The project file still only compiles a test binary, but it should be easy to separate out a library project from here.

Thanks again to Bryan Boreham <bjboreham@gmail.com> for much help getting this working.

Extended discussion:
https://groups.google.com/d/msg/capnproto/lRlWBOglQv4/8-Qo96AcZQIJ

Extended discussion:
https://groups.google.com/d/msg/capnproto/lRlWBOglQv4/8-Qo96AcZQIJ

To use, pass --disable-reflection to the configure script.

This produces a smaller runtime library. However, using it for this purpose is not recommended. The main purpose of lite mode is to define a subset of Cap'n Proto which might plausibly compile under MSVC. MSVC still lacks full support for constexpr and expression SFINAE; luckily, most of our use of these things relates to reflection, and not all users need reflection.

Cap'n Proto lite mode inherits its name from Protocol Buffers' lite mode. However, there are some key differences:

- Protobuf generated code included global constructors related to registering descriptors and extensions. For many people, this was the main reason to use lite mode: to get rid of these global constructors and achieve faster startup times. Cap'n Proto, on the other hand, never had global constructors in the first place.

- Schemas are actually still available in lite mode, though only in their raw (Cap'n Proto structure) form. Only the schema API (which wraps the raw schemas in a more convenient interface) and reflection API (which offers a convenient way to use the schemas) are unavailable.

- Lite mode is enabled in an application by defining CAPNP_LITE rather than by specifying an annotation in the schema file. This better-reflects real-world usage patterns, where you typically want to enable lite mode application-wide anyway.

- We do not build the lite mode library by default. You must request it by passing --disable-reflection to the configure script. Before you can do that, you must have a prebuilt Cap'n Proto compiler binary available, since the compiler can't be built without reflection.

- Relatedly, the lite mode library is built with the same name as the full library. This library is not intended to be installed. If anything it should be statically linked. But, mostly the option only exists on non-MSVC platform to give us a way to test that we haven't broken lite mode.

Note that embedded schema structures in generated code are still incorrect.

Actually, let's use single digits for these otherwise they'll actually be printed in scientific notation with a '.', defeating the purpose of the test.

Appending 'f' to an integer literal doesn't make it a float. We have to add '.0' first.

Fixes #119.

For portions currently copyright by Kenton (most of it), transfer copyright to Sandstorm Development Group, Inc. (Kenton's company).

The license change is practically meaningless, as MIT and BSD 2-clause are legally equivalent. However, the BSD 2-clause license is sometimes confused for its ugly siblings, BSD 3-clause and BSD 4-clause. The MIT license is more immediately recognizeable for what it is.

Rémy Blank and Jason Choy (the two non-trivial contributors) are on record as approving this change:

https://groups.google.com/d/msg/capnproto/xXDd2HUOCcc/gbe_COIuXKYJ

Fixes #90. The bug causes floating-point fields with non-zero defaults to be corrupted if they are set to NaN on a MIPS machine.

Extend totalSizeInWords() to also return a count of capabilities, which helps when a separate capability table needs to be allocated as well.  Use this in the RPC system.

Eliminate the ability to have multiple threads working on building the same message -- performance penalty is too large, and applies even to single-threaded users.

Remove 'has' for primitive fields.  Dynamic version always returns true.  This also means that primitives are always included in text format, even if they are default-valued.

Define new KJ constants for min/max integer values and inf/nan, then remove all use of numeric_limits.

Revamp generated interface to Object fields.  Now there is an ObjectPointer::{Reader,Builder}.  This will simplify the dynamic API (next change) and also makes it easier to delegate decisions about the object type to a function that doesn't know about the context where the object lives (i.e. by passing an ObjectPointer::Builder).

Fix bug where only the first import made it into the import table.  Can't believe this wasn't tested.  Sigh.

Actually fix the bug, which was a doozy:  OrphanBuilder::tag was sometimes initialized using WirePointer::setKindAndTarget(), but since the tag didn't live inside the target segment, this used illegal pointer arithmetic.  The target is never read from an orphan tag anyway, so I thought it would be no big deal.  But it turns out Clang actually optimizes under the assumption that pointer arithmetic returns a whole value.  As a result, on 32-bit system where 64-bit values are only 32-bit aligned, the tag and target might not have been a whole number of words apart, and the extra bit actually found its way into the 'kind' bits, causing e.g. a struct pointer to become an invalid far pointer.  Crash.  The fix required refactoring to ensure that setKindAndOffset() is never used for orphan tags.

Delete default construcrtor on builders, so that people don't incorrectly try to just declare a Builder on the stack and then start setting its fields.  Instead, a Builder can be explicitly assigned to nullptr, which makes it clearer that it isn't safe to use.

Rename non-group again, now to 'slot', which is much nicer.  Also introduce fixed lists as a third kind of field (but not implemented yet).

Warn when an optimized build doesn't set NDEBUG and then #define it ourselves -- unless the build explicitly sets DEBUG.

New method isFoo() exists for union fields only and returns true if foo is the active union field.

Generate usable code for groups, although making schemas work is still an open problem.  I think I'm going to do the schema rewrite now.

Unnamed unions WIP.  About to change MemberInfo format, which will require re-bootstrapping the code generator.