Invocation
Quick start
Parsing from local file:
#include <fstream>
#include <kaitai/kaitaistream.h>
std::ifstream is("path/to/local/file.dat", std::ifstream::binary);
kaitai::kstream ks(&is);
example_t data(&ks);
Parsing from std::string
:
#include <fstream>
#include <kaitai/kaitaistream.h>
#include <sstream>
std::string buf;
std::istringstream is(buf);
kaitai::kstream ks(&is);
example_t data(&ks);
Auto-read
By default, invoking constructor with a stream argument assumes that you want to run parsing process and populate object’s fields with the data read from the stream:
example_t data(&ks);
data.some_attribute(); // already populated and available
However, sometimes you want more control and want to trigger that
process manually. In that case, you can supply --no-auto-read
argument for kaitai-struct-compiler, and you’ll have to trigger
reading manually using _read()
invocation:
example_t data(&ks);
data.some_attribute(); // not yet populated, may contain random garbage
data._read();
data.some_attribute(); // populated and available
There are several reasons why you’d want to separate object creation and object population:
-
Obviously, if you’re using read-write mode with the intent to populate the object manually and call
_write()
afterwards to serialize it. In this use-case, you probably won’t call_read()
at all. -
You want to reuse the same object and thus want to repeatedly call
_read()
on several times manually.
Ownership model
In all modes, Kaitai Struct follows the same ownership model:
-
If something is created during the parsing process, it belongs to the enclosing object which represents the user type. As soon as enclosing object will be deleted, it will take care of deletion (and cleanup) of all owned attributes.
-
If something is assigned to a user type using
set…()
methods for serialization process, afterset…()
is completed, user type assumes ownership of everything you’ve passed throughset…()
. -
Everything else passed in a constructor and/or any other invocations, is not owned by user type (and will not be cleaned up automatically), namely:
-
root object reference/pointer
-
parent object reference/pointer
-
IO object
-
type parameters
-
To illustrate these principles, using the following .ksy spec:
meta:
id: example
seq:
- id: foo
type: block
types:
block:
seq:
- id: bar
type: u1
You would always do reading this way:
// On stack
{
kaitai::kstruct in_stream(...); // belongs here
example_t example(&in_stream); // belongs here
// assumes auto-read; in case of --no-auto-read, add:
// example._read();
example_t::block_t* foo = example->foo();
// does not belong here, belongs to user type
// can use `foo` here:
int foo_bar = foo->bar();
// but must not delete it:
// delete foo; // ILLEGAL: will result in double free
// example & in_stream get deleted here as they go out of scope
}
// On heap
{
kaitai::kstruct* in_stream = new kaitai::kstruct(...); // belongs here
example_t* example = new example_t(in_stream); // belongs here
// assumes auto-read; in case of --no-auto-read, add:
// example->_read();
example_t::block_t* foo = example->foo();
// does not belong here, belongs to user type
// can use `foo` here:
int foo_bar = foo->bar();
// but must not delete it:
// delete foo; // ILLEGAL: will result in double free
// delete in reverse order to order of creation
delete example;
// also deletes `foo`, so `foo` should not be used after this point
// int foo_bar_2 = foo->bar(); // ILLEGAL: foo pointer is already deleted
delete in_stream;
}
Pointers model
TODO: raw pointers, unique+raw pointers
Primitive type mapping
Mapping KS types to C++ is pretty straight-forward:
type |
C++ type |
---|---|
no type |
std::string |
|
uint8_t |
|
uint16_t |
|
uint32_t |
|
uint64_t |
|
int8_t |
|
int16_t |
|
int32_t |
|
int64_t |
|
std::string |
Note that both byte arrays and strings are mapped to std::string
—
that’s because when we store byte array, we need something that would be
able to both hold the byte buffer and store it’s length (or at least
able to derive it).
String encoding
There’s no universal agreement on dealing with encodings in C++, so KS allows you to choose one of the few popular approaches. You can choose how to deal with string encoding using a compile-time define.
-
KS_STR_ENCODING_NONE
: Ignore encodings at all. In this mode, all string parsing operations just ignore any encoding specifications and pass raw bytes as a string to application. Note that in some cases it might break some .ksy files that actually depend on string being properly decoded / converted. -
Convert all incoming byte streams into strings in a single, one-size-fits-all encoding (for example, UTF8, as suggested by UTF8 Everywhere Manifesto). Since there’s no universal way to do it, KS would use one of platform-dependent ways (which can be also enforced by specifying specific defines):
-
KS_STR_ENCODING_ICONV
: Use POSIXiconv
library — usually preinstalled (or included in libc) on all POSIX systems, can be linked as external library on most other systems (i.e. Windows) -
(not implemented yet) Use Windows API functions MultiByteToWideChar and WideCharToMultiByte — obviously, available only on Windows platform
-
(not implemented yet) Use ICU library
-
Null values
In certain cases, namely when using if
with an expression that will be
false, a certain attribute won’t be parsed. For example:
seq:
- id: foo
type: u1
- id: bar
type: u1
if: foo == 42
If foo
is not 42, then an unsigned 1-byte integer bar
won’t be
parsed. By general convention, Kaitai Struct makes sure that bar
is
equal to a null
value, to be able to distinguish such a situation (as
opposed to having some value). However, it’s not possible to do so for
many primitive (non-reference) types in C++. In the example above, bar
will have type uint8_t
, and assigning null
to it would just set it
to 0, thus we won’t be able to distinguish a situation when bar
was
read and we’ve got 0, and bar
wasn’t read.
To work around this situation, Kaitai Struct generates special null
checking methods for every attribute that can be null:
_is_null_ATTRIBUTE
, where ATTRIBUTE
is the name of the attribute.
Thus, the proper way to use such nullable values is something like:
if (!r->_is_null_bar()) {
uint8_t bar = r->bar();
// `bar` is defined, use `bar` here
} else {
// `bar` is null because of failed `if` comparison
// note that accessing r->bar() will return an uninitialized value
// (i.e. random garbage)
}