fix(cli): normalize workspace path to forward slashes for CMake templates

[GuTS805]

Problem

workspace_dir() in binaries/cli/src/template/mod.rs resolved the dora
workspace root from env!("CARGO_MANIFEST_DIR") and returned it as a raw OS
path. On Windows that's backslash-separated (e.g. C:\Users\...). This
string was substituted verbatim — no escaping — into the __DORA_PATH__
placeholder in the C/C++ CMake templates
(binaries/cli/src/template/c/mod.rs, binaries/cli/src/template/cxx/mod.rs),
producing a generated CMakeLists.txt line like:

set(DORA_ROOT_DIR "C:\Users\alok0\...\dora" CACHE FILEPATH "Path to the root of dora")

CMake treats \ inside a quoted string as an escape character. \U, \D,
etc. are not recognized escapes, so cmake -B build failed immediately with
a parse error (Invalid character escape '\U'), before any compiler/generator
logic ran. This only affects dora new --lang c/cxx --internal-create-with-path-dependencies
(the flag used to build templates against a local checkout) — the default
dora new flow substitutes an empty string and git-clones instead, so it
never hit this.

This is why it was never caught in CI: .github/workflows/nightly.yml
exercises this exact dora new --lang c/cxx --internal-create-with-path-dependencies

• cmake -B flow, but is gated if: runner.os == 'Linux', where paths are
  already forward-slash.

Fix

workspace_dir() now normalizes \ to / before returning, since both
CMake and Windows accept forward slashes. Added a regression test
(workspace_dir_has_no_backslashes) asserting the returned path never
contains a backslash.

Verification

• Built the actual dora CLI and ran
  dora new test_c_project --lang c --internal-create-with-path-dependencies
  — the generated CMakeLists.txt now contains
  set(DORA_ROOT_DIR "C:/Users/.../dora" ...) (forward slashes).
• cmake -B build on the generated project no longer hits the
  backslash-escape parse error (progresses past it to the project() step;
  the only remaining error is an unrelated missing MSVC/nmake toolchain on
  this machine).
• cargo fmt --all -- --check — clean
• cargo clippy -p dora-cli -- -D warnings — clean
• cargo test -p dora-cli workspace_dir_has_no_backslashes — pass

Fixes #3038

[trunk-io]

😎 Merged successfully - details.

[phil-opp]

I reviewed both parts of this diff.

The CMake path normalization is correct: replacing \ with / yields a path that CMake (which treats \ as a string-escape character) and Windows both accept, and the C/C++ template callers are updated to borrow the now-owned String.

Two things worth raising:

1. An unrelated change is bundled in. Beyond the CMake templates, this PR also boxes CachedResult::Cached in binaries/coordinator/src/state.rs to silence a clippy::large_enum_variant warning. That is the same change already proposed on its own in fix(coordinator): box CachedResult::Cached to fix large_enum_variant on Windows #3001, so the two PRs overlap and will conflict, and per the repo convention ("Don't fix unrelated warnings in PRs") it doesn't belong in a CMake-path fix. I'd drop it here and let fix(coordinator): box CachedResult::Cached to fix large_enum_variant on Windows #3001 carry it (or close whichever is redundant). The boxing itself is correct and behavior-preserving — all CachedResult::Cached sites still compile (the send_result_to(&result, …) call happens before the value is boxed, and the is_ok()/is_err() guards work through the Box).

2. The added test can't catch the regression on PR CI. workspace_dir_has_no_backslashes is trivially true on Linux (Linux paths contain no \), so the Linux-only PR CI never exercises the replace('\\', "/") path — the fix is only really guarded on the Windows nightly job. Asserting the replacement against a constructed sample path that contains backslashes would give the test teeth on every platform.

---

🤖 Fully automated review by Claude (Claude Code) — no human has reviewed these findings. Please verify before relying on them.

Generated by Claude Code

---

Generated by Claude Code

[GuTS805]

Thanks for the review. Addressed both points:

1. Dropped the unrelated coordinator change. Reverted the CachedResult::Cached
   boxing in binaries/coordinator/src/state.rs — you're right that it duplicates
   fix(coordinator): box CachedResult::Cached to fix large_enum_variant on Windows #3001, so I'll let that PR own it and keep this one scoped to the CMake path fix.

2. Made the test OS-independent. Pulled the normalization logic into a
   normalize_for_cmake() helper and changed the test to assert against a
   hardcoded Windows-style sample path (C:\Users\example\dora) instead of the
   live workspace_dir() output. It now actually exercises the \ → /
   replacement on every platform instead of trivially passing on Linux CI.

Pushed both fixes. Diff is back to just the 3 template files.

[phil-opp]

🤖 Automated review by Claude — fully automated review, not verified by a human.

Reviewed the latest commit (b45176f). Both points from the prior review are addressed: the unrelated CachedResult::Cached boxing in binaries/coordinator/src/state.rs is dropped, and the regression test now runs normalize_for_cmake(r"C:\Users\example\dora") against a hardcoded Windows-style path, so it exercises the \ → / replacement on every platform instead of passing trivially on Linux CI. The diff is scoped to the three template files, and the &super::workspace_dir()? borrow updates match the new String return type. No new issues found.

---

Generated by Claude Code

[github-actions]

@GuTS805 the Trunk merge queue failed for this PR.

See the Trunk merge-status comment for details.

Posted as a new comment so GitHub sends an email — Trunk's sticky comment is edited in place and won't trigger a notification.