v3: clean up transform helpers/inits; cmd/v: fix v3.v bootstrap dispatch on macOS

[medvednikov]

What

Two related changes that came out of resolving the XTODO markers in vlib/v3/transform/transform.v and then verifying self-host performance.

1. v3: transform helper/init cleanup (resolves the 3 XTODOs)

• Remove the arr1/arr2/arr3/arr4 wrappers and node_kind_id, replacing every call site in the transform module with plain array literals ([a, b, ...]) and int(node.kind).
• Drop the empty map[...]{} initializers from new_transformer_view — V auto-initializes those struct map fields, so only the AST/type-checker views, used_fns, and the pointer-backed lookup caches (which would otherwise be nil) need explicit values.
• Keep the V3_STR_CAP experimental autostr-cap override (documented in fn.v) and replace the // XTODO is this needed marker with an explanatory comment.

2. cmd/v: route a bare v3.v bootstrap to the compat compiler on macOS

The macOS dispatch gates only recognized the v3 bootstrap by the literal path vlib/v3/v3.v. Building it as a bare v3.v from inside vlib/v3 (e.g. ../../v -prealloc -prod -o v3 v3.v) was treated as an ordinary .v build and delegated to the embedded V3 driver instead of the compatibility compiler (which then crashes and only survives via the auto-fallback retry). A shared is_macos_v3_compiler_bootstrap() helper now also resolves a file literally named v3.v through its real path, and both is_macos_v3_relevant_command and macos_v3_force_requested use it so they can't drift. The extra real-path lookup only runs for a v3.v target.

Verification

• v fmt -verify clean on all changed files; v3 builds.

• Transform unit tests (v test vlib/v3/transform/) pass 7/7.

• New dispatcher test test_macos_v3_compiler_bootstrap_is_detected_from_any_cwd; before/after, the bare-v3.v invocation goes from "embedded V3 invoked → segfault → fallback" to a clean compat build.

• No self-host performance regression (-prod, 5 runs each, baseline @ HEAD vs. this branch):

  metric	baseline	this PR
  transform, median	121.4 ms	118.6 ms
  transform, min	111.4 ms	109.3 ms
  transform RSS	~1.09 GB	~1.09 GB
  cgen, mean	~152.8 ms	~152.9 ms
  prod v3 binary size	14,096,904 B	14,096,904 B (byte-identical)

🤖 Generated with Claude Code

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 562f9f5c48

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Ascend one more directory before testing the bare path

With this test located at cmd/v/macos_v3_test.v, two os.dir calls resolve repo_root to <repo>/cmd, so v3_dir becomes <repo>/cmd/vlib/v3. The existence guard is therefore false in a normal checkout, and the key bare and dotted assertions never run, allowing the exact bare-v3.v dispatch regression this test targets to pass unnoticed.

Useful? React with 👍 / 👎.

[medvednikov]

Good catch — the two os.dir calls resolved repo_root to <repo>/cmd, so the vlib/v3 existence guard was always false and the key assertions never ran.

Fixed in 3b22b60 by dropping the @FILE-relative guard entirely: the test now builds an isolated <tmp>/vlib/v3/v3.v (plus a sibling non-bootstrap v3.v) and chdirs into it, so the real-path resolution is exercised unconditionally regardless of where the test file lives. I also confirmed the test now fails on assert bare if the helper's real-path branch is removed, so the regression can no longer pass unnoticed.