scip-python 0.6.6: SymbolInformation missing for @dataclass class symbols and stdlib symbols, causing scip expt-convert to hard-fail

[lichv7]

Summary

scip-python 0.6.6 (latest) does not generate SymbolInformation for @dataclass class symbols themselves (only for their fields/methods), nor for many standard library symbols. When scip expt-convert (v0.8.1) encounters a definition occurrence with an enclosing range for one of these missing symbols, it hard-errors with "has definition occurrence, but no SymbolInformation" and aborts the entire conversion, making the index unusable.

Reproduction

Minimal two-file project:

pyproject.toml

[project]
name = "scip-final-repro"
version = "0.1.0"

contracts.py

"""File A: defines a @dataclass with self-validation in __post_init__."""
from __future__ import annotations

from dataclasses import dataclass
from numbers import Real
import math


def _validate(value: object) -> None:
    if not isinstance(value, Foo):
        raise ValueError("must be Foo")
    if not isinstance(value.x, Real) or not math.isfinite(value.x):
        raise ValueError("x must be finite real")


@dataclass(frozen=True)
class Foo:
    x: float

    def __post_init__(self) -> None:
        _validate(self)

consumer.py

"""File B: imports and uses Foo from file A."""
from __future__ import annotations

from contracts import Foo


def make_foo(x: float) -> Foo:
    return Foo(x=x)

Steps

scip-python index . --output index.scip --quiet
scip expt-convert --output index.db index.scip

Observed

In the generated index.scip:

1. Foo# (the class symbol itself) has 8 occurrences in contracts.py, but SymbolInformation is only generated for 3 sub-symbols (Foo#x, Foo#__post_init__, Foo#__post_init__().(self)). The class-level Foo# symbol is missing from doc.symbols.

2. Foo# in consumer.py has 4 occurrences but zero SymbolInformation entries — cross-file references are completely missing symbol metadata.

3. Standard library symbols also missing SymbolInformation:

   • dataclasses/dataclass()
   • math/isfinite()
   • numbers/Real#
   • math/__init__:

Converter behavior

For this minimal repro, scip expt-convert succeeds (exit 0) because the missing symbols don't have enclosing ranges. However, in a large project (477 Python files, 160K occurrences), some missing symbols do have enclosing ranges, triggering the hard error in convert.go:395:

return fmt.Errorf("symbol %q has definition occurrence, but no SymbolInformation", occ.Symbol)

This aborts the entire conversion, producing only a 36KB partial database instead of the expected ~13MB.

Expected behavior

scip-python should generate SymbolInformation for:

1. @dataclass class symbols themselves (not just their fields/methods)
2. Standard library symbols referenced in the indexed code

Environment

• scip-python: 0.6.6 (latest on npm)
• scip converter: v0.8.1
• Python: 3.12.13
• Node.js: v22.22.3
• macOS: Darwin x86_64

Related issues

• Indexer fails to create symbols for assignments with type annotations #161 — "Indexer fails to create symbols for assignments with type annotations" (closed, not planned)
• Cross-package references silently dropped in src-layout projects (autoSearchPaths is never enabled) #221 — "Cross-package references silently dropped in src-layout projects" (open)
• scip-code/scip PR Find implementations (for parent classes) #36 — relaxed this check for lint but expt-convert still hard-fails
• scip-code/scip #464 — expt-convert missing fields population (open)

Workaround

We've written a patch script that parses the .scip protobuf, detects all symbols with occurrences but no SymbolInformation, and inserts minimal placeholder SymbolInformation entries before running expt-convert. This produces a complete ~13MB database and scip-cli analyze works correctly on the patched index.

[linear]

OSS-24217