[Bug]: Playwright workers force ANSI colors via FORCE_COLOR=1, breaking color-disable behavior in CI/CD

[veged]

Version

@playwright/test 1.58.2

Steps to reproduce

@playwright/test 1.58.2 (also reproducible on current main source where workerHost.ts still sets FORCE_COLOR: '1')

1. Set CI no-color env in parent process, for example:
   • CI=1
   • FORCE_COLOR=0
   • NO_COLOR=1 (or just FORCE_COLOR=0)
2. Run Playwright tests with a worker-based run:
   • npx playwright test --reporter=dot
3. Log env in test config (or worker-side) and compare parent vs worker process values.
4. Observe source behavior in worker bootstrap:
   • workerHost.ts#L52

Expected behavior

Actual behavior

Environment

Если хочешь, сразу дам ещё ultra-short вариант (совсем компактный, без лишнего текста) для вставки в issue form.

Expected behavior

Playwright workers should respect user/CI color settings inherited from parent process (e.g. keep FORCE_COLOR=0 when explicitly set), or at least not override them unconditionally.

Actual behavior

Parent process starts with FORCE_COLOR=0, but Playwright worker processes are started with FORCE_COLOR=1 due to hardcoded env in workerHost.ts. This re-enables colored output in CI and can conflict with no-color setups.

Additional context

No response

Environment

- Playwright: `@playwright/test 1.58.2`
- OS: Linux CI runner
- Node.js: `v24.14.1`
- CI env: `CI=1`, `FORCE_COLOR=0`, `PLAYWRIGHT_FORCE_TTY=0` (and in some setups `NO_COLOR=1`)
- Reporter: `dot`

[Skn0tt]

This is intended behaviour from our end, so that CI artifacts like the HTML report have coloured output. For the CI logs itself, colours are stripped by the host process. Are you observing a bug around this?

[veged]

This is intended behaviour from our end, so that CI artifacts like the HTML report have coloured output. For the CI logs itself, colours are stripped by the host process. Are you observing a bug around this?

Yes, I think the bug is specifically that workers unconditionally override an explicit no-color setting from the parent process.

Having colored output by default for HTML artifacts is fine, but there still needs to be a way to disable ANSI colors for environments where colors are unsupported or undesirable. Host-side stripping only addresses one path (CI log rendering), while worker output can also be consumed by other tools/integrations that expect plain text and should not receive ANSI escapes.

So the request is mainly to preserve an explicit opt-out such as FORCE_COLOR=0 (and ideally NO_COLOR=1), instead of always forcing FORCE_COLOR=1.

Also, right next to this line, DEBUG_COLORS already respects the environment when it is set, instead of overriding it unconditionally. FORCE_COLOR could follow the same pattern: default to 1 only when unset, but keep the inherited value when the user/CI explicitly disables colors.

[pavelfeldman]

Host-side stripping only addresses one path (CI log rendering), while worker output can also be consumed by other tools/integrations that expect plain text and should not receive ANSI escapes.

escapes should not reach other tools / integrations. can you provide an example?

[veged]

Yes — this is the bug I mean.

I made a minimal public repro here:

• https://sourcecraft.dev/veged/playwrite-bug

And this CI run shows it on a clean remote environment:

• https://sourcecraft.dev/veged/playwrite-bug/cicd/runs/1

In that run, please look at the logs of the repro cube. The relevant line is:

worker-child {"forceColor":"1","noColor":"1","sample":"\\u001b[31mRED\\u001b[39m"}

So even though the parent process starts with FORCE_COLOR=0 and NO_COLOR=1, the Playwright worker still gives its child process FORCE_COLOR=1, which re-enables ANSI-colored output.

This is also visible locally with the same minimal repo:

npm ci
npx playwright install chromium
npm run repro

So the issue is not just host-side CI log rendering — there is currently no way to keep colors disabled inside the worker process tree when the environment explicitly opts out.

[pavelfeldman]

These links are bad (do not open).

When you pass FORCE_COLOR=0 or NO_COLOR=1 we just promise you to not send ansi color into your stdout/stderr, which we hold. The fact that there are worker processes is Playwright's implementation detail. If you CI is getting ansi in stdout despite passing FORCE_COLOR=0, please file an issue with a repro.

[veged]

@pavelfeldman Thanks, this clarifies the distinction you are making, and I think I understand it now.

If the intended contract is only: when FORCE_COLOR=0 / NO_COLOR=1 is set, Playwright should not emit ANSI into the final stdout / stderr that the runner host process owns, then I agree that my earlier repro was not sufficient by itself. That one proved the internal env override inside the worker process tree, but not yet a visible contract violation in the final CI logs.

To address exactly that point, I made a GitHub-native repro:

• repo: https://github.com/veged/playwright-force-color-repro
• Actions run: https://github.com/veged/playwright-force-color-repro/actions/runs/24140550588

In the repro job, in the Run npm run repro step, the relevant lines are:

worker-child {"forceColor":"1","noColor":"1","sample":"\\u001b[31mRED\\u001b[39m"}
worker-raw �[31mRAW_RED�[39m

The first line is the internal evidence from before: even though the parent process starts with FORCE_COLOR=0 and NO_COLOR=1, the Playwright worker still gives its child process FORCE_COLOR=1.

The second line is the reason I still think this is a real bug: a child process spawned from inside the Playwright worker writes raw ANSI to stdout, and that ANSI reaches the final GitHub Actions logs. So in this setup, host-side stripping does not seem to be preventing escapes from reaching the final CI stdout / stderr.

That is also the practical problem for me. I intentionally need to spawn isolated child executors from the worker when running not-fully-trusted code, for example when checking student homework submissions. In that workflow, preserving an explicit no-color opt-out matters because those isolated runs should inherit the caller's plain-text logging policy.

So I think there are two separate but related issues:

1. Playwright workers override an explicit parent opt-out (FORCE_COLOR=0 / NO_COLOR=1).
2. In at least this GitHub Actions repro, raw ANSI from a worker child process reaches the final CI logs.

If helpful, I can reduce this even further, but I wanted first to answer your narrower contract interpretation directly with a GitHub-native repro.