Filter Objects

A filter object looks like this:

struct filter {
    filter_read_t       *read;          /* read data out of filter */
    filter_write_t      *write;         /* write data into filter */
    filter_end_t        *end;           /* signal end of data */
    filter_convert_t    *convert;       /* map # bytes in <-> out */
    filter_destroy_t    *destroy;       /* destroy filter */
    void                *private;       /* object private data */
};

typedef int   filter_read_t(struct filter *f, void *buf, int len);
typedef int   filter_write_t(struct filter *f,
                  const void *data, int len);
typedef int   filter_end_t(struct filter *f);
typedef int   filter_convert_t(struct filter *f,
                  int num, int forward);
typedef void  filter_destroy_t(struct filter **fp);

Note: these functions must be implemented to be thread-safe. For example, two threads should be able to write to and read from the same filter object simultaneously.

The read() method should read data from the filter and return the number of bytes read (up to len bytes), or 0 if more data needs to be written to the filter (i.e., the filter's internal buffer is empty). If an error is encountered, it should return -1 with errno set appropriately.

The write() method inputs data into the filter, returning the number of bytes input (up to len bytes). It should return 0 if the filter's internal buffer is full, or -1 and set errno on error.

The end() method indicates to the filter that no more data will be input. It should return 0 on success or -1 and set errno on error. After this method is called, any calls to write() should return -1 with errno set to EPIPE.

The convert() method provides estimates of the ratio of input length to output length, and vice-versa. If forward is non-zero, convert() should return an upper bound on the number of bytes of that num bytes of input will generate. Otherwise, it should return an upper bound on the number of bytes of input that are required to generate num (or more) bytes of output.

The destroy() method should free all resources associated with the filter and set *fp to NULL. If *fp is already equal to NULL, destroy() should do nothing.

The filter_read(), filter_write(), filter_end(), filter_convert() and filter_destroy() functions are convenience wrappers for the corresponding object methods.

By implementing these methods and providing a constructor function to create new instances, user-defined filters may be used with the functions below.

Filter Functions

filter_fopen() pushes a filter on top of a uni-directional stream, returning a new stream. Data read from or written to the newly created stream will pass through the filter. The mode argument is as described for fopen(3), but is restricted to being either "r" or "w".

If flags is zero, calling fclose(3) on the newly created stream causes the underlying stream fp to be closed and filter to be destroyed. Otherwise, the flags value may contain any of the following values OR'd together:

FILTER_NO_CLOSE_STREAM    fclose() does not close underlying stream
FILTER_NO_DESTROY_FILTER  fclose() does not destroy the filter

filter_process() sends len bytes of data pointed to by input through the filter. Upon successful return, the results are placed into a newly-allocated buffer having typed_mem(3) type mtype and pointed to by *outputp; the caller must eventually free this buffer. If final is non-zero, then the filter's end() method is called after the last input byte is written.