ci: skip expensive jobs for docs-only changes

[benvinegar]

Summary

• add a shared CI change detector for docs/assets/LICENSE-only updates
• keep PR/main/Nix workflows running a cheap detector while skipping expensive jobs when there are no code changes
• gate Windows compatibility, typecheck/tests, package, prebuilt, binary, and Nix jobs on code changes

Testing

• bun run format:check
• manual detect-code-changes.sh checks in a temporary git repo for docs-only and code-change scenarios

This PR description was generated by Pi using OpenAI GPT-5

[greptile-apps]

Greptile Summary

This PR replaces per-workflow paths-ignore filters with a shared detect-code-changes.sh script and a lightweight changes gating job that runs first in each workflow, skipping all expensive jobs when only docs/assets/LICENSE files are touched.

• A new detect-code-changes.sh script compares base and head SHAs via git diff --name-only, handles shallow clones by fetching missing commits, and writes code_changed=true/false to GITHUB_OUTPUT.
• Three workflows (ci.yml, pr-ci.yml, nix.yml) each gain a changes job; all expensive jobs (typecheck, tests, packaging, Nix build, Windows compat) are gated with if: needs.changes.outputs.code == 'true'.
• Each workflow correctly resolves the base SHA for its event type: github.event.before for push-to-main events and github.event.pull_request.base.sha for pull-request events.

Confidence Score: 4/5

Safe to merge — the change only affects when expensive CI jobs run, not how the project is built or tested.

The logic is sound: the script correctly handles shallow clones, zero SHAs, and both push/PR event contexts. The only issue is a pair of unreachable docs/** and assets/** alternatives in the case statement that can never be reached after the preceding docs/*/assets/* arms — cosmetic but harmless.

.github/scripts/detect-code-changes.sh has the redundant case patterns; all three workflow files look correct.

Important Files Changed

Filename	Overview
.github/scripts/detect-code-changes.sh	New script that diffs base/head SHAs and outputs code_changed; handles shallow clones and zero SHAs correctly, but has redundant docs/**/assets/** case patterns (unreachable after docs/*/assets/*).
.github/workflows/ci.yml	Replaces paths-ignore with a changes gating job; all expensive jobs correctly depend on needs.changes.outputs.code == 'true'. BASE_SHA from github.event.before is appropriate for push events and the zero-SHA edge case is handled by the script.
.github/workflows/nix.yml	Adds the same changes job pattern; correctly uses a ternary expression to pick pull_request.base.sha vs github.event.before depending on the trigger event.
.github/workflows/pr-ci.yml	Adds changes job using github.event.pull_request.base.sha as the base, which is the correct anchor for a PR diff; all expensive jobs properly gated.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Push to main / PR opened] --> B[changes job\ndetect-code-changes.sh]
    B --> C{code_changed?}
    C -- false\ndocs/assets/LICENSE only --> D[All expensive jobs SKIPPED]
    C -- true\ncode files touched --> E[validate / pr-validate]
    C -- true --> F[tty-smoke]
    C -- true --> G[pack-npm / prebuilt-npm]
    C -- true --> H[build-bin]
    C -- true --> I[windows-compat]
    C -- true --> J[Nix package]
    subgraph detect-code-changes.sh
        K[resolve zero base SHA → empty tree] --> L[fetch missing SHAs\nif shallow clone]
        L --> M[git diff --name-only\nbase..head]
        M --> N{any non-docs path?}
        N -- yes --> O[code_changed=true]
        N -- no --> P[code_changed=false]
    end

Loading

Prompt To Fix All With AI

Fix the following 1 code review issue. Work through them one at a time, proposing concise fixes.

---

### Issue 1 of 1
.github/scripts/detect-code-changes.sh:22
In bash `case` patterns, `*` matches any character sequence **including** `/`, so `docs/*` already matches `docs/subfolder/file.png` and any deeper path. The `docs/**` and `assets/**` alternatives are therefore dead branches — they can never be reached because `docs/*` always matches first and exits the case block. The same applies to `assets/**` following `assets/*`.

```suggestion
    *.md | docs/* | assets/* | LICENSE)
```

_{Reviews (1): Last reviewed commit: "ci: skip expensive jobs for docs-only ch..." | Re-trigger Greptile}

[greptile-apps]

In bash case patterns, * matches any character sequence including /, so docs/* already matches docs/subfolder/file.png and any deeper path. The docs/** and assets/** alternatives are therefore dead branches — they can never be reached because docs/* always matches first and exits the case block. The same applies to assets/** following assets/*.

Suggested change

	*.md | docs/* | docs/** | assets/* | assets/** | LICENSE)
	*.md | docs/* | assets/* | LICENSE)

Prompt To Fix With AI

This is a comment left during a code review.
Path: .github/scripts/detect-code-changes.sh
Line: 22

Comment:
In bash `case` patterns, `*` matches any character sequence **including** `/`, so `docs/*` already matches `docs/subfolder/file.png` and any deeper path. The `docs/**` and `assets/**` alternatives are therefore dead branches — they can never be reached because `docs/*` always matches first and exits the case block. The same applies to `assets/**` following `assets/*`.

```suggestion
    *.md | docs/* | assets/* | LICENSE)
```

How can I resolve this? If you propose a fix, please make it concise.

[benvinegar]

Good catch — simplified the Bash case pattern to remove the unreachable docs/** / assets/** alternatives while preserving recursive matching via docs/* and assets/*.

This comment was generated by Pi using OpenAI GPT-5