httpbrowser-compat.turnstile.workers.dev/ makes Pale Moon crash

[Goodydino]

Why does https://browser-compat.turnstile.workers.dev/ make Pale Moon crash? It does not crash 5 other browsers. Two of those were derived from modern Firefox, one is a Blink browser, one is a WebKit browser, and one is SeaMonkey.

[billmcct]

It also crashes my PM on both Win 10 and Win 7.
I discovered that about 2 weeks ago.

[Night Wing]

After clicking on the link, it took me to the page where I had to wait about 45 seconds to get the TurnStile to complete. After completion I took a screenshot and placed it in my Pictures folder and as soon as I had finished, Pale Moon crashed. I felt fortunate to get the screenshot.

[Moonchild]

One of cloudflare's challenges can make Pale Moon crash.
The turnstile troubleshooter does not exclude this particular challenge.

I would investigate but here's the big deal:
CloudFlare refuses to disclose any test scripts, challenge code or even the most basic details to me.
They are also not telling me what makes the turnstile test suite hang, other than some vague "it's got something to do with CSP" without giving me any details (CSP is a pretty big spec, so it's impossible to start making wild guesses or find out "by trial and error"; I'd probably be here for many months with no guarantee I hit the cause of their sore spot in their scripting) which makes it impossible for me to do anything about because as far as I know our CSP implementation is spec compliant.
They know which test crashes Pale Moon; they are not giving me the details to reproduce or solve it (or any other cloudflare problem, for that matter!) because they are apparently afraid it "may go public". The turnstile for me just hangs and fails to complete, as Night Wing indicated as well. So I can't even reproduce the crash at this point because their test suite is broken in all sorts of ways.

[Kris_88]

https://browser-compat.turnstile.workers.dev/

Palemoon 33.8.2, 32bit.

First attempt:

Checking browser storage (6/10)

Problem signature:
Problem Event Name: APPCRASH
Application Name: palemoon.exe
Application Version: 6.7.0.9363
Application Timestamp: 68a5ee0b
Fault Module Name: xul.dll
Fault Module Version: 6.7.0.9363
Fault Module Timestamp: 68a5eec5
Exception Code: c0000005
Exception Offset: 0131e082
OS Version: 6.3.9600.2.0.0.256.48

Second attempt:
Test OK.

Third attempt:
Test OK, APPCRASH

Problem signature:
Problem Event Name: APPCRASH
Application Name: palemoon.exe
Application Version: 6.7.0.9363
Application Timestamp: 68a5ee0b
Fault Module Name: xul.dll
Fault Module Version: 6.7.0.9363
Fault Module Timestamp: 68a5eec5
Exception Code: c0000005
Exception Offset: 0131e082
OS Version: 6.3.9600.2.0.0.256.48

[Kris_88]

Palemoon 34.2.2 32bit.

All check passed, waiting turnstile to complete.

Problem signature:
Problem Event Name: APPCRASH
Application Name: palemoon.exe
Application Version: 6.8.1.9615
Application Timestamp: 69f22be1
Fault Module Name: xul.dll
Fault Module Version: 6.8.1.9615
Fault Module Timestamp: 69f22c90
Exception Code: c0000005
Exception Offset: 00cdda64
OS Version: 6.3.9600.2.0.0.256.48

[Moonchild]

Palemoon 33.8.2, 32bit.

Please update. We did solve a CF-triggered crash in the meantime that is known.

[Kris_88]

Please update.

The next post is for the latest version.

[Moonchild]

Yes and you didn't crash there

[athenian200]

So, I tested a few versions of Pale Moon, Windows 10, 64-bit edition...

34.2.2: Crashes immediately, seemingly while checking Canvas.
34.2.1: Same result or similar.
34.2.0: Check fails, but no browser crash. Lots of lag and temporary "not responding" stuff, though.
34.0.0: Check fails, for some reason with less lag.

Don't know if this helps narrow it down at all, but those are the results I'm getting. It seemed like something changed in the changes done for the point releases that causes more crashing, which is kind of odd because there's nothing obvious in those that would seem to cause a big problem like that.

[Kris_88]

Yes and you didn't crash there

Good.

Overall, the problem is caught quite reliably in the 32-bit version. At first, everything runs fine, the tests pass, then the script times out, then waits and crashes. I tried setting dom.max_script_run_time to 15 seconds and 60 seconds, but the result is the same—it times out.

[Gemmaugr]

Pale Moon 34.2.2 64-bit Win 10 crashes:

Code: Select all

Faulting application name: palemoon.exe, version: 6.8.1.9615, time stamp: 0x69f22ae9
Faulting module name: xul.dll, version: 6.8.1.9615, time stamp: 0x69f22c42
Exception code: 0xc0000005
Fault offset: 0x0000000001066dfe
Faulting process ID: 0x1048
Faulting application start time: 0x01dce15dddb69df6
Faulting application path: C:\Program Files\Pale Moon 64-bit\palemoon.exe
Faulting module path: C:\Program Files\Pale Moon 64-bit\xul.dll

Code: Select all

Faulting application name: palemoon.exe, version: 6.8.1.9615, time stamp: 0x69f22ae9
Faulting module name: xul.dll, version: 6.8.1.9615, time stamp: 0x69f22c42
Exception code: 0xc0000005
Fault offset: 0x0000000001066dfe
Faulting process ID: 0x233c
Faulting application start time: 0x01dce16c96475114
Faulting application path: C:\Program Files\Pale Moon 64-bit\palemoon.exe
Faulting module path: C:\Program Files\Pale Moon 64-bit\xul.dll

I've never seen these checks other people show, not even when it didn't used to crash on me. Now it's just an immediate CTD.

[Moonchild]

I'll see if I can figure out if any of these crash addresses make sense, but once again without CF working with me and only having their pretty random full test suite that has yet to crash at all for me to work with, it's going to be difficult.

[Moonchild]

Oh, seems something did change with their turnstile troubleshooter and it crashes trying to report back a failure by doing something weird with the requestsubmit API (javascript's new form submission API that was implemented very recently in Pale Moon). Funny thing is it crashes in a corner case situation I specifically tried to work around a crash with, but now the workaround crashes due to what they are apparently doing XD

[Moonchild]

34.2.2: Crashes immediately, seemingly while checking Canvas.

I did by chance get an immediate crash on our master branch this morning, and the immediate one seems to be related to doing something weird with simulating range selection with "thumb dragging" (a mobile concept) in a way that would never occur in the real world... so they are doing something really really niche there; but a nullcheck should fix that.

[back2themoon]

I thought that Turnstile Troubleshooter page is not meant for Pale Moon at the moment, i.e. we are still on some sort of "exception/patch/special case" mode i.e. it is not a reliable test?

One of cloudflare's challenges can make Pale Moon crash. The turnstile troubleshooter does not exclude this particular challenge.

This is the interesting part. So, do we keep hammering the Troubleshooter page or hunt for that specific challenge, wherever it may be? I haven't yet spotted such a website/CF test that crashes the browser.

Also, Pale Moon is not on their "supported browsers" list yet. I guess that's going to take a while...

[Moonchild]

I thought that Turnstile Troubleshooter page is not meant for Pale Moon at the moment

It is, but it is not the test suite that is used "in the wild" for us at the moment. But it is what they would prefer to use globally without specific carveouts for individual browsers. Unfortunately it's not helping our development if we are left to guess what is going wrong with their "unmitigated" test. After all, their test page passes all API tests on Pale Moon, but there's some other weird thing going on with their test suite in code surrounding it (i.e. not the individual tests, except one of the suite that caused the crash). Once the crash is solved they can use that test as part of their suite on production, which is (apparently?) such an important thing to test for that bots "can't easily work around" or what not; although I'm still pretty convinced bots these days are just Chrome instances that are being puppeteered through WebDriver or what not (which we don't actually support for this very reason among others)

[back2themoon]

So, the Turnstile's inner workings are kept like a trade secret. Like the Coca-Cola formula.

[Moonchild]

So, the Turnstile's inner workings are kept like a trade secret. Like the Coca-Cola formula.

Exactly. And it's been the only tool CF has made available in any way shape or form.

And worse yet, I was just now told that CF "provide the troubleshooter as a quality-of-life tool and maintain it with best effort." So they aren't even committed to making it work while that was one of the very things that kept coming up in our meeting as it was the only tool remotely available to be able to test Pale Moon (other than testing internally which they seem unwilling to consider, as well) prior to them pushing things to production, so they are already going back on their promises.

I'm really running out of what little patience I have left with CF. They have effectively not addressed our problems, not provided answers, not told me what behaviour is wrong or broken in Pale Moon (according to them), not provided any tools, and are now playing down the one thing we could use in some fashion as "best effort without commitments", while continuing to be gatekeepers for the Internet and access to large swathes of it. EU's Digital Markets Act is pretty clear about how that is not acceptable behaviour -- even if it was written initially to deal with "preferred bundled software" for operating systems, it does lay the groundwork for addressing unfair practices by other types of gatekeepers like CF.

[Kand_in_Sky]

So, do we keep hammering the Troubleshooter page

Does not work for a very long time. When try so send:
"An unknow error occurs. Please try again later"

At least for me, PM 34.2.2 (64-bit, Win7)