sg_vector_create, sg_vector_clear, sg_vector_resize, sg_vector_free, sg_vector_clone, sg_vector_clone_into, sg_vector_compute_diff, sg_prove_vector, sg_get_nelements, sg_free_stats_buf — statgrab vector management


#include "statgrab.h"
#include "vector.h"
struct sg_vector *fsfuncsg_vector_create(block_size,  
size_t block_size;
size_t alloc_count;
size_t initial_used;
const sg_vector_init_info * const info;
void fsfuncsg_vector_clear(vector); 
struct sg_vector *vector;
struct sg_vector *fsfuncsg_vector_resize(vector); 
struct sg_vector *vector;
void fsfuncsg_vector_free(vector); 
struct sg_vector *vector;
struct sg_vector *fsfuncsg_vector_clone(src); 
const struct sg_vector *src;
sg_error fsfuncsg_vector_clone_into(dest,  
struct sg_vector **dest;
const struct sg_vector *src;
sg_error fsfuncsg_vector_compute_diff(dest,  
struct sg_vector **dest;
const struct sg_vector *cur_vector;
const struct sg_vector *last_vector;
sg_error fsfuncsg_prove_vector(vec); 
const struct sg_vector *vec;
size_t fsfuncsg_get_nelements(data); 
const void *data;
sg_error fsfuncsg_free_stats_buf(data); 
void *data;


sg_vector_create() allocates and initialises a new statgrab vector with initial_used elements ready for use. Space for alloc_count elements is initially allocated (to avoid too many calls to realloc() during later sg_vector_resize() calls). The value of block_size must be a power of 2, it's rounded up to the next power of 2 when it's not. If alloc_count is not a multiple of block_size, it's rounded up to the next multiple of block_size. It returns a pointer to the newly created vector.

sg_vector_clear() destroys all elements contained in the given vector. In opposite to sg_vector_resize( x, 0 ) the allocated size of the vector remains untouched.

sg_vector_resize() increases or decreases the amount of allocated elements in the specified vector. The amount of allocated elements is always a multiple of the intialisation parameter block_size. In the special case, sg_vector_resize() is called with 0 in argument new_count, the vector is freed after all vector elements had been destroyed. It returns the pointer to the resized vector.

sg_vector_free() destroys all vector elements and deallocates the storage belonging to the given vector.

sg_vector_clone() clones all elements of the given vector into a new vector created with the same specification as the referenced one. It returns a pointer to the cloned vector.

sg_vector_clone_into() clones all elements of the given source vector into the given target vector. The target vector must be created for the same element data type as the source vector. It returns an error code != to SG_ERROR_NONE if something went wrong.

sg_vector_compute_diff() computes a difference vector between the vector containing current statistics and another vector containing older statistics. If an element exists in the current vector but not in the opposite one, it's cloned into the result vector. If an element exists only in the opposite vector, it doesn't appear in the target vector. sg_vector_compute_diff() returns an error code != to SG_ERROR_NONE if something went wrong.

sg_prove_vector() proves whether a pointer to a vector really points to a vector. In case the given vector pointer points to corrupted data, the program is aborted. When sg_prove_vector() returns, it returns SG_ERROR_NONE.

sg_get_nelements() returns the number of elements the given data area, encompasses by a statgrab vector, contains. The vector head is internally calculated from the given pointer to the first vector element.

sg_free_stats_buf() frees the vector emcompassing the given data area.


Except sg_get_nelements() and sg_free_stats_buf() none of above functions can be called from outside of the libstatgrab sources. The documented structures and APIs may change without warning. The description of all other API is intended to be read from libstatgrab developers only.

Each vector is created from two elements: the vector information and the list of elements:

template <class T, class Impl>
struct sg_vector {
        size_t used_count;
        size_t alloc_count;
        size_t block_shift;
        Impl vector_implementation;
        T elements[alloc_count];

Of course, it is not valid C, so being tricky was the solution:

typedef struct sg_vector {
	size_t used_count;
	size_t alloc_count;
	size_t block_shift;
	struct sg_vector_init_info info;
} sg_vector;

struct sg_vector_size_helper {
	struct sg_vector v;
	long long ll;

#define VECTOR_SIZE offsetof(struct sg_vector_size_helper,ll)

/* Return the data ptr of a vector */
#define VECTOR_DATA(vector) \
	(vector ? (void *)(((char *)vector)+VECTOR_SIZE) : NULL)

#define VECTOR_ADDR_ARITH(ptr) \
	(sg_vector *)(((char *)(ptr))-VECTOR_SIZE)
/* Return the vector for a data */
#define VECTOR_ADDRESS(ptr) \
	((ptr) ? (SG_ERROR_NONE == sg_prove_vector(VECTOR_ADDR_ARITH(ptr)) ? VECTOR_ADDR_ARITH(ptr) : NULL ) : NULL)

This also allows user functions as sg_get_nelements() and sg_free_stats_buf() to switch easily between the vector structure and the content.

The vector specialisation structure

As mentioned, the vector implementation uses strategies from the object oriented programming concept named "polymorphism". A vector is described by a small object containing inherent attributes like element size and a bunch of "virtual methods" to do element related tasks like initialising or destroying elements.

typedef void (*vector_init_function)(void *item);
typedef sg_error (*vector_copy_function)(const void *src, void *dst);
typedef sg_error (*vector_compute_diff_function)(void *dst, const void *src);
typedef int (*vector_compare_function)(const void *a, const void *b);
typedef void (*vector_destroy_function)(void *item);

struct sg_vector_init_info {
        size_t item_size;
        vector_init_function init_fn;
        vector_copy_function copy_fn;
        vector_compute_diff_function compute_diff_fn;
        vector_compare_function compare_fn;
        vector_destroy_function destroy_fn;

The instances of struct sg_vector_init_info are conceptional statically initialised by using either the preprocessor macro VECTOR_INIT_INFO_FULL_INIT(type) or VECTOR_INIT_INFO_EMPTY_INIT(type). Here're some examples to demonstrate how it's meant:

Working with vectors

To simplify the working with the vector management functions, some preprocessor macros are available. They are shown here as if they were functions to ease understanding.

struct sg_vector *fsfuncVECTOR_CREATE(type,  
identifier type;
size_t block_size;
void fsfuncVECTOR_CLEAR(vector); 
struct sg_vector *vector;
struct sg_vector *fsfuncVECTOR_CREATE_OR_RESIZE(vector,  
struct sg_vector *vector;
size_t new_count;
identifier type;
void fsfuncVECTOR_UPDATE(vectorptr,  
struct sg_vector **vectorptr;
size_t new_count;
datatype *data;
identifier datatype;
void fsfuncVECTOR_ITEM_COUNT(vector); 
struct sg_vector *vector;

VECTOR_CREATE() calls sg_vector_create() with alloc_count = block_size and initial_used = 0 using the vector specialisation type##_vector_init_info.

VECTOR_CLEAR() simply calls sg_vector_clear(). This macro exists only for conformity.

VECTOR_CREATE_OR_RESIZE() calls sg_vector_create() when the given vector pointer points to NULL or sg_vector_resize() otherwise. The result of the appropriate function is returned.

VECTOR_UPDATE() calls VECTOR_CREATE_OR_RESIZE() and sets data to the first element of the resulting vector when a non-NULL pointer got, to NULL otherwise. When VECTOR_CREATE_OR_RESIZE() returns a NULL pointer and new_count is not equal to 0 (zero), the intructions from the macro VECTOR_UPDATE_ERROR_CLEANUP are executed to cleanup before returning from current subroutine with the error which has been occurred.

VECTOR_ITEM_COUNT() returns 0 for a non-existing vector (vector == 0) and the number of containing elements otherwise.