feat: Turn int and float into core types #225

mark-koch · 2024-05-28T08:38:42Z

This PR adds a new NumericType to the type hierarchy that now represents int and float.

Closes #250

croyzor · 2024-05-30T08:37:43Z

guppylang/tys/builtin.py

@@ -79,6 +86,23 @@ def check_instantiate(
        return NoneType()


+@dataclass(frozen=True)
+class _NumericTypeDef(TypeDef):
+    """Type definition associated with the builtin `None` type.


associated with the Numeric type!

croyzor · 2024-05-30T08:40:10Z

guppylang/prelude/builtins.py


 builtins = GuppyModule("builtins", import_builtins=False)

 T = guppy.type_var(builtins, "T")
 L = guppy.type_var(builtins, "L", linear=True)


+# Define the nat type so scripts can import it
+nat = nat_type_def


I'd expect this to implement some numeric dunder methods, like __add__, __eq__, __gt__?

croyzor · 2024-05-30T08:43:46Z

guppylang/prelude/_internal.py

@@ -145,16 +122,12 @@ def synthesize(self, args: list[ast.expr]) -> tuple[ast.expr, Type]:

        for i in range(len(args)):
            args[i], ty = ExprSynthesizer(self.ctx).synthesize(args[i])
-            if isinstance(ty, OpaqueType) and ty.defn == self.ctx.globals["int"]:
+            if isinstance(ty, NumericType) and ty.kind == NumericType.Kind.Int:


Shouldn't nat be valid too? Maybe bool?

arguably anything that implements __float__...

Yeah, we need a better way to figure out which types are automatically coreced (also applies to function tuples etc.). These CustomCallCheckers are quite hacky...

I'm tempted to just leave it as is until we have a proper solution?

arguably anything that implements float...

This only works as long as float is at the top of the numeric hierarchy. For example, if we add complex in the future, it might also have a __float__ method but shouldn't be coreced automtically.

Also we'd need coercsion from nat to int etc...

I think the better interim solution is that we coerce NumericType instead of just int, then come up with something better down the line

croyzor

I'm surprised not to see any special handling for nat in the type checker, and that it doesn't have any methods implemented by default. Are we ever checking that nats >= 0?

mark-koch · 2024-05-30T09:19:01Z

My plan was to leave all of this to another PR so this one doesn't become too big. But I could also add it here if you'd prefer to look at everything?

croyzor · 2024-05-30T09:23:03Z

I think the better way to split it up would be NumericType in this PR, then the one that adds the functionality also adds nat

codecov-commenter · 2024-05-30T12:52:10Z

Codecov Report

Attention: Patch coverage is 95.52239% with 3 lines in your changes missing coverage. Please review.

Please upload report for BASE (main@b2901d8). Learn more about missing BASE report.

Files	Patch %	Lines
guppylang/checker/core.py	87.50%	1 Missing ⚠️
guppylang/tys/builtin.py	91.66%	1 Missing ⚠️
guppylang/tys/ty.py	96.87%	1 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #225   +/-   ##
=======================================
  Coverage        ?   90.69%           
=======================================
  Files           ?       44           
  Lines           ?     4730           
  Branches        ?        0           
=======================================
  Hits            ?     4290           
  Misses          ?      440           
  Partials        ?        0

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

guppylang/prelude/_internal.py

croyzor · 2024-05-30T13:43:02Z

guppylang/prelude/_internal.py

+        return func
+
+    def synthesize(self, args: list[ast.expr]) -> tuple[ast.expr, Type]:
+        args, _, inst = synthesize_call(self.func.ty, args, self.node, self.ctx)


I don't really understand what this call is doing?

It checks args against the function inputs and potentially infers some type args. The call to self._get_func().synthesize_call() below delegates to the corresponding integer function.

qartik · 2024-05-30T13:52:30Z

bools are now treated as a numeric values as in Python.

What's the motivation for this change?

mark-koch · 2024-05-30T14:19:00Z

What's the motivation for this change?

Mainly consistency with Python. Adding __add__ etc to bool is easy to do and Imo we should follow Python as closely as possible unless we have a good reason not to.

I don't feel super strongly though, so we could reconsider if numeric bools are controversial?

qartik · 2024-05-30T14:30:32Z

If we are considering guppy as a strongly typed language, I think bool should not be implicitly treated as a numeric type. We could certainly provide casting/conversion functions for explicit coercion, if they don't exist already.

Co-authored-by: Craig Roy <croyzor@users.noreply.github.com>

mark-koch · 2024-06-24T13:52:45Z

@qartik I moved the bool change to another PR #263 and created issue #264 for discussion.

@croyzor Could you have another at this PR? Sorry for moving things around so much

croyzor · 2024-06-24T15:36:39Z

guppylang/tys/builtin.py

@@ -145,6 +175,10 @@ def bool_type() -> OpaqueType:
    return OpaqueType([], bool_type_def)


+def int_type() -> NumericType:
+    return NumericType(NumericType.Kind.Int)


Should this not contain some reference to int_type_def?

Is this used? I see a couple of places it'd work in this PR but it's just being inlined

Should this not contain some reference to int_type_def?

No, only the int_type_def has a reference to a NumericType. The inverse mapping from type to type def is implemeted here:

guppylang/guppylang/checker/core.py

Lines 82 to 112 in cf8a529

def get_instance_func(self, ty: Type | TypeDef, name: str) -> CallableDef | None:

"""Looks up an instance function with a given name for a type.

Returns `None` if the name doesn't exist or isn't a function.

"""

type_defn: TypeDef

match ty:

case TypeDef() as type_defn:

pass

case BoundTypeVar() | ExistentialTypeVar() | SumType():

return None

case NumericType(kind):

match kind:

case NumericType.Kind.Int:

type_defn = int_type_def

case NumericType.Kind.Float:

type_defn = float_type_def

case kind:

return assert_never(kind)

case FunctionType():

type_defn = callable_type_def

case OpaqueType() as ty:

type_defn = ty.defn

case StructType() as ty:

type_defn = ty.defn

case TupleType():

type_defn = tuple_type_def

case NoneType():

type_defn = none_type_def

case _:

return assert_never(ty)

Is this used? I see a couple of places it'd work in this PR but it's just being inlined

Good point, it's indeed unused. I'll remove it 👍

croyzor · 2024-06-24T15:52:12Z

guppylang/prelude/_internal.py

-            args=num_params * [tys.TypeArg(tys.BoundedNatArg(n=INT_WIDTH))],
-            parent=UNDEFINED,
-        )
+        ops.CustomOp(extension=ext, op_name=op_name, args=args, parent=UNDEFINED)


🤖 I have created a release *beep* *boop* --- ## [0.6.0](v0.5.2...v0.6.0) (2024-07-02) ### Features * Add array type ([#258](#258)) ([041c621](041c621)) * Add nat type ([#254](#254)) ([a461a9d](a461a9d)) * Add result function ([#271](#271)) ([792fb87](792fb87)), closes [#270](#270) * Allow constant nats as type args ([#255](#255)) ([d706735](d706735)) * Generate constructor methods for structs ([#262](#262)) ([f68d0af](f68d0af)), closes [#261](#261) * Return already-compiled hugrs from `GuppyModule.compile` ([#247](#247)) ([9d01eae](9d01eae)) * Turn int and float into core types ([#225](#225)) ([99217dc](99217dc)) ### Bug Fixes * Add missing test file ([#266](#266)) ([75231fe](75231fe)) * Loading custom polymorphic function defs as values ([#260](#260)) ([d15b2f5](d15b2f5)), closes [#259](#259) --- This PR was generated with [Release Please](https://github.com/googleapis/release-please). See [documentation](https://github.com/googleapis/release-please#release-please).

mark-koch force-pushed the feat/numeric-tys branch 2 times, most recently from 8c04951 to d22d4bb Compare May 28, 2024 09:39

feat: Add a nat type and make int/float core types

179b115

mark-koch force-pushed the feat/numeric-tys branch from d22d4bb to 179b115 Compare May 28, 2024 10:21

mark-koch requested a review from croyzor May 29, 2024 09:04

croyzor reviewed May 30, 2024

View reviewed changes

mark-koch added 4 commits May 30, 2024 10:33

Revert addition of nat type

0ea6ac2

Fix _NumericTypeDef docstring

506de84

Support float coercion for all numeric types

1a61ca8

Turn bool into a numeric type

078aa25

mark-koch changed the title ~~feat: Add nat type and turn int and float into core types~~ feat: Make bool a numeric type and turn int and float into core types May 30, 2024

mark-koch requested a review from croyzor May 30, 2024 12:49

croyzor reviewed May 30, 2024

View reviewed changes

guppylang/prelude/_internal.py Outdated Show resolved Hide resolved

croyzor reviewed May 30, 2024

View reviewed changes

mark-koch mentioned this pull request Jun 17, 2024

Add a type for unsigned integers #228

Closed

mark-koch and others added 3 commits June 18, 2024 17:14

Fix comment

999f121

Co-authored-by: Craig Roy <croyzor@users.noreply.github.com>

Merge remote-tracking branch 'origin/main' into feat/numeric-tys

13154dd

Undo turning bool into a numeric type

cf8a529

mark-koch changed the title ~~feat: Make bool a numeric type and turn int and float into core types~~ feat: Turn int and float into core types Jun 24, 2024

mark-koch requested a review from croyzor June 24, 2024 13:52

qartik mentioned this pull request Jun 24, 2024

Turn bool into a numeric type? #264

Open

croyzor reviewed Jun 24, 2024

View reviewed changes

croyzor approved these changes Jun 24, 2024

View reviewed changes

Remove int_type function

7389aa4

mark-koch merged commit 99217dc into main Jun 24, 2024
3 checks passed

mark-koch deleted the feat/numeric-tys branch June 24, 2024 16:03

hugrbot mentioned this pull request Jun 18, 2024

chore(main): release 0.6.0 #248

Merged

hugrbot mentioned this pull request Oct 21, 2024

chore(main): release 0.13.0 #518

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat: Turn int and float into core types #225

feat: Turn int and float into core types #225

mark-koch commented May 28, 2024 •

edited

Loading

croyzor May 30, 2024

croyzor May 30, 2024

croyzor May 30, 2024

croyzor May 30, 2024

mark-koch May 30, 2024

mark-koch May 30, 2024

croyzor May 30, 2024

croyzor left a comment •

edited

Loading

mark-koch commented May 30, 2024

croyzor commented May 30, 2024

codecov-commenter commented May 30, 2024 •

edited

Loading

croyzor May 30, 2024

mark-koch Jun 18, 2024

qartik commented May 30, 2024

mark-koch commented May 30, 2024

qartik commented May 30, 2024

mark-koch commented Jun 24, 2024

croyzor Jun 24, 2024

croyzor Jun 24, 2024

mark-koch Jun 24, 2024

croyzor Jun 24, 2024

	def get_instance_func(self, ty: Type \| TypeDef, name: str) -> CallableDef \| None:
	"""Looks up an instance function with a given name for a type.

	Returns `None` if the name doesn't exist or isn't a function.
	"""
	type_defn: TypeDef
	match ty:
	case TypeDef() as type_defn:
	pass
	case BoundTypeVar() \| ExistentialTypeVar() \| SumType():
	return None
	case NumericType(kind):
	match kind:
	case NumericType.Kind.Int:
	type_defn = int_type_def
	case NumericType.Kind.Float:
	type_defn = float_type_def
	case kind:
	return assert_never(kind)
	case FunctionType():
	type_defn = callable_type_def
	case OpaqueType() as ty:
	type_defn = ty.defn
	case StructType() as ty:
	type_defn = ty.defn
	case TupleType():
	type_defn = tuple_type_def
	case NoneType():
	type_defn = none_type_def
	case _:
	return assert_never(ty)

feat: Turn int and float into core types #225

feat: Turn int and float into core types #225

Conversation

mark-koch commented May 28, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

croyzor left a comment • edited Loading

Choose a reason for hiding this comment

mark-koch commented May 30, 2024

croyzor commented May 30, 2024

codecov-commenter commented May 30, 2024 • edited Loading

Codecov Report

Choose a reason for hiding this comment

Choose a reason for hiding this comment

qartik commented May 30, 2024

mark-koch commented May 30, 2024

qartik commented May 30, 2024

mark-koch commented Jun 24, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

mark-koch commented May 28, 2024 •

edited

Loading

croyzor left a comment •

edited

Loading

codecov-commenter commented May 30, 2024 •

edited

Loading