Re: [Gnash-commit] gnash ChangeLog libamf/element.cpp libnet/cque....

[Rob Savoye]

strk wrote:
> Rob, I provided a clean patch for ::merge.

  It didn't work "as was" for me. You can attempt to apply it to this
current version of ::merge(), I'm done staring at it. To be honest this
whole thing about singular iterators is too obscure. My current patch
fixes all the aborts, but still leaks a bit of memory still, which is
next to track down. I have a pretty good idea where though, but wanted
to fix the aborts from -D_GLIBCXX_DEBUG first.

  I do now see why we've been getting different results. You use
-D_GLIBCXX_DEBUG, and I usually never do. Without this the test cases
worked just fine, with this they'd core dump.

> It wasn't deleting elements as I wasn't sure about ownership, but would be
> pretty easy to modify to remove the merged lines.

  Basically, here's the idea of merging buffers. A large packet will be
several Buffers, as the network will only read a set amount of data. The
final packet in a series is always less than the default size.

  Some protocols, like RTMP, need the entire data as one packet, so I
added the ::merge method to do this. This way the size of the data
required doesn't have to be large enough to read all the data in one
operation. Ultimately, we'll probably also need a ::split() method as
well as some buffers will contain two separate data packets if the
network read was for a large amount of data. In those cases, since RTMP
is stupid, we'll have to split Buffers into their distinct RTMP messages.

> If we have test for merging it should be easy to verify it's functionality.

  The test in test_cque is lame, and needs to be improved. All it
currently does is makes sure the new buffer is the correct size. Better
than nothing, but that's all it does.

        - rob -