[tausbn]

Top-level rule guards currently get a mutable copy of the ctx user context. If the guard succeeds, this copy is used to replace the existing context and the rule body is executed.

However, this design is somewhat problematic: first of all, it's questionable to do such mutation in the first place, and secondly if we ever want to embed guards deeper inside of a rule, then for consistency we would expect these to be mutable as well. This would entail a lot of unnecessary context cloning and bookkeeping for a "feature" that will likely never be used.

Thus, I opted to simply make the access to the user context read-only inside of the guard body (as in: we only pass it an immutable reference).

[Copilot]

🟡 Changes recommended

The test duplicates existing enabled-context coverage, leaving nested context propagation and caller isolation untested.

1 open issue

• Test nested context propagation and caller isolation · New

What changed in this PR

Makes YEAST rule guards borrow user context immutably, avoiding context cloning for rejected guards while preserving private, mutable context clones for transforms.

Changes:

• Change the guard API and generated closures to use immutable context references.
• Move context cloning until after a guard succeeds.
• Update tests, documentation, and examples for read-only guard access.

File	Description
shared/​yeast/​tests/​test.rs	Adapt guard tests to read context without mutation.
shared/​yeast/​src/​lib.rs	Make guards immutable and delay transform-context cloning.
shared/​yeast/​doc/​yeast.md	Explain read-only context access and revise the example.
shared/​yeast-macros/​src/​lower.rs	Generate immutable context parameters for guard closures.
shared/​yeast-macros/​src/​lib.rs	Update macro documentation to describe read-only guards.

_{Review effort: Max}

[paldepind]

LGTM. If we don't need the mutability, then this is definitely simpler.