetChildMemberWithName('width').GetValueAsUnsigned(0); return 'Area: %d' % (height*width)" Rectangle
  • using the –python-function (-F) option to type summary add and giving the name of a Python function with the correct prototype. Most probably, you will define (or have already defined) the function in the interactive interpreter, or somehow loaded it from a file, using the command script import command. LLDB will emit a warning if it is unable to find the function you passed, but will still register the binding.

Regular Expression Typenames#

As you noticed, in order to associate the custom summary string to the array types, one must give the array size as part of the typename. This can long become tiresome when using arrays of different sizes, Simple [3], Simple [9], Simple [12], …

If you use the -x option, type names are treated as regular expressions instead of type names. This would let you rephrase the above example for arrays of type Simple [3] as:

(lldb) type summary add --summary-string "${var[].x}" -x "Simple \[[0-9]+\]"
(lldb) frame variable
(Simple [3]) sarray = [1,4,7]
(Simple [2]) sother = [3,6]

The above scenario works for Simple [3] as well as for any other array of Simple objects.

While this feature is mostly useful for arrays, you could also use regular expressions to catch other type sets grouped by name. However, as regular expression matching is slower than normal name matching, LLDB will first try to match by name in any way it can, and only when this fails, will it resort to regular expression matching.

One of the ways LLDB uses this feature internally, is to match the names of STL container classes, regardless of the template arguments provided. The details for this are found at FormatManager.cpp

The regular expression language used by LLDB is the POSIX extended language, as defined by the Single UNIX Specification, of which macOS is a compliant implementation.

Names Summaries#

For a given type, there may be different meaningful summary representations. However, currently, only one summary can be associated to a type at each moment. If you need to temporarily override the association for a variable, without changing the summary string for to its type, you can use named summaries.

Named summaries work by attaching a name to a summary when creating it. Then, when there is a need to attach the summary to a variable, the frame variable command, supports a –summary option that tells LLDB to use the named summary given instead of the default one.

(lldb) type summary add --summary-string "x=${var.integer}" --name NamedSummary
(lldb) frame variable one
(i_am_cool) one = int = 3, float = 3.14159, char = 69
(lldb) frame variable one --summary NamedSummary
(i_am_cool) one = x=3

When defining a named summary, binding it to one or more types becomes optional. Even if you bind the named summary to a type, and later change the summary string for that type, the named summary will not be changed by that. You can delete named summaries by using the type summary delete command, as if the summary name was the datatype that the summary is applied to

A summary attached to a variable using the –summary option, has the same semantics that a custom format attached using the -f option has: it stays attached till you attach a new one, or till you let your program run again.

Synthetic Children#

Summaries work well when one is able to navigate through an expression path. In order for LLDB to do so, appropriate debugging information must be available.

Some types are opaque, i.e. no knowledge of their internals is provided. When that’s the case, expression paths do not work correctly.

In other cases, the internals are available to use in expression paths, but they do not provide a user-friendly representation of the object’s value.

For instance, consider an STL vector, as implemented by the GNU C++ Library:

(lldb) frame variable numbers -T
(std::vector<int>) numbers = {
   (std::_Vector_base<int, std::allocator<int> >) std::_Vector_base<int, std::allocator<int> > = {
      (std::_Vector_base<int, std::allocator&tl;int> >::_Vector_impl) _M_impl = {
            (int *) _M_start = 0x00000001001008a0
            (int *) _M_finish = 0x00000001001008a8
            (int *) _M_end_of_storage = 0x00000001001008a8
      }
   }
}

Here, you can see how the type is implemented, and you can write a summary for that implementation but that is not going to help you infer what items are actually stored in the vector.

What you would like to see is probably something like:

(lldb) frame variable numbers -T
(std::vector<int>) numbers = {
   (int) [0] = 1
   (int) [1] = 12
   (int) [2] = 123
   (int) [3] = 1234
}

Synthetic children are a way to get that result.

The feature is based upon the idea of providing a new set of children for a variable that replaces the ones available by default through the debug information. In the example, we can use synthetic children to provide the vector items as children for the std::vector object.

In order to create synthetic children, you need to provide a Python class that adheres to a given interface (the word is italicized because Python has no explicit notion of interface, by that word we mean a given set of methods must be implemented by the Python class):

class SyntheticChildrenProvider:
   def __init__(self, valobj, internal_dict):
      this call should initialize the Python object using valobj as the variable to provide synthetic children for
   def num_children(self):
      this call should return the number of children that you want your object to have
   def get_child_index(self,name):
      this call should return the index of the synthetic child whose name is given as argument
   def get_child_at_index(self,index):
      this call should return a new LLDB SBValue object representing the child at the index given as argument
   def update(self):
      this call should be used to update the internal state of this Python object whenever the state of the variables in LLDB changes.[1]
      Also, this method is invoked before any other method in the interface.
   def has_children(self):
      this call should return True if this object might have children, and False if this object can be guaranteed not to have children.[2]
   def get_value(self):
      this call can return an SBValue to be presented as the value of the synthetic value under consideration.[3]

As a warning, exceptions that are thrown by python formatters are caught silently by LLDB and should be handled appropriately by the formatter itself. Being more specific, in case of exceptions, LLDB might assume that the given object has no children or it might skip printing some children, as they are printed one by one.

[1] This method is optional. Also, a boolean value must be returned (since lldb 3.1.0). If False is returned, then whenever the process reaches a new stop, this method will be invoked again to generate an updated list of the children for a given variable. Otherwise, if True is returned, then the value is cached and this method won’t be called again, effectively freezing the state of the value in subsequent stops. Beware that returning True incorrectly could show misleading information to the user.

[2] This method is optional (since lldb 3.2.0). While implementing it in terms of num_children is acceptable, implementors are encouraged to look for optimized coding alternatives whenever reasonable.

[3] This method is optional (since lldb 3.5.2). The SBValue you return here will most likely be a numeric type (int, float, …) as its value bytes will be used as-if they were the value of the root SBValue proper. As a shortcut for this, you can inherit from lldb.SBSyntheticValueProvider, and just define get_value as other methods are defaulted in the superclass as returning default no-children responses.

If a synthetic child provider supplies a special child named $$dereference$$ then it will be used when evaluating operator * and operator -> in the frame variable command and related SB API functions. It is possible to declare this synthetic child without including it in the range of children displayed by LLDB. For example, this subset of a synthetic children provider class would allow the synthetic value to be dereferenced without actually showing any synthetic children in the UI:

class SyntheticChildrenProvider:
    [...]
    def num_children(self):
        return 0
    def get_child_index(self, name):
        if name == '$$dereference$$':
            return 0
        return -1
    def get_child_at_index(self, index):
        if index == 0:
            return <valobj resulting from dereference>
        return None

For examples of how synthetic children are created, you are encouraged to look at examples/synthetic in the LLDB trunk. Please, be aware that the code in those files (except bitfield/) is legacy code and is not maintained. You may especially want to begin looking at this example to get a feel for this feature, as it is a very easy and well commented example.

The design pattern consistently used in synthetic providers shipping with LLDB is to use the __init__ to store the SBValue instance as a part of self. The update function is then used to perform the actual initialization. Once a synthetic children provider is written, one must load it into LLDB before it can be used. Currently, one can use the LLDB script command to type Python code interactively, or use the command script import fileName command to load Python code from a Python module (ordinary rules apply to importing modules this way). A third option is to type the code for the provider class interactively while adding it.

For example, let’s pretend we have a class Foo for which a synthetic children provider class Foo_Provider is available, in a Python module contained in file ~/Foo_Tools.py. The following interaction sets Foo_Provider as a synthetic children provider in LLDB:

(lldb) command script import ~/Foo_Tools.py
(lldb) type synthetic add Foo --python-class Foo_Tools.Foo_Provider
(lldb) frame variable a_foo
(Foo) a_foo = {
   x = 1
   y = "Hello world"
}

LLDB has synthetic children providers for a core subset of STL classes, both in the version provided by libstdcpp and by libcxx, as well as for several Foundation classes.

Synthetic children extend summary strings by enabling a new special variable: ${svar.

This symbol tells LLDB to refer expression paths to the synthetic children instead of the real ones. For instance,

(lldb) type summary add --expand -x "std::vector<" --summary-string "${svar%#} items"
(lldb) frame variable numbers
(std::vector<int>) numbers = 4 items {
   (int) [0] = 1
   (int) [1] = 12
   (int) [2] = 123
   (int) [3] = 1234
}

It’s important to mention that LLDB invokes the synthetic child provider before invoking the summary string provider, which allows the latter to have access to the actual displayable children. This applies to both inlined summary strings and python-based summary providers.

As a warning, when programmatically accessing the children or children count of a variable that has a synthetic child provider, notice that LLDB hides the actual raw children. For example, suppose we have a std::vector, which has an actual in-memory property __begin marking the beginning of its data. After the synthetic child provider is executed, the std::vector variable won’t show __begin as child anymore, even through the SB API. It will have instead the children calculated by the provider. In case the actual raw children are needed, a call to value.GetNonSyntheticValue() is enough to get a raw version of the value. It is import to remember this when implementing summary string providers, as they run after the synthetic child provider.

In some cases, if LLDB is unable to use the real object to get a child specified in an expression path, it will automatically refer to the synthetic children. While in summaries it is best to always use ${svar to make your intentions clearer, interactive debugging can benefit from this behavior, as in:

(lldb) frame variable numbers[0] numbers[1]
(int) numbers[0] = 1
(int) numbers[1] = 12

Unlike many other visualization features, however, the access to synthetic children only works when using frame variable, and is not supported in expression:

(lldb) expression numbers[0]
Error [IRForTarget]: Call to a function '_ZNSt33vector<int, std::allocator<int> >ixEm' that is not present in the target
error: Couldn't convert the expression to DWARF

The reason for this is that classes might have an overloaded operator [], or other special provisions and the expression command chooses to ignore synthetic children in the interest of equivalency with code you asked to have compiled from source.

Filters#

Filters are a solution to the display of complex classes. At times, classes have many member variables but not all of these are actually necessary for the user to see.

A filter will solve this issue by only letting the user see those member variables they care about. Of course, the equivalent of a filter can be implemented easily using synthetic children, but a filter lets you get the job done without having to write Python code.

For instance, if your class Foobar has member variables named A thru Z, but you only need to see the ones named B, H and Q, you can define a filter:

(lldb) type filter add Foobar --child B --child H --child Q
(lldb) frame variable a_foobar
(Foobar) a_foobar = {
   (int) B = 1
   (char) H = 'H'
   (std::string) Q = "Hello world"
}

Callback-based type matching#

Even though regular expression matching works well for the vast majority of data formatters (you normally know the name of the type you’re writing a formatter for), there are some cases where it’s useful to look at the type before deciding what formatter to apply.

As an example scenario, imagine we have a code generator that produces some classes that inherit from a common GeneratedObject class, and we have a summary function and a synthetic child provider that work for all GeneratedObject instances (they all follow the same pattern). However, there is no common pattern in the name of these classes, so we can’t register the formatter neither by name nor by regular expression.

In that case, you can write a recognizer function like this:

def is_generated_object(sbtype, internal_dict):
  for base in sbtype.get_bases_array():
    if base.GetName() == "GeneratedObject"
      return True
  return False

And pass this function to type summary add and type synthetic add using the flag --recognizer-function.

(lldb) type summary add --expand --python-function my_summary_function --recognizer-function is_generated_object
(lldb) type synthetic add --python-class my_child_provider --recognizer-function is_generated_object

Objective-C Dynamic Type Discovery#

When doing Objective-C development, you may notice that some of your variables come out as of type id (for instance, items extracted from NSArray). By default, LLDB will not show you the real type of the object. it can actually dynamically discover the type of an Objective-C variable, much like the runtime itself does when invoking a selector. In order to be shown the result of that discovery that, however, a special option to frame variable or expression is required: --dynamic-type.

--dynamic-type can have one of three values:

  • no-dynamic-values: the default, prevents dynamic type discovery

  • no-run-target: enables dynamic type discovery as long as running code on the target is not required

  • run-target: enables code execution on the target in order to perform dynamic type discovery

If you specify a value of either no-run-target or run-target, LLDB will detect the dynamic type of your variables and show the appropriate formatters for them. As an example:

(lldb) expr @"Hello"
(NSString *) $0 = 0x00000001048000b0 @"Hello"
(lldb) expr -d no-run @"Hello"
(__NSCFString *) $1 = 0x00000001048000b0 @"Hello"

Because LLDB uses a detection algorithm that does not need to invoke any functions on the target process, no-run-target is enough for this to work.

As a side note, the summary for NSString shown in the example is built right into LLDB. It was initially implemented through Python (the code is still available for reference at CFString.py). However, this is out of sync with the current implementation of the NSString formatter (which is a C++ function compiled into the LLDB core).

Categories#

Categories are a way to group related formatters. For instance, LLDB itself groups the formatters for the libstdc++ types in a category named gnu-libstdc++. Basically, categories act like containers in which to store formatters for a same library or OS release.

By default, several categories are created in LLDB:

  • default: this is the category where every formatter ends up, unless another category is specified

  • objc: formatters for basic and common Objective-C types that do not specifically depend on macOS

  • gnu-libstdc++: formatters for std::string, std::vector, std::list and std::map as implemented by libstdcpp

  • libcxx: formatters for std::string, std::vector, std::list and std::map as implemented by libcxx

  • system: truly basic types for which a formatter is required

  • AppKit: Cocoa classes

  • CoreFoundation: CF classes

  • CoreGraphics: CG classes

  • CoreServices: CS classes

  • VectorTypes: compact display for several vector types

If you want to use a custom category for your formatters, all the type … add provide a –category (-w) option, that names the category to add the formatter to. To delete the formatter, you then have to specify the correct category.

Categories can be in one of two states: enabled and disabled. A category is initially disabled, and can be enabled using the type category enable command. To disable an enabled category, the command to use is type category disable.

The order in which categories are enabled or disabled is significant, in that LLDB uses that order when looking for formatters. Therefore, when you enable a category, it becomes the second one to be searched (after default, which always stays on top of the list). The default categories are enabled in such a way that the search order is:

  • default

  • objc

  • CoreFoundation

  • AppKit

  • CoreServices

  • CoreGraphics

  • gnu-libstdc++

  • libcxx

  • VectorTypes

  • system

As said, gnu-libstdc++ and libcxx contain formatters for C++ STL data types. system contains formatters for char* and char[], which reflect the behavior of older versions of LLDB which had built-in formatters for these types. Because now these are formatters, you can even replace them with your own if so you wish.

There is no special command to create a category. When you place a formatter in a category, if that category does not exist, it is automatically created. For instance,

(lldb) type summary add Foobar --summary-string "a foobar" --category newcategory

automatically creates a (disabled) category named newcategory.

Another way to create a new (empty) category, is to enable it, as in:

(lldb) type category enable newcategory

However, in this case LLDB warns you that enabling an empty category has no effect. If you add formatters to the category after enabling it, they will be honored. But an empty category per se does not change the way any type is displayed. The reason the debugger warns you is that enabling an empty category might be a typo, and you effectively wanted to enable a similarly-named but not-empty category.

Finding Formatters 101#

Searching for a formatter (including formats, since lldb 3.4.0) given a variable goes through a rather intricate set of rules. Namely, what happens is that LLDB starts looking in each enabled category, according to the order in which they were enabled (latest enabled first). In each category, LLDB does the following:

  • If there is a formatter for the type of the variable, use it

  • If this object is a pointer, and there is a formatter for the pointee type that does not skip pointers, use it

  • If this object is a reference, and there is a formatter for the referred type that does not skip references, use it

  • If this object is an Objective-C class and dynamic types are enabled, look for a formatter for the dynamic type of the object. If dynamic types are disabled, or the search failed, look for a formatter for the declared type of the object

  • If this object’s type is a typedef, go through typedef hierarchy (LLDB might not be able to do this if the compiler has not emitted enough information. If the required information to traverse typedef hierarchies is missing, type cascading will not work. The clang compiler, part of the LLVM project, emits the correct debugging information for LLDB to cascade). If at any level of the hierarchy there is a valid formatter that can cascade, use it.

  • If everything has failed, repeat the above search, looking for regular expressions instead of exact matches

If any of those attempts returned a valid formatter to be used, that one is used, and the search is terminated (without going to look in other categories). If nothing was found in the current category, the next enabled category is scanned according to the same algorithm. If there are no more enabled categories, the search has failed.

Warning: previous versions of LLDB defined cascading to mean not only going through typedef chains, but also through inheritance chains. This feature has been removed since it significantly degrades performance. You need to set up your formatters for every type in inheritance chains to which you want the formatter to apply.