C memmove equivalent of for loop - segfault

Question

I'm implementing ring buffer with dynamic resizing. When tail is behind head, data from the end of the buffer has to be moved to new end of buffer after resize. In order to do so I made following code:

memmove(self->broadcaster.events+self->broadcaster.events_head+self->broadcaster.events_size,
              self->broadcaster.events+self->broadcaster.events_head,
              self->broadcaster.events_size-self->broadcaster.events_head);

where self->broadcaster.events_size is old size (new_size/2). Unfortunately it results in segmentation fault. I thought it's quivalent of this code:

for (i = 0 ; i < self->broadcaster.events_size - self->broadcaster.events_head ; ++i)
        self->broadcaster.events[self->broadcaster.events_size+self->broadcaster.events_head+i]=
            self->broadcaster.events[self->broadcaster.events_head+i];

but such naive for-loop implementation works properly so it seems I don't know how to use memmove properly. How those two pieces of code are different?

Why does data have to be moved in a ring buffer? Is this an XY question? — Weather Vane, Sep 15 '18 at 23:45
@WeatherVane when ring buffer turns out to be too short often, I'm increasing its size dynamically. If buffer is not continuos (it's looped) then after reallocation of buffer its rear part has to be moved to new end in order to make such ring buffer "continuos" in sense of ring buffer. To depict it: `OO---OO` --(realloc)-> `OO---OO-------` --(memmove)-> `OO----------OO`. — Lapsio, Sep 15 '18 at 23:48

score 1 · Accepted Answer · 2018-09-16T08:40:35.027

They're only equivalent if sizeof(*self->broadcaster.events) == 1.

For clarity, I've replaced self->broadcaster with b and events with e and added some whitespace to your code.

memmove(b.e + b.e_head + b.e_size, b.e + b.e_head, b.e_size - b.e_head);

will copy just b.e_size - b.e_head bytes, and the loop:

for (i = 0 ; i < b.e_size - b.e_head ; ++i)
        b.e[b.e_size + b.e_head + i] = b.e[b.e_head + i];

will copy (b.e_size - b.e_head) * sizeof *b.e bytes, because each b.e[...] = b.e[...] assignment is moving through sizeof *b.e bytes, and each ++i is advancing the address of b.e[... + i] by sizeof *b.e bytes.

You'll be best served if you define a macro:

#define MOVE(dst, src, count)  memmove((dst), (src), (count) * sizeof *(src))

and use it instead of memmove.

But you can of course just change the last argument of memmove to

(self->broadcaster.events_size-self->broadcaster.events_head)*sizeof*self->broadcaster.events

C memmove equivalent of for loop - segfault

1 Answers1