why don't Readers have copy constructors?

[Jason E. Aten]

In the code generated by capnp from a pp.capnp file, it seems that Readers are returned by value, and so should have copy-constructors. But the pp.capnp.h generated file doesn't appear to declare copy constructors for Readers... so I'm getting the compiler-written default-shallow copy constructor.

I'm getting crashes from lack of deep copy as a result...

See common_pp.cpp:86 in the attached minimized test case.

[Kenton Varda]

Hi Jason,

This is working as intended.

As described in the docs, Readers and Builders both behave like pointers -- they point at something else, but they don't own the data.  Copying a Reader is like copying a pointer; it's a shallow copy.

Doing a deep copy is complicated because Cap'n Proto objects are not allocated on the heap.  You need to copy *to* some other message.  What you can do is something like:

    MallocMessageBuilder newMessage;

    newMessage.setRoot(someReader);

This will make a deep copy of someReader, making it the root of newMessage.  Similarly, struct-typed fields have "set" accessors that make a deep copy.

-Kenton

[Jason E. Aten]

Hi Kenton,

I'm still puzzled, then, as to why I am seeing the crash at line 67 of clinano.cpp.  If Reader is just a pointer, then this should succeed, but it crashes with:

terminate called after throwing an instance of 'kj::ExceptionImpl'                                                                                           
  what():  src/capnp/layout.c++:1904: requirement not met: expected boundsCheck(segment, ptr, ptr + roundBytesUpToWords(ref->listRef.elementCount() * (1 * BY\
TES / ELEMENTS))); Message contained out-of-bounds text pointer.   

- Jason

[Kenton Varda]

It looks like receiveZsample() is returning a Reader that points into a message that no longer exists after the function returns.  At least, I don't see any MessageBuilder instances allocated on the stack.

[Jason E. Aten]

Ah, I think I see. The FlatArrayMessageReader at line 73 in common_pp.cpp seems to "own" the message,because things don't crash if I new it up instead of making the FlatArrayMessageReader stack allocated.

In general is it correct to surmise that the MessageBuilder or the FlatArrayMessageReader will "own" the messages?

[Kenton Varda]

Yes, the MessageBuilder/MessageReader is the owner of the message content.

[Jason E. Aten]

On Sat, Oct 19, 2013 at 4:02 PM, Kenton Varda <temp...@gmail.com> wrote:

Yes, the MessageBuilder/MessageReader is the owner of the message content.

Very good -- thanks for clarifying that.

Btw, I'm using nanomsg presently, but would be happy to try your rpc if you would benefit from some early testing on it.

[Kenton Varda]

Hopefully Cap'n Proto RPC will be ready within a week or two...