Inconsistent inference results unless `jedi.Script` is reinitialized per token

[jatinarora96]

Hi,

I'm working on extracting references from a Python codebase using Jedi. For each token in a file, I use jedi.infer(line, column) to get inference results.

Here's a simplified version of my setup:

class JediInferenceEngine:
    """Performs Jedi inference on a given Python file."""

    def __init__(self, file_path, project_root):
        self.file_path = file_path
        self.project_root = project_root
        try:
            raw_content = file_path.read_text(encoding="utf-8")
            self.source = raw_content.encode('ascii', 'ignore').decode('ascii')
            self.source_lines = self.source.splitlines()
            self.script = jedi.Script(
                code=self.source,
                path=str(file_path),
                project=jedi.Project(path=str(project_root))
            )
        except Exception as e:
            logger.error(f"Failed to initialize JediInferenceEngine for {file_path}: {e}")
            raise

    def safe_jedi_infer(self, line, col, token_str=None):
        try:
            if line < 1 or line > len(self.source_lines):
                return []
            line_content = self.source_lines[line - 1]
            col = max(0, min(col, len(line_content)))
            return self.script.infer(line=line, column=col)
        except Exception as e:
            logger.warning(f"Jedi inference failed at {self.file_path}:{line}:{col} - {e}")
            return []

In my analysis loop, I initially reused the same JediInferenceEngine instance for all tokens in a file. However, I noticed that some references were missing. When I reinitialize the JediInferenceEngine for each token, the inferences are accurate and complete.

for token in tokens:
    self.jedi_engine = JediInferenceEngine(self.file_path, self.project_root)
    results = self.jedi_engine.safe_jedi_infer(line, col, token_str)
    ...

This behavior seems unexpected. Is there any internal caching or state in jedi.Script that could cause inference results to vary unless reinitialized? Or am I missing something in how Jedi is intended to be used?

Any insights would be appreciated!

[davidhalter]

This behavior seems unexpected. Is there any internal caching or state in jedi.Script that could cause inference results to vary unless reinitialized? Or am I missing something in how Jedi is intended to be used?

Yes. You have discovered Jedi's limitations :). There are a couple different ways for Jedi to limit the time it takes to calculate results and apparently you are running into them. I'm happy to review pull requests if you reset these defaults to reasonable values on every infer call. But it's kind of hard, because Jedi has not really been designed for this kind of stuff. It's been mostly designed for working in an IDE where the content of Script always changes.