diff options
Diffstat (limited to 'libbb/xrealloc_vector.c')
| -rw-r--r-- | libbb/xrealloc_vector.c | 14 |
1 files changed, 12 insertions, 2 deletions
diff --git a/libbb/xrealloc_vector.c b/libbb/xrealloc_vector.c index ccc961b7b..342ae536e 100644 --- a/libbb/xrealloc_vector.c +++ b/libbb/xrealloc_vector.c @@ -14,7 +14,6 @@ * #define magic packed two parameters into one: * sizeof = sizeof_and_shift >> 8 * shift = (sizeof_and_shift) & 0xff - * (TODO: encode "I want it zeroed" in lowest bit?) * * Lets say shift = 4. 1 << 4 == 0x10. * If idx == 0, 0x10, 0x20 etc, vector[] is resized to next higher @@ -23,14 +22,25 @@ * * In other words: after xrealloc_vector(v, 4, idx) it's ok to use * at least v[idx] and v[idx+1], for all idx values. + * + * New elements are zeroed out, but only if realloc was done + * (not on every call). You can depend on v[idx] and v[idx+1] being + * zeroed out if you use it like this: + * v = xrealloc_vector(v, 4, idx); + * v[idx].some_fields = ...; - the rest stays 0/NULL + * idx++; + * If you do not advance idx like above, you should be more careful. + * Next call to xrealloc_vector(v, 4, idx) may or may not zero out v[idx]. */ void* FAST_FUNC xrealloc_vector_helper(void *vector, unsigned sizeof_and_shift, int idx) { int mask = 1 << (uint8_t)sizeof_and_shift; if (!(idx & (mask - 1))) { - sizeof_and_shift >>= 8; + sizeof_and_shift >>= 8; /* sizeof(vector[0]) */ vector = xrealloc(vector, sizeof_and_shift * (idx + mask + 1)); + vector += idx; + memset(vector, 0, sizeof_and_shift * (mask + 1)); } return vector; } |