(C API) Segfault when calling rocksdb_slicetransform_destroy() after rocksdb_options_set_prefix_extractor()

[warrenfalk]

The code below will segfault.

The expectation is that a call to rocksdb_slicetransform_create_fixed_prefix() should be paired with a call to rocksdb_slicetransform_destroy().

However, once the rocksdb_slicetransform_t is associated with an rocksdb_options_t, it gets destroyed when the rocksdb_options_t is destroyed, and so if you call rocksdb_slicetransform_destroy() it segfaults.

This makes language bindings exceedingly clumsy as you have to keep track of whether the rocksdb_slicetransform_t has been associated with a rocksdb_options_t somewhere.

Is this desired behavior?

/*
 The code below segfaults.
 Lines marked (B) and (C) appear to be mutually exclusive.  Is this the desired behavior?
*/
#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#include <assert.h>
#include "rocksdb/c.h"
int main(int argc, char **argv) {
  rocksdb_slicetransform_t *prefix = rocksdb_slicetransform_create_fixed_prefix(8);  // <-- (A) Create Prefix Extractor
  rocksdb_options_t *options = rocksdb_options_create();
  rocksdb_options_set_create_if_missing(options, 1);
  rocksdb_options_set_prefix_extractor(options, prefix); // <-- (B) Set Prefix Extractor
  /* ... */
  rocksdb_options_destroy(options);
  rocksdb_slicetransform_destroy(prefix); // <-- (C) Destroy Prefix Extractor
  return 0;
}

[warrenfalk]

The segfault also occurs with the bloom filter policies also, and possibly more. However it does not occur with block based table options. And here is what appears to be the difference:

The following seems to be OK:

void rocksdb_options_set_block_based_table_factory(
    rocksdb_options_t *opt,
    rocksdb_block_based_table_options_t* table_options) {
  if (table_options) {
    opt->rep.table_factory.reset(
        rocksdb::NewBlockBasedTableFactory(table_options->rep));
  }
}

The following eventually results in segfault:

void rocksdb_options_set_prefix_extractor(
    rocksdb_options_t* opt, rocksdb_slicetransform_t* prefix_extractor) {
  opt->rep.prefix_extractor.reset(prefix_extractor);
}

Is there any reason why the latter can't be made to work like the former?

[gfosco]

Closing this via automation due to lack of activity. If discussion is still needed here, please re-open or create a new/updated issue.