GitLab

added QDatastream bench (best perf but implicit structure)

fixed all issues but compile issue

- remove QBind::bind::overloads that do not allow both definitions to
coexist
- handle T&&, T& and const T& universally using perfect forwarding,
- factor out static_assert when QBind Read to const T&,
- factor out QBind Write const_cast<T&>

Removed optimizations from *Result shortcut methods (no noticeable perf
improvement)

hard to understand for implementers) with an optional setChoice() and
ScopedChoice that tags all alternatives in a choice to avoid reporting
"normal" mismatches as errors
Removed QCborWriter
Specialised QMovedResult into QMovedWriter as it is unlikely to be used
with BindMode::Read

Fixed a few things
Simplified QBind<_,QJsonValue> reusing QBind<_,QJsonArray>

detection of expected shape
Introduced min cardinality in fluent interface envisioning the addition
of item(2,3).any() syntax...

explicitely constructing it...

Added one decimal to better measure performance which is now close to a
few µsec

Avoiding building superfluous Val<> does not seem to pay for the additional friendship declarations on the "user" side.

Fixed a few errors

Added T>Cbor test
Attempted to add Json>Cbor which does not compile for now

Added CborWriter:QMovedResult for writing Cbor without heap allocation (QCborWriter can be used when a pimpled class is needed)

Added QJsonReader JSON whitespace support

Simplified QMovedResult for direct use, requiring just an extra .value() call from the user but no "facade" class to hide protected impl

Drawbacks are:
- users must call .result() to close opened data structures
- it is more difficult to implement TResult and TImpl correctly
- TResult is not pimpled by default

simplified implementations interface (always return bool)

WIP: replaced allocation of TResultShared with move semantic (copy only used for nested QBind) for very slight perf improvement

Grouped QJson* support in QJson.hpp
Kept use cases and benchmarks in main.cpp