feat: Add span and diagnostics base classes

[mark-koch]

Closes #536.

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 0% with 83 lines in your changes missing coverage. Please review.

Project coverage is 89.90%. Comparing base (37e1633) to head (56dec3e).

Files with missing lines	Patch %	Lines
guppylang/span.py	0.00%	42 Missing ⚠️
guppylang/diagnostic.py	0.00%	41 Missing ⚠️

Additional details and impacted files

@@                 Coverage Diff                  @@
##           feat/diagnostics     #548      +/-   ##
====================================================
- Coverage             91.09%   89.90%   -1.20%     
====================================================
  Files                    59       61       +2     
  Lines                  6264     6347      +83     
====================================================
  Hits                   5706     5706              
- Misses                  558      641      +83     

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

[croyzor]

Loc is meant to be 1-indexed, so it seems good that line_offset starts at 1?

[mark-koch]

But x.lineno also starts at 1

Both of these come from the python ast module 😅

[croyzor]

Why have this be a ClassVar and not just a field?

[mark-koch]

Otherwise Python will complain that non-default fields follow default ones when defining a subclass. E.g.

@dataclass(frozen=True)
class TypeMismatchError(Error):
    expected: Type
    actual: Type

    title: str = "Type mismatch"
    span_label: str = "Expected `{expected}`, got `{actual}`"

This is because the field order is dermined by the base classes, so the title and span_label constructor arguments would actually come before the expected and actual args. We would need to do

@dataclass(frozen=True)
class TypeMismatchError(Error):
    expected: Type
    actual: Type

    title: str = field(default="Type mismatch", init=False)
    span_label: str = field(defaut="Expected `{expected}`, got `{actual}`", init=False)

which would be quite cumbersome and looks noisy. Making them ClassVars tells the @dataclass decorator that these fields are static and shouldn't be initialised by the constructor.

[croyzor]

Okay, I think the disconnect is that you're intending for these classes to be inherited from rather than used directly, so it makes sense for a the TypeMismatchError subclass to have a class variable.