a253560600 / protobuf-net

Automatically exported from code.google.com/p/protobuf-net
Other
0 stars 0 forks source link

Better usage guide / FAQ #42

Open GoogleCodeExporter opened 9 years ago

GoogleCodeExporter commented 9 years ago
All common scenarios, including inheritance, inferred tags, remoting, etc

Original issue reported on code.google.com by marc.gravell on 11 Feb 2009 at 8:04

GoogleCodeExporter commented 9 years ago
"Better usage guide" implies that there is a usage guide somewhere. I can't 
find one. Did I miss a link somewhere? Also, I installed the VS10 msi but can't 
see that it added anything to the VS environment.

I'm working on a project to provide allow us to rapidly deploy a hardware 
interface for HW developers to test embedded systems. These systems can be 
hosted on anything from a full blown Windows7 running .Net to linux (C++ or 
mono) running on the equivalent of a washing machine CPU. The developers will 
be working in .Net. I need to provide them remote access to the host. Protobuf 
seems like the perfect solution. If the host is resource limited, it could be 
running pb in native C++ and the clients would just need to choose the 
protobuf-net serialization method.

I have a proof of concept started using .Net remoting channels with custom data 
contracting encapsulating the HW registers. Now I want to get it working with 
protobuf .Net. I assume I just need to provide the new Client/Server 
binaryformatter classes but I'm at a loss as to how to get started given the 
sparse documentation. My deadline for delivering the proof of concept is 
rapidly approaching. Any pointers to documentation/examples would be greatly 
appreciated!

Original comment by jpat34...@gmail.com on 1 Jun 2011 at 12:34

GoogleCodeExporter commented 9 years ago
The installer should add support for .proto files. Whether you need this or not 
depends entirely on the context. Re BinaryFormatter / remoting - well, that 
will *work*, but probably wouldn't be my first choice. However, if you *want* 
to use protobuf-net within remoting, you simply implement ISerializable and 
invoke Serializer.Serializer / Serializer.Merge (in the ctor).

If you can give more context as to what is running at each end, I might be able 
to help more; protobuf-net is but a single implementation of protobuf, and is 
aimed at .NET developers. If you are using .NET (I'm not entirely sure, due to 
the "native C++" mentions) it should be pretty easy to throw together. Note 
that it is a *serialization* API, not a *remoting* API (although interoperable 
RPC stacks are mentioned regularly)

Original comment by marc.gravell on 1 Jun 2011 at 1:04

GoogleCodeExporter commented 9 years ago
Hi Marc - thanks for the quick reply. The idea is that the HW developers will 
be using the framework our group is developing for algorithm development and 
debugging. They will be using the provided API (that provides a high level 
abstraction of the host's HW registers, PCI devices etc.) These are the 
clients. They basically translate high level API calls to a series of 
address/data pairs remoted to the host which writes them to HW. As the HW 
engineers develop the algorithms necessary to get the HW to do its thing, they 
pass them off to the software developers who are working in parallel to build 
the host application. In other words, the client code will never ship so 
performance is not a huge concern.

The server has to run on the embedded HW which might be a complete .NET4 
platform, or a CF or even linux (hence the C++ reference above). Thus I need 
the clients to be agnostic. If the HW is being deployed on a .NET platform the 
client will use Remoting or WCF. If the host is resource starved the server 
would run protobuf-net/CF or perhaps protobuf in native C++. The client API has 
to adapt to the project's choice and place as few restrictions as possible on 
the host software developers.

The problem now is that each project rolls its own development environment. My 
group is trying to provide the infrastructure to make this process quick and 
uniform. Bus transfer speed is not an issue here. The data to serialize will be 
mostly address/int pairs that configure the HW. What is important is cross 
platform interoperability and rapid deployment of the API. The idea is to 
provide data contracts (in whatever flavor the project needs i.e. WCF contract, 
proto file etc.) which model the various types of HW devices we need to access 
and then configure client back end to use the selected serialization protocol. 
The easiest way to accomplish this I think is to plug in the appropriate 
serialization scheme into the .Net channel model. Better ideas would be 
appreciated!

Thanks again,
Jeff

Original comment by jpat34...@gmail.com on 1 Jun 2011 at 2:36

GoogleCodeExporter commented 9 years ago
I would stay away from remoting, personally. Raw sockets or a basic http 
payload (just POST the serialized binary) would be my suggestions. Easy to port 
between random platforms.

Original comment by marc.gravell on 1 Jun 2011 at 3:07

GoogleCodeExporter commented 9 years ago
I'd request that the wiki be fleshed out with the following info (or if this 
info is in source code, the wiki should say so)...

- A table of contents(which the "Getting Started" page should link to or 
include at the end)
- Usage "without attributes". The "Getting Started" page still implies this is 
not possible.
- How are shared references or circular references handled?
- How are references to "object" handled?
- How are collections handled, and can I change collection types between 
versions? If I write "SortedList<A,B> Foo;" in version one and "Dictionary<A,B> 
Foo;" in version two, they should ideally be compatible. Assuming collections 
are detected automatically, what if I have a class that happens to implement 
IEnumerable but I want ordinary field-and-property-based serialization?
- What happens when I change data types between versions? (hmm, I remember 
reading somewhere in the protocol buffer documentation about how protocol 
buffers handle various kinds of data type changes... wasn't able to locate that 
info in a 2-minute search)
- [ProtoInclude(7, typeof(SomeDerivedType)] is an attribute on the base class. 
Why can't this be an attribute on the derived class instead?
- When and how should one use RuntimeTypeModel?

Original comment by qwertie...@gmail.com on 17 Aug 2013 at 6:01