Foldable take a TypeConstructor. The first argument must be a
NatNum.
The FunctionSignatureRouter wasn't completely on point, instead
this commit adds an TypeClassArgsRouter lookup router. This
makes sense since the only available arguments we have to find
a router is the list of type class arguments.
Before this commit, finding the implementation for a type
class method was done with a simple lookup table.
This commit adds a router based on function signature.
This also paves the way for adding type constructor
arguments in function signatures.
And it removes quite a few references to the prelude out
of the compiler.
Also adds a bunch of helper methods to render signatures
as strings.
By annotating types with the constructor application
that was used to create them.
Later on we can use the router to replace compiler's
INSTANCES or for user defined types.
A lot of isinstance checks no longer did anything, since
the referred to variable was always a type.
In some places, we used it to check if a type was internal,
it's unclear if we will need to rebuild those checks in
the future.
Before, a type class was a property of a type.
But that doesn't make any sense for multi parameter
type classes.
Had to make a hacky way for type classes with
type constructors.
They look a lot like placeholders, but they exist before
the typing system takes place. And there's a (much smaller)
context to deal with.
For now, removes Placeholders in user function definitions
as they were not implemented.
Adds signature to function to try to get them closer to
type class methods. Already seeing some benefit in the
constraint generator.
Stricter zipping for safety.
Also adds the remaining unexposed WebAssembly opcodes as
comments (eqz, clz, ctz, popcnt, copysign).
This also means that BinaryOp.operator is now always
a type class method.
"An arithmetic shift is usually equivalent to
multiplying the number by a positive or a negative
integral power of the radix." -- Federal Standard 1037C
"This is the same as multiplying x by 2**y."
"A right shift by n bits is defined as floor division
by pow(2,n). A left shift by n bits is defined as
multiplication with pow(2,n)."
This is because Haskell defines negate, abs and signum
for Num, but they don't work with our unsigned number
types. (abs would be a noop.) Haskell has Word32 / Word64,
but there negate doesn't make much sense to me.
Implemented neg and abs.
Implemented a type class inheritance check.
Removed Integral from u8 and i8 since it wasn't implemented.
Prior to this PR, each type would have its own handwritten
test suite. The end result was that not all types were tested
for all situations.
This PR adds a framework based on a Markdown file, which
generates the basic tests for the types defined in json
files. These are auto generated and updated by the Makefile
before the test suite is run.
Also, a number of unsupported type combinations are now
supported.
Also, we now support negative literals.
Also, allocation calculation fixes for nested types.
Also, the test helpers can now properly import and export
typed variables such as bytes, static arrays and tuples. This
may come in handy when it comes to phasm platform wanting to
route data.
Also, adds better support for i8 type.
Also, started on a runtime.py, since there's quite some code
now that deals with compile time handling of WebAssembly stuff.
Also, minor improvement to the type constrains, namely we
better match 'tuple' literals with static array types.
Also, reduced spam when printing the type analysis results;
constraints that go back on the backlog are now no longer
printed one by one. It now also prints the end results of
the typing analysis.
Also, reorganized the big test_primitives test into type
classes.
Also, replaced pylint with ruff.
Prior to this PR, the compiler would call stdlib.types's
__subscript_bytes__.
However, that function performs some checks we do not need.
After this MR, folding iterates directly over the bytes
memory, saving the memory access checks and the function
calls. This gets us a speedup of about 43% less CPU time
used on Firefox.
Also, by default, the CRC32 page runs a shorter timing test.
Doesn't give right answer yet and out of bound check fails.
No constructor yet for static arrays, but module constants
work. Which don't work yet for tuples and structs.
Also, u32 for indexing please.
Also, more module constant types.
