Remove legacy HFS writer

[bernalde]

Summary

• Removed the unused legacy HFS writer source under src/wrapper/hfs.
• Removed source and Chimera doc comments that described the disabled HFS writer.
• Added a Chimera wrapper test guard that the HFS writer is not shipped or exported.

Tests

• julia --project=. -e 'using Pkg; Pkg.instantiate(); include("test/chimera.jl")' passed.
• julia --project=. -e 'using Pkg; Pkg.test()' passed.

Notes: local tests warned that DWAVE_API_TOKEN is unset, so the DWave.Optimizer cloud tests were skipped by the existing test harness.

Branch Hygiene

• Base branch: main
• Source branch point: 091d47e (origin/main after git pull --ff-only origin main)
• Stacked status: not stacked; branch created directly from main
• Prerequisite PRs: none

Closes #48

[bernalde]

Blocking issues: None.

Nonblocking issues: None.

Questions: None.

Tests run and outcomes:

• Verified PR #50 against issue #48; the removal path fully resolves the requested decision and Closes #48 is appropriate.
• Confirmed the local merge-base diff totals match GitHub: 5 files changed, 5 insertions, 64 deletions.
• julia --project=. -e 'using Pkg; Pkg.instantiate(); include("test/chimera.jl")' passed in clean worktree /tmp/dwave-pr50-review-50.
• julia --project=. -e 'using Pkg; Pkg.test()' passed in clean worktree /tmp/dwave-pr50-review-50. Local DWave.Optimizer cloud tests were skipped by the existing harness because DWAVE_API_TOKEN is unset.
• CI is green on Julia 1 and 1.10 across Ubuntu and Windows.

Merge-readiness: I found no actionable issues and would consider this merge-ready. Because I authored this PR, this review is submitted as COMMENT; formal approval must come from another maintainer.

[bernalde]

Review of #50 (remove legacy HFS writer, Closes #48).

Note on event type: I am the PR author, so GitHub only permits a COMMENT review. A formal APPROVE / REQUEST_CHANGES must come from another maintainer. There are no blocking issues.

Intent vs issue #48

Issue #48 asked to either remove src/wrapper/hfs/ or revive it behind a guarded load with a round-trip test. This PR takes the remove path, which is a valid resolution. The writer was already dead code: architecture.jl has not included hfs/format.jl since #46, and hfs/printer.jl did not compile against the current API (it called a nonexistent Chimera(model, ...) constructor and read chimera.linear_terms/quadratic_terms fields the struct does not have). So removal carries no behavioral risk. Closes #48 is appropriate.

Verification

• Confirmed the removal is complete: git grep for hfs, HFS, write_model, :hfs, AbstractFormat at head finds no remaining references except the legitimate [^alex1770] citation in the Chimera docstring (the source of the coordinate convention, not writer code).
• architecture.jl has no orphan include after removing the trailing comment.
• src/wrapper/hfs/ contained only format.jl and printer.jl, so deleting both removes the directory; verified absent at head.
• src/wrapper/README.md does not mention HFS (rewritten in #46), so no stale doc reference.

Nonblocking

1. The new guard test is fine but lightly coupled to the source tree. !isdir(joinpath(pkgdir(DWave), "src", "wrapper", "hfs")) asserts on repo layout rather than behavior, so it would spuriously fail or pass if the wrapper directory is ever reorganized; and !isdefined(DWave, :HFS) is trivially true today since HFS was never loaded into the module. Both are harmless as regression guards against re-introduction, so this is optional - consider whether the directory-existence assertion earns its keep, or rely on the isdefined symbol guard alone.

Observation (pre-existing, optional follow-up)

With the writer gone, the Chimera docstring still opens with "The format of the instance-description file starts with a line giving the size of the Chimera graph" - describing a file format the package no longer reads or writes. That line is pre-existing and not modified by this PR, so it is out of scope here, but a future docs pass could reframe it to document the coordinate convention rather than implying the package handles that file format.

Tests run (clean worktree at head d412614)

• julia --project=. test/chimera.jl: all pass - new "Legacy HFS writer is removed" guard 2/2, plus 13/13, 12/12, dnx-equivalence 1/1, 4/4, 7/7.
• julia --project=. -e 'using Pkg; Pkg.test()': passed (exit 0); package precompiles and loads after the removals; cloud DWave.Optimizer tests skipped (no token, expected).

Merge-readiness

No blocking issues. Clean, complete, well-scoped removal of dead code with a regression guard; tests pass on a clean checkout. Mergeable on the merits, pending another maintainer's formal verdict.

[bernalde]

Commits pushed:

• f5cc87b test: avoid HFS source layout assertion (#50)

Main changes made:

• Removed the pkgdir(DWave)/src/wrapper/hfs directory-existence assertion from the HFS regression test.
• Kept the durable module-level guard that DWave.HFS is not loaded.
• Renamed the testset to match the remaining assertion.

Tests run and results:

• julia --project=. -e 'using Pkg; Pkg.instantiate(); include("test/chimera.jl")' passed.
• julia --project=. -e 'using Pkg; Pkg.test()' passed. The existing local cloud tests skipped because DWAVE_API_TOKEN is unset.
• No separate lint or type-check command is documented in this repo beyond the Julia test workflow.

Comments intentionally not addressed:

• None.

Remaining risks or follow-up items:

• GitHub Actions restarted for f5cc87b and was still in progress when this was posted.

[bernalde]

Blocking issues: none.

Nonblocking issues: none.

Questions: none.

Tests run:

• Targeted PR-head Chimera tests passed: include("test/chimera.jl").
• Full PR-head package suite passed: Pkg.test(), including the live D-Wave quantum sampler checks.
• Targeted merge-result Chimera tests passed against current main, resolving with DWave v0.7.1, QUBOTools v0.15.1, and QUBODrivers v0.6.4.

Merge-readiness: the PR diff matches GitHub's reported 5 changed files, +4/-64; Closes #48 is complete for the removal path; a merge simulation against current main was clean; CI is green. This is approvable on the merits, but because this PR is authored by bernalde, formal approval needs to come from another maintainer.

[bernalde]

Reviewed as maintainer. Note: I am the PR author, so GitHub only permits a COMMENT event here; a formal APPROVE/REQUEST_CHANGES must come from another maintainer. There are no blocking issues, so this is approvable on the merits.

Scope and intent

The PR removes the legacy HFS writer (src/wrapper/hfs/format.jl, printer.jl), the dead include-tracking comment in src/wrapper/architecture.jl, and the related Chimera doc comment, and adds a guard test. This matches one of the two paths issue #48 explicitly offered ("remove src/wrapper/hfs/ and the commented include if there is no near-term plan to revive the writer"). Closes #48 is appropriate: the issue asks for a decision, and removal fully resolves it, including its request that the architecture.jl comment no longer imply an opt-in mechanism exists.

Correctness verification

• Confirmed the removed code was dead, not live: on the merge-base (091d47e), src/wrapper/architecture.jl includes only device.jl, topology.jl, and chimera.jl. The hfs/ files were never included, so HFS, format(::Val{:hfs}), and write_model(..., ::HFS) were never in DWave's namespace. Removing them changes no public or loaded behavior.
• Confirmed no dangling references remain at the head: git grep -i hfs over the whole tree returns only the new test. The src/wrapper/hfs/ directory is fully gone. No docs, includes, or format/write_model call sites reference it.

Nonblocking observations (not actionable, recorded for context)

• The new guard Test.@test !isdefined(DWave, :HFS) would already have passed before this PR (the symbol was never defined), so it is a forward regression guard rather than a test of behavior changed by this PR. That is reasonable and the right intent.
• The second commit ("test: avoid HFS source layout assertion") deliberately replaced a source-file-layout assertion with the symbol-definition check. That is the better choice (asserts behavior, not on-disk layout), and I would not suggest reverting to a path-based check.

Tests run

• Clean detached worktree at the PR head (f5cc87b), isolated JULIA_DEPOT_PATH, DWAVE_API_TOKEN unset.
• This PR does not touch Project.toml, so it resolved against the base compat (QUBOTools 0.13.x), which is the intended environment for this branch.
• Pkg.test() passed (exit 0). The new "Legacy HFS writer is not loaded" testset passed (1/1), and the full Neal/Greedy/Random/Tabu conformance, parity, metadata, chimera, and topology testsets passed. The hardware-backed DWave.Optimizer set is token-gated and was skipped, matching the PR description.

Merge-readiness

• No blocking issues. Diff verified against merge-base 091d47e: 5 files, +4/-64, matching the PR's reported totals. Clean, well-scoped dead-code removal with a sensible forward guard. Approvable on the merits; because this is my own PR, the formal approval must come from a second maintainer.