Commit 452d01c8 authored by Yolan Uyttenhove's avatar Yolan Uyttenhove
Browse files

fix stringop-overread errors

parent 60e80004
......@@ -226,7 +226,7 @@ struct io_props {
* Do not call this function directly. Use the macro defined above.
*/
INLINE static struct io_props io_make_input_field_(
const char name[FIELD_BUFFER_SIZE], enum IO_DATA_TYPE type, int dimension,
const char *name, enum IO_DATA_TYPE type, int dimension,
enum DATA_IMPORTANCE importance, enum unit_conversion_factor units,
char* field, size_t partSize, const float default_value) {
struct io_props r;
......@@ -273,9 +273,9 @@ INLINE static struct io_props io_make_input_field_(
* Do not call this function directly. Use the macro defined above.
*/
INLINE static struct io_props io_make_output_field_(
const char name[FIELD_BUFFER_SIZE], enum IO_DATA_TYPE type, int dimension,
const char *name, enum IO_DATA_TYPE type, int dimension,
enum unit_conversion_factor units, float a_exponent, char* field,
size_t partSize, const char description[DESCRIPTION_BUFFER_SIZE]) {
size_t partSize, const char *description) {
struct io_props r;
bzero(&r, sizeof(struct io_props));
......@@ -324,11 +324,11 @@ INLINE static struct io_props io_make_output_field_(
* Do not call this function directly. Use the macro defined above.
*/
INLINE static struct io_props io_make_output_field_convert_part_INT(
const char name[FIELD_BUFFER_SIZE], enum IO_DATA_TYPE type, int dimension,
const char *name, enum IO_DATA_TYPE type, int dimension,
enum unit_conversion_factor units, float a_exponent, size_t partSize,
const struct part* parts, const struct xpart* xparts,
conversion_func_part_int functionPtr,
const char description[DESCRIPTION_BUFFER_SIZE]) {
const char *description) {
struct io_props r;
bzero(&r, sizeof(struct io_props));
......@@ -370,11 +370,11 @@ INLINE static struct io_props io_make_output_field_convert_part_INT(
* Do not call this function directly. Use the macro defined above.
*/
INLINE static struct io_props io_make_output_field_convert_part_FLOAT(
const char name[FIELD_BUFFER_SIZE], enum IO_DATA_TYPE type, int dimension,
const char *name, enum IO_DATA_TYPE type, int dimension,
enum unit_conversion_factor units, float a_exponent, size_t partSize,
const struct part* parts, const struct xpart* xparts,
conversion_func_part_float functionPtr,
const char description[DESCRIPTION_BUFFER_SIZE]) {
const char *description) {
struct io_props r;
bzero(&r, sizeof(struct io_props));
......@@ -416,11 +416,11 @@ INLINE static struct io_props io_make_output_field_convert_part_FLOAT(
* Do not call this function directly. Use the macro defined above.
*/
INLINE static struct io_props io_make_output_field_convert_part_DOUBLE(
const char name[FIELD_BUFFER_SIZE], enum IO_DATA_TYPE type, int dimension,
const char *name, enum IO_DATA_TYPE type, int dimension,
enum unit_conversion_factor units, float a_exponent, size_t partSize,
const struct part* parts, const struct xpart* xparts,
conversion_func_part_double functionPtr,
const char description[DESCRIPTION_BUFFER_SIZE]) {
const char *description) {
struct io_props r;
bzero(&r, sizeof(struct io_props));
......@@ -462,11 +462,11 @@ INLINE static struct io_props io_make_output_field_convert_part_DOUBLE(
* Do not call this function directly. Use the macro defined above.
*/
INLINE static struct io_props io_make_output_field_convert_part_LONGLONG(
const char name[FIELD_BUFFER_SIZE], enum IO_DATA_TYPE type, int dimension,
const char *name, enum IO_DATA_TYPE type, int dimension,
enum unit_conversion_factor units, float a_exponent, size_t partSize,
const struct part* parts, const struct xpart* xparts,
conversion_func_part_long_long functionPtr,
const char description[DESCRIPTION_BUFFER_SIZE]) {
const char *description) {
struct io_props r;
bzero(&r, sizeof(struct io_props));
......@@ -516,10 +516,10 @@ INLINE static struct io_props io_make_output_field_convert_part_LONGLONG(
* Do not call this function directly. Use the macro defined above.
*/
INLINE static struct io_props io_make_output_field_convert_gpart_INT(
const char name[FIELD_BUFFER_SIZE], enum IO_DATA_TYPE type, int dimension,
const char *name, enum IO_DATA_TYPE type, int dimension,
enum unit_conversion_factor units, float a_exponent, size_t gpartSize,
const struct gpart* gparts, conversion_func_gpart_int functionPtr,
const char description[DESCRIPTION_BUFFER_SIZE]) {
const char *description) {
struct io_props r;
bzero(&r, sizeof(struct io_props));
......@@ -559,10 +559,10 @@ INLINE static struct io_props io_make_output_field_convert_gpart_INT(
* Do not call this function directly. Use the macro defined above.
*/
INLINE static struct io_props io_make_output_field_convert_gpart_FLOAT(
const char name[FIELD_BUFFER_SIZE], enum IO_DATA_TYPE type, int dimension,
const char *name, enum IO_DATA_TYPE type, int dimension,
enum unit_conversion_factor units, float a_exponent, size_t gpartSize,
const struct gpart* gparts, conversion_func_gpart_float functionPtr,
const char description[DESCRIPTION_BUFFER_SIZE]) {
const char *description) {
struct io_props r;
bzero(&r, sizeof(struct io_props));
......@@ -602,10 +602,10 @@ INLINE static struct io_props io_make_output_field_convert_gpart_FLOAT(
* Do not call this function directly. Use the macro defined above.
*/
INLINE static struct io_props io_make_output_field_convert_gpart_DOUBLE(
const char name[FIELD_BUFFER_SIZE], enum IO_DATA_TYPE type, int dimension,
const char *name, enum IO_DATA_TYPE type, int dimension,
enum unit_conversion_factor units, float a_exponent, size_t gpartSize,
const struct gpart* gparts, conversion_func_gpart_double functionPtr,
const char description[DESCRIPTION_BUFFER_SIZE]) {
const char *description) {
struct io_props r;
bzero(&r, sizeof(struct io_props));
......@@ -645,10 +645,10 @@ INLINE static struct io_props io_make_output_field_convert_gpart_DOUBLE(
* Do not call this function directly. Use the macro defined above.
*/
INLINE static struct io_props io_make_output_field_convert_gpart_LONGLONG(
const char name[FIELD_BUFFER_SIZE], enum IO_DATA_TYPE type, int dimension,
const char *name, enum IO_DATA_TYPE type, int dimension,
enum unit_conversion_factor units, float a_exponent, size_t gpartSize,
const struct gpart* gparts, conversion_func_gpart_long_long functionPtr,
const char description[DESCRIPTION_BUFFER_SIZE]) {
const char *description) {
struct io_props r;
bzero(&r, sizeof(struct io_props));
......@@ -697,10 +697,10 @@ INLINE static struct io_props io_make_output_field_convert_gpart_LONGLONG(
* Do not call this function directly. Use the macro defined above.
*/
INLINE static struct io_props io_make_output_field_convert_spart_INT(
const char name[FIELD_BUFFER_SIZE], enum IO_DATA_TYPE type, int dimension,
const char *name, enum IO_DATA_TYPE type, int dimension,
enum unit_conversion_factor units, float a_exponent, size_t spartSize,
const struct spart* sparts, conversion_func_spart_int functionPtr,
const char description[DESCRIPTION_BUFFER_SIZE]) {
const char *description) {
struct io_props r;
bzero(&r, sizeof(struct io_props));
......@@ -740,10 +740,10 @@ INLINE static struct io_props io_make_output_field_convert_spart_INT(
* Do not call this function directly. Use the macro defined above.
*/
INLINE static struct io_props io_make_output_field_convert_spart_FLOAT(
const char name[FIELD_BUFFER_SIZE], enum IO_DATA_TYPE type, int dimension,
const char *name, enum IO_DATA_TYPE type, int dimension,
enum unit_conversion_factor units, float a_exponent, size_t spartSize,
const struct spart* sparts, conversion_func_spart_float functionPtr,
const char description[DESCRIPTION_BUFFER_SIZE]) {
const char *description) {
struct io_props r;
bzero(&r, sizeof(struct io_props));
......@@ -783,10 +783,10 @@ INLINE static struct io_props io_make_output_field_convert_spart_FLOAT(
* Do not call this function directly. Use the macro defined above.
*/
INLINE static struct io_props io_make_output_field_convert_spart_DOUBLE(
const char name[FIELD_BUFFER_SIZE], enum IO_DATA_TYPE type, int dimension,
const char *name, enum IO_DATA_TYPE type, int dimension,
enum unit_conversion_factor units, float a_exponent, size_t spartSize,
const struct spart* sparts, conversion_func_spart_double functionPtr,
const char description[DESCRIPTION_BUFFER_SIZE]) {
const char *description) {
struct io_props r;
bzero(&r, sizeof(struct io_props));
......@@ -826,10 +826,10 @@ INLINE static struct io_props io_make_output_field_convert_spart_DOUBLE(
* Do not call this function directly. Use the macro defined above.
*/
INLINE static struct io_props io_make_output_field_convert_spart_LONGLONG(
const char name[FIELD_BUFFER_SIZE], enum IO_DATA_TYPE type, int dimension,
const char *name, enum IO_DATA_TYPE type, int dimension,
enum unit_conversion_factor units, float a_exponent, size_t spartSize,
const struct spart* sparts, conversion_func_spart_long_long functionPtr,
const char description[DESCRIPTION_BUFFER_SIZE]) {
const char *description) {
struct io_props r;
bzero(&r, sizeof(struct io_props));
......@@ -878,10 +878,10 @@ INLINE static struct io_props io_make_output_field_convert_spart_LONGLONG(
* Do not call this function directly. Use the macro defined above.
*/
INLINE static struct io_props io_make_output_field_convert_bpart_INT(
const char name[FIELD_BUFFER_SIZE], enum IO_DATA_TYPE type, int dimension,
const char *name, enum IO_DATA_TYPE type, int dimension,
enum unit_conversion_factor units, float a_exponent, size_t bpartSize,
const struct bpart* bparts, conversion_func_bpart_int functionPtr,
const char description[DESCRIPTION_BUFFER_SIZE]) {
const char *description) {
struct io_props r;
bzero(&r, sizeof(struct io_props));
......@@ -921,10 +921,10 @@ INLINE static struct io_props io_make_output_field_convert_bpart_INT(
* Do not call this function directly. Use the macro defined above.
*/
INLINE static struct io_props io_make_output_field_convert_bpart_FLOAT(
const char name[FIELD_BUFFER_SIZE], enum IO_DATA_TYPE type, int dimension,
const char *name, enum IO_DATA_TYPE type, int dimension,
enum unit_conversion_factor units, float a_exponent, size_t bpartSize,
const struct bpart* bparts, conversion_func_bpart_float functionPtr,
const char description[DESCRIPTION_BUFFER_SIZE]) {
const char *description) {
struct io_props r;
bzero(&r, sizeof(struct io_props));
......@@ -964,10 +964,10 @@ INLINE static struct io_props io_make_output_field_convert_bpart_FLOAT(
* Do not call this function directly. Use the macro defined above.
*/
INLINE static struct io_props io_make_output_field_convert_bpart_DOUBLE(
const char name[FIELD_BUFFER_SIZE], enum IO_DATA_TYPE type, int dimension,
const char *name, enum IO_DATA_TYPE type, int dimension,
enum unit_conversion_factor units, float a_exponent, size_t bpartSize,
const struct bpart* bparts, conversion_func_bpart_double functionPtr,
const char description[DESCRIPTION_BUFFER_SIZE]) {
const char *description) {
struct io_props r;
bzero(&r, sizeof(struct io_props));
......@@ -1007,10 +1007,10 @@ INLINE static struct io_props io_make_output_field_convert_bpart_DOUBLE(
* Do not call this function directly. Use the macro defined above.
*/
INLINE static struct io_props io_make_output_field_convert_bpart_LONGLONG(
const char name[FIELD_BUFFER_SIZE], enum IO_DATA_TYPE type, int dimension,
const char *name, enum IO_DATA_TYPE type, int dimension,
enum unit_conversion_factor units, float a_exponent, size_t bpartSize,
const struct bpart* bparts, conversion_func_bpart_long_long functionPtr,
const char description[DESCRIPTION_BUFFER_SIZE]) {
const char *description) {
struct io_props r;
bzero(&r, sizeof(struct io_props));
......@@ -1059,10 +1059,10 @@ INLINE static struct io_props io_make_output_field_convert_bpart_LONGLONG(
* Do not call this function directly. Use the macro defined above.
*/
INLINE static struct io_props io_make_output_field_convert_sink_INT(
const char name[FIELD_BUFFER_SIZE], enum IO_DATA_TYPE type, int dimension,
const char *name, enum IO_DATA_TYPE type, int dimension,
enum unit_conversion_factor units, float a_exponent, size_t sinkSize,
const struct sink* sinks, conversion_func_sink_int functionPtr,
const char description[DESCRIPTION_BUFFER_SIZE]) {
const char *description) {
struct io_props r;
bzero(&r, sizeof(struct io_props));
......@@ -1102,10 +1102,10 @@ INLINE static struct io_props io_make_output_field_convert_sink_INT(
* Do not call this function directly. Use the macro defined above.
*/
INLINE static struct io_props io_make_output_field_convert_sink_FLOAT(
const char name[FIELD_BUFFER_SIZE], enum IO_DATA_TYPE type, int dimension,
const char *name, enum IO_DATA_TYPE type, int dimension,
enum unit_conversion_factor units, float a_exponent, size_t sinkSize,
const struct sink* sinks, conversion_func_sink_float functionPtr,
const char description[DESCRIPTION_BUFFER_SIZE]) {
const char *description) {
struct io_props r;
bzero(&r, sizeof(struct io_props));
......@@ -1145,10 +1145,10 @@ INLINE static struct io_props io_make_output_field_convert_sink_FLOAT(
* Do not call this function directly. Use the macro defined above.
*/
INLINE static struct io_props io_make_output_field_convert_sink_DOUBLE(
const char name[FIELD_BUFFER_SIZE], enum IO_DATA_TYPE type, int dimension,
const char *name, enum IO_DATA_TYPE type, int dimension,
enum unit_conversion_factor units, float a_exponent, size_t sinkSize,
const struct sink* sinks, conversion_func_sink_double functionPtr,
const char description[DESCRIPTION_BUFFER_SIZE]) {
const char *description) {
struct io_props r;
bzero(&r, sizeof(struct io_props));
......@@ -1188,10 +1188,10 @@ INLINE static struct io_props io_make_output_field_convert_sink_DOUBLE(
* Do not call this function directly. Use the macro defined above.
*/
INLINE static struct io_props io_make_output_field_convert_sink_LONGLONG(
const char name[FIELD_BUFFER_SIZE], enum IO_DATA_TYPE type, int dimension,
const char *name, enum IO_DATA_TYPE type, int dimension,
enum unit_conversion_factor units, float a_exponent, size_t sinkSize,
const struct sink* sinks, conversion_func_sink_long_long functionPtr,
const char description[DESCRIPTION_BUFFER_SIZE]) {
const char *description) {
struct io_props r;
bzero(&r, sizeof(struct io_props));
......
Supports Markdown
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment