tensorflow_io namespace? about io HOT 5 CLOSED

I think we could create a different namespace, or anom namespace if possible. The tensorflow-io is always complied with tensorflow, so even if there is a collision we will be able to find out quickly I believe.

from io.

I am a bit more worried about tensorflow changes causing collisions with current/past versions of tensorflow-io then the other way round.

from io.

@suphoff At the moment we pin to a specific version of tensorflow for each of the past releases. For example, in the past we pin to 1.12.0 in tensorflow-io 0.1.0-0.3.0. For the upcoming tensorflow-io 0.4.0 we plan to pin to 1.13.0.

The reason is that while our python code has largely been forward compatible, there are still some C++ changes in tensorflow from time to time to make it backward/forward incompatible. One example is #54 (comment)

Since tensorlow-io tends to release faster than tensorflow (at least in the near future), I would be less worried about the namespace collision for now. But if there is anything we could do to make it future compatible, e.g., have a namespace tensorflow_io or anonymous namespace, it would be great as well.

from io.

@yongtang : I am mechanically changing the name space of my prototype for #109
from:

namespace tensorflow {
// ....
}  // tensorflow

to

namespace tensorflow_io {
using namespace ::tensorflow;
// ....
}  // tensorflow_io

As sources are distributed over several files anonymous namespaces are not really an option.
The prototype should be ready in a few days and can then serve as an example. (Good or bad)

from io.

With the recently updates, most of the C++ files should have the correct namespace:

namespace tensorflow {
namespace io {
namespace { // if applicable use anonymous namespace

} // namespace
} // namespace io
} // namespace tensorflow

Think this issue could be closed.

from io.