Segmentation fault occurs when sequence array element is undefined or partially defined

[Orbuild]

• When I use the code from the master branch, I ran a simple publish-subscribe program, where the message format is as follows:

  module MyTestData
  { 
  struct Msg {
  @key long id;
  sequence<octet> data[10];
  };
  };

• If the sequence array is not defined or only some elements in the array are defined, the segmentation fault will occur in the program. I would like to ask if this is the agreed behavior (all elements in the sequence array need to be defined), or is this a problem? Looking forward to your response!

  for (int i = 0; i < 5; i++) {
  msg->data[i]._buffer = dds_alloc(100);
  msg->data[i]._length = 100;
  msg->data[i]._release = true;
  }

  === [MyTestPublisher]  Waiting for a reader to be discovered ...
  [MyTestPublisher]  Writing : Segmentation fault (core dumped)

[eboasson]

Yes, in general samples that you pass into the API need to be valid. That means:

• strings must point to a null-terminated string or be a null pointer
• sequences need _length ≤ _maximum and _buffer may not be a null pointer if _length > 0
• the sequence buffer will not be allocated/resized/freed if _release is false except that if everything is 0 it will take over ownership in read/take and allocate and that it is somewhat forgiving of _maximum < _length
• all elements of an array need to be initialized
• the union case identified by the discriminant must be correctly initialized
• @external pointers may never be a null pointer on write.

Maybe I forgot something, maybe not 😂 I'm sure you understand. That exception noted above about sequences exists so you can memset everything 0 and you'll be good for read and take.

There are few functions that only care about the values of the key fields, for example dds_dispose. Those ignore all other fields, so in your example, you can call dds_dispose with id set correctly and everything else garbage.