Is it possible to have zlib read from and write to the same memory buffer?

Question

I have a character buffer that I would like to compress in place. Right now I have it set up so there are two buffers and zlib's deflate reads from the input buffer and writes to the output buffer. Then I have to change the input buffer pointer to point to the output buffer and free the old input buffer. This seems like an unnecessary amount of allocation. Since zlib is compressing, the next_out pointer should always lag behind the next_in pointer. Anyway, I can't find enough documentation to verify this and was hoping someone had some experience with this. Thanks for your time!

Buntline

Answer 1

It can be done, with some care. The routine below does it. Not all data is compressible, so you have to handle the case where the output data catches up with the input data. It takes a lot of incompressible data, but it can happen (see comments in code), in which case you have to allocate a buffer to temporarily hold the remaining input.

/* Compress buf[0..len-1] in place into buf[0..*max-1].  *max must be greater
   than or equal to len.  Return Z_OK on success, Z_BUF_ERROR if *max is not
   enough output space, Z_MEM_ERROR if there is not enough memory, or
   Z_STREAM_ERROR if *strm is corrupted (e.g. if it wasn't initialized or if it
   was inadvertently written over).  If Z_OK is returned, *max is set to the
   actual size of the output.  If Z_BUF_ERROR is returned, then *max is
   unchanged and buf[] is filled with *max bytes of uncompressed data (which is
   not all of it, but as much as would fit).

   Incompressible data will require more output space than len, so max should
   be sufficiently greater than len to handle that case in order to avoid a
   Z_BUF_ERROR. To assure that there is enough output space, max should be
   greater than or equal to the result of deflateBound(strm, len).

   strm is a deflate stream structure that has already been successfully
   initialized by deflateInit() or deflateInit2().  That structure can be
   reused across multiple calls to deflate_inplace().  This avoids unnecessary
   memory allocations and deallocations from the repeated use of deflateInit()
   and deflateEnd(). */
int deflate_inplace(z_stream *strm, unsigned char *buf, unsigned len,
                    unsigned *max)
{
    int ret;                    /* return code from deflate functions */
    unsigned have;              /* number of bytes in temp[] */
    unsigned char *hold;        /* allocated buffer to hold input data */
    unsigned char temp[11];     /* must be large enough to hold zlib or gzip
                                   header (if any) and one more byte -- 11
                                   works for the worst case here, but if gzip
                                   encoding is used and a deflateSetHeader()
                                   call is inserted in this code after the
                                   deflateReset(), then the 11 needs to be
                                   increased to accomodate the resulting gzip
                                   header size plus one */

    /* initialize deflate stream and point to the input data */
    ret = deflateReset(strm);
    if (ret != Z_OK)
        return ret;
    strm->next_in = buf;
    strm->avail_in = len;

    /* kick start the process with a temporary output buffer -- this allows
       deflate to consume a large chunk of input data in order to make room for
       output data there */
    if (*max < len)
        *max = len;
    strm->next_out = temp;
    strm->avail_out = sizeof(temp) > *max ? *max : sizeof(temp);
    ret = deflate(strm, Z_FINISH);
    if (ret == Z_STREAM_ERROR)
        return ret;

    /* if we can, copy the temporary output data to the consumed portion of the
       input buffer, and then continue to write up to the start of the consumed
       input for as long as possible */
    have = strm->next_out - temp;
    if (have <= (strm->avail_in ? len - strm->avail_in : *max)) {
        memcpy(buf, temp, have);
        strm->next_out = buf + have;
        have = 0;
        while (ret == Z_OK) {
            strm->avail_out = strm->avail_in ? strm->next_in - strm->next_out :
                                               (buf + *max) - strm->next_out;
            ret = deflate(strm, Z_FINISH);
        }
        if (ret != Z_BUF_ERROR || strm->avail_in == 0) {
            *max = strm->next_out - buf;
            return ret == Z_STREAM_END ? Z_OK : ret;
        }
    }

    /* the output caught up with the input due to insufficiently compressible
       data -- copy the remaining input data into an allocated buffer and
       complete the compression from there to the now empty input buffer (this
       will only occur for long incompressible streams, more than ~20 MB for
       the default deflate memLevel of 8, or when *max is too small and less
       than the length of the header plus one byte) */
    hold = strm->zalloc(strm->opaque, strm->avail_in, 1);
    if (hold == Z_NULL)
        return Z_MEM_ERROR;
    memcpy(hold, strm->next_in, strm->avail_in);
    strm->next_in = hold;
    if (have) {
        memcpy(buf, temp, have);
        strm->next_out = buf + have;
    }
    strm->avail_out = (buf + *max) - strm->next_out;
    ret = deflate(strm, Z_FINISH);
    strm->zfree(strm->opaque, hold);
    *max = strm->next_out - buf;
    return ret == Z_OK ? Z_BUF_ERROR : (ret == Z_STREAM_END ? Z_OK : ret);
}

Codger

Awesome!! Love this place. And I don't have long incompressible streams, so even better! Thanks a ton! – Buntline