How it works

Compile SBE schema using sbeppc to get C++ headers.
Integrate them into your project.
Link to sbepp::sbepp library.
Use it!

sbepp has two main goals:

provide fast, direct access to SBE data
provide convenient representation of SBE entities like types, messages, etc.

As a result, almost all types in sbepp are reference semantics wrappers over memory buffer. They never manage underlying memory in any way, only treat the data according to SBE standard. The idea was to provide a thing similar to plain struct with getters/setters and leave rest to a client. The only thing which is handled implicitly is a schema extension mechanism, everything else must be done by a client. For a boiler-plate things like header filling sbepp provides helper utilities. Basically, all it does under the hood is offset calculation and memory read/write.

It's worth mentioning that at the moment sbepp cannot generate schema representation at run-time like other similar software.

How it compares to RealLogic

RealLogic is more like a platform rather than a simple tool. It can generate code in different languages, load schema at run-time, validate XML, etc. sbepp is focused solely on generating C++ code.

Both tools generate view-like classes which don't manage memory. However, RealLogic requires non-const char* buffer even for decoding, while sbepp takes byte type as a template parameter and provides read-only access when it's const-qualified.

The performance of the code generated by both tools is quite similar, at least in my benchmarks I don't see a significant difference. In compiler-explorer experiments I found that in some cases RealLogic's code results in a noticeably more assembly lines while the code generated by sbepp is nearly identical to the assembly of a hand-written code.

RealLogic requires forward-only access when one needs to touch each field either by reading or skipping it with explicit call. sbepp provides both, random and forward-only access. Since most messages have simple structure, random access performs just as well for them.

RealLogic has inconvenient 2-step process for message creation. sbepp needs only a single line to create a message wrapper and one more call to fill message header in case of encoding.

In general, RealLogic provides poor C++ abstractions for SBE types and mixes various helper functions with getters/setters. This results in message interface being polluted with methods like field_nameId(), field_nameEncodingOffset(), field_nameNullValue() along with field accessors. sbepp provides convenient STL-like wrappers for each SBE entity and never mixes accessors with internal methods so auto-suggestion in IDE shows you only schema names and nothing else. This interface problem holds for group and fixed/variable length array interfaces. They just don't work in a C++ way.

RealLogic uses exceptions to report boundary check violations (if checks are enabled). sbepp uses assertions and they are supposed to be enabled only in Debug mode.

RealLogic provides iostream-based operator<< with fixed formatting. sbepp provides only stringification primitives and Visit API which allows clients to build their own to_string() with desired format and output mechanism.

License

Distributed under the MIT license.