GH-43258: [C++][Flight] Use a Base CRTP type for the types used in RPC calls

[felipecrv]

Rationale for this change

Flight should eventually be buildable without a specific protobuf version. As such, all the protobuf types are wrapped in hand-written classes. Uniformity of the interface is not enforced even though it's desirable. Extending the interface requires adding functions to every struct. A common base class can mitigates risks and reduce the amount of hand-written code.

What changes are included in this PR?

• Definition of a BaseType<> template (a CRTP)
• Add constructors that aren't implicitly defined anymore
• Having a default value for some fields that would be undefined values otherwise
• Leverage this structure to add SerializeToBuffer to all the types
• ~30KiB in binary size reduction

Are these changes tested?

By existing tests.

• GitHub Issue: [C++][Flight] Add a base CRTP class template to serializable types #43258

[lidavidm]

Can we make the impls protected and/or update the docstrings to make it clear that these are internal?

[felipecrv]

Making them private breaks self()->Do... calls in the CRTP base class. I could work around this with friend annotations, but to be frank, it was my intention to have these as public functions as well because if I ever deserialize object in a loop, I can re-use an std::string and avoid mallocs.

I will add docstrings telling users that they can call Seria... because they are more convenient.

[lidavidm]

I'd really rather not have so many different methods. If we're going to keep them, please at least make them overloads (so drop the 'Do') since they are just the same methods implemented in terms of each other.

[lidavidm]

Just reiterating the comment above.

[felipecrv]

I will drop the Do- prefix.

I'd really rather not have so many different methods.

If I'm going to automatically generate code (for FlightSQL), the functions that take std::string* produce much less binary bloat than Result<T> which is why this is the pattern adopted in protobuf itself.

[lidavidm]

Ok, sounds good to me.

[felipecrv]

bloaty report:

$ ninja libarrow_flight.a && bloaty -d symbols -C full -n 0 ./release/libarrow_flight.a -- BEFORE-libarrow_flight.a
[17/17] Linking CXX static library release/libarrow_flight.a
    FILE SIZE        VM SIZE
 --------------  --------------
  -1.0% -28.1Ki  [ = ]       0    [AR Non-ELF Member File]
  -1.0% -28.1Ki  [ = ]       0    TOTAL

It's small compared to the gigantic arrow library, but any reduction is good IMO.

[felipecrv]

I think I broke the MSVC build by merging this. I will try fixing it here -> #43330 #43332

[conbench-apache-arrow]

After merging your PR, Conbench analyzed the 0 benchmarking runs that have been run so far on merge-commit 9fb1129.

None of the specified runs were found on the Conbench server.

The full Conbench report has more details.