Rust: Regenerate models

[paldepind]

This PR regenerates the auto-generated models for Rust.

A few changes where necessary in order to re-run the model generator. Most of them should be understandable by the commit message, but "Disable dynamic dispatch when generating models" might require a bit more explanation:

The dynamic dispatch that we do for calls that resolve to trait functions caused some problematic sinks to pop up. As an example of where this happened consider parse:

pub fn parse<F: FromStr>(&self) -> Result<F, F::Err> {
    FromStr::from_str(self)
}

from the standard library. This essentially just wraps FromStr::from_str and dynamically dispatching at this call causes parse to get generated sinks from all implementations of FromStr.

To fix the problem this PR disables dynamic dispatch when running the model generator. This is achieved by creating a parameterized module around Rust's data flow library which allows one to disable dynamic dispatch. This feels a bit heavy-handed but actually didn't require that many changes and I can't think of another way of achieving the goal.

[paldepind]

Gen is for "generic" or something. Suggestions for better names are very welcome 🙏

[Copilot]

Pull request overview

This PR regenerates auto-generated models for Rust by disabling dynamic dispatch during model generation. The change prevents problematic sinks from appearing when trait function implementations are dynamically dispatched (e.g., preventing parse from inheriting sinks from all FromStr implementations).

Changes:

• Adds a parameterized module system to control dynamic dispatch in data flow analysis
• Updates model generation to disable dynamic dispatch when generating models
• Regenerates models for multiple Rust libraries (serde, smallvec, once_cell, reqwest)
• Updates test expectations to reflect the new generated models

Reviewed changes

Copilot reviewed 16 out of 30 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
ModelGeneratorImpl.qll	Adds applyReadStepsAsTaintSteps() predicate to control read step behavior
CaptureModels.qll	Creates parameterized modules for disabling dynamic dispatch, adds path filtering for semicolons
TaintTrackingImpl.qll	Converts to parameterized module to support dynamic dispatch control
Test expectations	Updates model references to match regenerated models
Generated models	Regenerates models with dynamic dispatch disabled
reqwest.model.yml	Adds exclusion for timeout field taint step

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[hvitved]

LGTM. Now that we generate more accurate models, these two manual models can be removed.

[hvitved]

private?

[paldepind]

It's used over in TaintTrackingImpl.qll.

[hvitved]

What does this achieve?

[paldepind]

Fixes a compile error on this line:

final class ParameterPosition extends TParameterPosition {

with the error being:

classes may not extend outside their parameterised module; consider extending through a final alias

[hvitved]

The dynamic dispatch that we do for calls that resolve to trait functions caused some problematic sinks to pop up.

@michaelnebel : Do you know if we have seen something similar in Java or C#?

[michaelnebel]

The dynamic dispatch that we do for calls that resolve to trait functions caused some problematic sinks to pop up.

@michaelnebel : Do you know if we have seen something similar in Java or C#?

I believe, we have seen issues like this for source generation, which we worked around by introducing the (poorly named) predicate irrelevantSourceSinkApi, which "ensures" a unique call chain (to get around the issue with dynamic dispatch).

[paldepind]

LGTM. Now that we generate more accurate models, these two manual models can be removed.

Nice, thanks!

[geoffw0]

Thanks for doing this.

I'm still not convinced we should be removing manual models when they are generated, as generated models are potentially more fragile / less trustworthy. Though the opposite argument is that they're easier to update when libraries change, and less prone to human error. I guess I don't feel we ever properly decided what the right approach is.

That aside, I'm very pleased to see the model generator working again, and pleased to see a bunch of new results on DCA! 🚀