Since testcases are de facto documentation, add a testcase for the new __tuple function to show how it's used.
Possibly also have a section on the wiki about static unspooling of loops?

mkdir examples
git mv *.nt examples/
cat > Makefile

build script.

dmd --blah
^Z

Neat has different exception/error syntax than D, and needs documentation:
Explain how instead of "throw new Exception str" it uses "raise new Error str", how set-handler/define-exit/all those other weird labels work, and then feel bad for not using try {} catch {} instead.

[todo: chc4 needs patching to not be dumb]

Importing std.time fails at compile-time. Log is:

opt -o - .obj/std/time.ll |opt -mattr=-avx,-avx2 -march=x86  - -mcpu=nocona  |llc -O1 -mattr=-avx,-avx2 -march=x86 -mcpu=nocona - -filetype=obj -o .obj/std/time.o
opt: .obj/std/time.ll:24:17: error: invalid cast opcode for cast from 'i32*' to 'i32 addrspace(1)*'
%var9 = bitcast i32* %var8 to i32 addrspace(1)*
                ^
ERROR: Compilation failed with 256 25

LLVM version:

LLVM (http://llvm.org/):
  LLVM version 3.4
  Optimized build.
  Built Jan  6 2014 (22:53:34).
  Default target: x86_64-unknown-linux-gnu
  Host CPU: corei7

I assume that this is a error with how FCC generates LLVM bytecode.

When using "using ", mvalue being an expression that supports assignment but not referencing, which of those should be done?

1. Export the mvalue as-is. Problem: if it's a struct, calls on it will create a temporary.
2. Export the mvalue, except if it's a struct (in which case allocate a temporary). Probably the DWIM thing to do, but hacky.
3. Create a temporary, then reassign it at scope end. Can have powerful effects involving casting reftuples to structs, but will probably lead to unexpected behavior.
4. Create a temporary, but don't reassign it.

I'll go with option two for now, but would appreciate feedback.

Seriously. Also link to the pre-built fcc binary in it and warn them not to try and build it themselves.

Currently, attempting to coerce a struct into a string with $foo throws an error message (which is good) containing the text Matching rule 'tree.scope': Can't format 5359 (*({struct __vtable<void*>@0, classinfo, saddr<sockaddr_in:sockaddr_in>@1}*: t)).saddr of sockaddr_in:sockaddr_in (which is bad).
The error message should be more friendly and only mention the friendly name of the struct, in this case sockaddr_in

When attempting to read socket.accept() in an iterator, I tried for Socket client <- tcpServer.accept().loop(), which segfaults FCC.
If this isn't the correct function to use for it, it should throw an error message instead of crashing. I don't see any other good way to loop over socket.accept() since while need a bool if condition and Neat doesn't support while(client = tcpServer.accept())

size-of, string-of, mangle-of, type-of, ReturnType, ParamTypes, types-equal (T,U), type-is, is-defined, implicitly-converts-to all need to be documented somewhere OTHER then then ast/obscure_file.d source. Including documentation about what arguments they take (I'm looking at you type-is tuple T)

Also include the stack guard constructs set-handler, define-exit, onExit, onFailure, onSuccess, etc. since they are pretty important, seeing how they are the entire error-handling system. Consider this a merge of #7.

int[] intTest = [0,0,0]; intTest[0] = 3; writeln "$(intTest[0])";
Crashes when it attempts to set intTest[0].
If you use a long[](or string[]) instead of int[], it works perfectly fine.
long[] longTest = [long:0,0,0]; longTest[0] = long:3; writeln "$(longTest[0])";