Hacktoberfest 2026: the issues maintainers tagged for October, open and beginner-friendly. Browse Hacktoberfest issues

RCA: `set () = { … }` drops its value's dynamic kind; code is lost

Open
  • Is something wrong?
#4,002 0 comments 0 reactions 0 assignees View on GitHub

Maintainers usually reply within 1 day

@domorale is already working on this.

Since Oct 22, 2026.

  • #4004 by @domorale — open

Assessment

Difficulty
4/5
Estimated time
3-5 days
Newbie friendliness
20/100
Issue type
Bug
Clarity
Clearly specified
Activity status
Stale
Tech stack
rust
Domain
compilers

Research direction

Start with qsc_rca/src/core.rs, especially analyze_expr_assign around lines 149–171, and trace how assignment kinds are computed from the assignee and value. The issue reports that an open pull request (#4004) already addresses this bug. The described fix should restore dynamic features for assignments to (); the two repros should then produce 5 and true on Adaptive_RI, Adaptive_RIF, and Adaptive.

Written by the indexing model from the issue text.

Description

bug needs triage

Summary

RCA computes the kind of an assignment only from the locals it updates. set () = … updates no local, so the statement is Static even when its value contains measurement-dependent code. Partial evaluation then evaluates the block classically: a store into a mutable and gates inside the block are silently dropped. This gives wrong results on Adaptive_RI, Adaptive_RIF and Adaptive.

Minimal repro

operation Main() : Int {
    use q = Qubit();
    X(q);
    let d = MResetZ(q) == One;
    mutable z = 3;
    set () = { if d { set z = 5; } () };   // assigns nothing; means: if d { set z = 5; }
    return z;
}

Gate variant (the X(t) is dropped):

operation Main() : Bool {
    use (q, t) = (Qubit(), Qubit());
    X(q);
    let d = MResetZ(q) == One;
    set () = { if d { X(t); } () };
    return MResetZ(t) == One;
}

Run with python check.py prog.qs "Main()"; check.py is in #4001. It runs the program on the simulator, then compiles it for Adaptive_RI, Adaptive_RIF and Adaptive and runs the QIR with qir-runner, one shot each.

Expected vs actual

Program Simulator (expected) Adaptive_RI / Adaptive_RIF / Adaptive
store 5 3
gate variant true false

Output on main 1b8de0d17 (same on the earlier main commits a90b8f777 and 48449939c and on open PR #3902 577dc9382; qir-runner 0.9.7):

### store
simulator: 5
Adaptive_RI: OUTPUT	INT	3	0_i
Adaptive_RIF: OUTPUT	INT	3	0_i
Adaptive: OUTPUT	INT	3	0_i
### gate variant
simulator: True
Adaptive_RI: OUTPUT	BOOL	false	0_b
Adaptive_RIF: OUTPUT	BOOL	false	0_b
Adaptive: OUTPUT	BOOL	false	0_b

The results are the same with #3977 (head 6228ff9ee): that PR does not change this path.

Root cause

analyze_expr_assign builds the statement's kind only from update_locals_compute_kind, the kind of the locals in the assignee (qsc_rca/src/core.rs#L149-L171, lines 160–169). With the assignee (), no local is updated, so the value's runtime features are discarded and the statement is Static.

History (Adaptive_RI): at #3733 (95c039f4b) and #3885 (45d8e3038) both programs were rejected with PartialEval.EvaluationFailed; since #3887 (96388a0da) they compile and give the wrong answer.

Suggested fix

In analyze_expr_assign, also aggregate the value expression's runtime features into the assignment's kind. For an assignee without () parts this adds nothing new, since the value's kind already reaches the statement through the updated locals (reasoned; consistent with no existing expect-test changing); for () (or a tuple with a () element) it restores the value's dynamic features.

I have this implemented (9 lines in core.rs, plus tests) and will open a PR. With it, the repros give 5 and true on Adaptive_RI, Adaptive_RIF and Adaptive. It is independent of #3977 and merges with it without conflicts.

How it was found

While proving properties of a formal model of the lowering: the proof required the kind of an assignment to include the kind of its value. Random differential testing did not find it (our program generator has no set ()).

Related issues

  • No existing issue found. Searches: "set () unit", "unit tuple assignment", "assign to unit", "set () unit tuple", "assign unit tuple dynamic".
  • #3887: the partial-evaluation change after which these programs compile to wrong QIR instead of being rejected.
  • #3976 / #3977: a different RCA bug in the right operand of a dynamic and/or; #3977 does not fix this one.
Dominant language
Rust
Stars
1k
Forks
219
Avg merge
3d 19h
Merged PRs (30d)
93

Getting set up

Open in Codespaces

Starts the project's dev container in your browser, under your own GitHub account.

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from microsoft/qdk

All issues in microsoft/qdk

Similar issues

More Rust issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.