[BUG] PDF Viewer does not work in headless Chromium and leads to download the file instead

[klhex]

Context:

• Playwright Version: 1.10.0
• Operating System: MacOS 11.3
• Node.js version: 15.11.0
• Browser: Chromium (works fine w/ Firefox)

Code Snippet

1. Create the following HTML file in your local webserver's DocumentRoot:

<a href="test.pdf" target="_blank">Link to PDF file</a>

2. Put a PDF file with the name "test.pdf" in the same directory.

3. Run the following PW code:

const { chromium } = require('playwright');

(async () => {
  const browser = await chromium.launch({
    headless: true,
  });
  const context = await browser.newContext();
  const page = await context.newPage();
  await page.setViewportSize({
    width: 1200,
    height: 1200,
  });

  await page.goto('localhost/a.html');

  const [pagePDF] = await Promise.all([
    context.waitForEvent('page'),
    await page.click('text=Link to PDF file'),
  ]);

  await pagePDF.waitForLoadState();
  console.log('Load state reached.');
  await browser.close();
})();

Describe the bug

After 30 seconds you will get the following error message:

/[...]/node_modules/playwright/lib/client/waiter.js:57
            const result = await Promise.race([promise, ...this._failures]);
                           ^
TimeoutError: page.waitForLoadState: Timeout 30000ms exceeded.
=========================== logs ===========================
  "networkidle" event fired
============================================================
Note: use DEBUG=pw:api environment variable to capture Playwright logs.
    at /[...]/node_modules/playwright/lib/client/waiter.js:49:51
    at Waiter.waitForPromise (/[...]/node_modules/playwright/lib/client/waiter.js:57:28)
    at /[...]/node_modules/playwright/lib/client/frame.js:150:13
    at Frame._wrapApiCall (/[...]/node_modules/playwright/lib/client/channelOwner.js:77:28)

Process finished with exit code 1

The expected behavior is instead to get the following console log messages:

Load state reached.

Process finished with exit code 0

The above code works fine when run with Chromium in headful mode or when run with Firefox either in headless or headful mode.

[mxschmitt]

Seems like it's because Firefox internal PDF viewer does emit a load event and Firefox tries to display the PDF file. Disabling the internal PDF viewer or adding the download attribute to the link element should lead to a more consistent behavior over all the browsers.

Related #3509 (comment)

[klhex]

Chromium emits a load event as well, but only in headful mode it appears. So would this have to be considered as an upstream bug?

Also, disabling the internal PDF viewer is not a good option for me since I write e2e tests which simulate the average customer's behavior/setup. And the average customer (user) does not disable the browser's internal PDF view.

And changing the application's functionality in order to make test code work ("adding the download attribute to the link element") would be considered as an anti-pattern in my team.

[mxschmitt]

Oh I thought its Firefox related, my mistake. After digging more into it I found out that on headless Chromium the PDF viewer does not get used and it leads to download events instead, this is an upstream issue on Chromium side. To workaround that you could start it headful with xvfb-run when you are using it on Linux, that should work or use the download event too for your use-case. Related issue.

See here for running Playwright with headful with Linux: https://playwright.dev/docs/next/ci#running-headed

[klhex]

Yes, indeed, I am usually running my tests in headful mode. However, since headless is a bit faster and since I am running my tests within a CI/CD environment, I wanted to switch to headless. But I guess I will have to postpone this for a while. ;-)

[mxschmitt]

Someone found another work around: microsoft/playwright-python#675 (comment)

[klhex]

@mxschmitt Thanks for the link to the workaround, that does the job so far. Although, from an e2e (simulating the end-user/customer experience) point of view, being able to use the internal PDF viewer in headless mode as well (like Firefox seems to do it) would better reflect the real-world scenario whereby the end-user/customer would also (in most cases) use the internal PDF viewer.

BTW: I think you mixed headed and headless when adjusting this issue's title. PDF viewer works fine in headed Chromium but does not get used in headless Chromium.