1498257 - Infinite loop loading Flipkart.com because the site's regex fails to match Firefox's JS error message

Status: RESOLVED FIXED

(Web Compatibility :: Site Reports, defect, P1)

Description

[Denis Palmeiro [:denispal]]

Attachment: screencap.png

When loading www.flipkart.com using a geckoview_example build from revision ddcd7cc2f3cd, I see an infinite loop trying to load the web page.  

This is on a Samsung S7 edge running Android 8.0.

Comment 1

[David Bolter [:davidb] (NeedInfo me for attention)]

Over to Kevin for deeper triage (DUPE?)

Comment 2

[Chris Peterson [:cpeterson]]

Is this a regression?

James says he can repro in Fennec (64?).

Comment 3

[Kevin Brosnan [Ex-Mozilla]]

[Tracking Requested - why for this release]: looks to be a 63 regression

Comment 4

[Kevin Brosnan [Ex-Mozilla]]

Flipcart is a very large commerce site in India.

25:17.80 INFO: No more inbound revisions, bisection finished.
25:17.80 INFO: Last good revision: 5d6bf0312e088bb2424a8177589a9fd4aa44bfa8
25:17.80 INFO: First bad revision: f0c6e521429cfaff0585ec6eaf734e9fcf873f8a
25:17.80 INFO: Pushlog:
https://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=5d6bf0312e088bb2424a8177589a9fd4aa44bfa8&tochange=f0c6e521429cfaff0585ec6eaf734e9fcf873f8a

Comment 5

[Tooru Fujisawa [:arai]]

(In reply to Denis Palmeiro [:denispal] from comment #0)
> geckoview_example build from revision ddcd7cc2f3cd

where can I get it?

Comment 6

[Denis Palmeiro [:denispal]]

To build geckoview example app, I used the following build & install commands:

./mach build
./mach package
./mach android archive-geckoview
./mach gradle geckoview_example:installLocalWithGeckoBinariesNoMinApiDebug

The apk is also available on taskcluster:
https://tools.taskcluster.net/index/gecko.v2.mozilla-central.latest.mobile/

Comment 7

[Tooru Fujisawa [:arai]]

in https://img1a.flixcart.com/www/linchpin/batman-returns/app.c7ed1ee40ce.bundle.js
there's the following function:

    function(t, e, n) {
        "use strict";

        function o(t, e) {
            try {
                return e(t)
            } catch (t) {
                if (t instanceof TypeError) {
                    if (i.test(t)) return null;
                    if (r.test(t)) return
                }
                throw t
            }
        }
        var i = /^null | null$|^[^(]* null /,
            r = /^undefined | undefined$|^[^(]* undefined /;
        o.default = o, t.exports = o
    }

it's checking the message, and the rules (i and r) are based on the previous messages which ends with "null" or "undefined", but now the message continues with "; ..."

Comment 8

[Tooru Fujisawa [:arai]]

the actual error thrown there is
  TypeError: t.jsBundles is undefined; can't access its "checkout" property
which doesn't match the above rule

Comment 9

[Tooru Fujisawa [:arai]]

this looks similar
https://github.com/ignivamanishbhudhraja/React-Native-Boilerplate/blob/83bc40ca1e549ae4dd12ec23db1b5fddd2d84a56/src/utilities/Idx.js#L62-L75

Comment 10

[Tooru Fujisawa [:arai]]

located
https://github.com/facebookincubator/idx/blob/master/packages/idx/src/idx.js

Comment 11

[Jason Orendorff [:jorendorff]]

Wow. This JS library actually looks for specific patterns in error messages, using RegExps. Our improved error message does not match what the library expects to see.

This whole idea seems to have unfixable bugs... just for starters:
  https://github.com/facebookincubator/idx/issues/51

Of course that doesn't help us much. This is a pretty big site; the bug happens not to affect desktop Firefox, but I guess only because they're doing UA sniffing (i.e. we got lucky).

What are our options? Treat this as tech evangelism; back out the change altogether; change the error message to whatever exact words V8 uses; change it to something else. The first two are options for Firefox 53.

I'm going to ask Product to make a call on this. It's pressing, obviously.

Comment 12

[Jason Orendorff [:jorendorff]]

Actually, on second thought, let's just back it out. :-P

Stealing the bug.

Comment 13

[Chris Peterson [:cpeterson]]

geckoview62=unaffected because this bug is a regression from bug 1259822 in Firefox 63.

Comment 14

[Kohei Yoshino [:kohei]]

This has already been documented at https://www.fxsitecompat.com/en-CA/docs/2018/improved-javascript-error-message-breaks-code-relying-on-the-legacy-format/

Comment 15

[Jason Orendorff [:jorendorff]]

Fixed by backout on mozilla-release:
https://hg.mozilla.org/releases/mozilla-release/rev/ea4b5c999c575532506dfe6c702f59c5a268983b

I think we already missed RC1, so the fix will be in FF63 RC2.

Comment 16

[Kohei Yoshino [:kohei]]

Marked the compat note reverted.

Comment 17

[Jason Orendorff [:jorendorff]]

Fixed by backout on mozilla-beta:
https://hg.mozilla.org/releases/mozilla-beta/rev/0e99081b5322d213fdba77a12ebbf6293f9c2a7f

We missed beta 1, so the fix will be in FF64 beta 2.

Comment 18

[Jason Orendorff [:jorendorff]]

OK. Now let's hand this over to Tech Evangelism to see if anything can be done.

If not, we give up and graft the backout from beta to Nightly. :-|

Comment 19

[Mike Taylor [:miketaylr]]

Hmm, I guess the first step is filing a bug against the FB lib. Fixing upstream is easy, but it's an uphill battle getting all the sites that might break to upgrade...

Comment 20

[Jason Orendorff [:jorendorff]]

The backout on beta failed. I'll look at it later today.

Comment 21

[Jason Orendorff [:jorendorff]]

I don't feel great about this, but my plan is to fix bug 1493560, then re-land the backout on beta.

This is complicated by the fact that I can't reproduce that bug locally, even in that exact revision -- only in test infra, where it seems to happen every time.

Comment 22

[Henrik Skupin [:whimboo][⌚️UTC+2]]

Jason, how is bug 1488554 involved here, which you set as being blocked by this bug? I assume it was a spelling mistake?

Comment 23

[Karl Dubost💡 :karlcow]

About flipkart, it is a regular of webcompat, multiple reports different issues, one after the other.
https://webcompat.com/issues/17588

Comment 24

[Mike Taylor [:miketaylr]]

Filed https://github.com/facebookincubator/idx/issues/55.

Comment 25

[Julien Cristau [:jcristau]]

Jason this still affects 64, did anything happen for this after comment 21?  Do we need to uplift 1493560 now and re-land the backout?

Comment 26

[Jason Orendorff [:jorendorff]]

Yes, that's exactly what needs to happen here. Sorry for dropping this.

The backout also needs to land in beta (65) and central. We just can't have nice things.

Comment 27

[Julien Cristau [:jcristau]]

Backed out of 64:
https://hg.mozilla.org/releases/mozilla-release/rev/290d09651022f524b7bcc5a5514dc616f93fe89a

Comment 28

[Jason Orendorff [:jorendorff]]

ni?me to make sure the backouts happen in beta and central.

Comment 29

[Ryan VanderMeulen [:RyanVM]]

Looks like the backout doesn't apply cleanly to 65+. Will need rebasing by you, Jason.

Comment 30

[Ryan VanderMeulen [:RyanVM]]

This has been backed out from all branches now.

Comment 31

[Ryan VanderMeulen [:RyanVM]]

Denis, can you confirm that everything works OK for you now?

Comment 32

[Denis Palmeiro [:denispal]]

Not sure how to interpret my experience.  It works using Focus 8.0.3 that has a 64.0.20181214004633 build of Geckoview.   However, it seems intermittent when using Fennec Nightly using a 2018-12-26 build.   Sometimes it loads, and other times it doesn't.  And when it does load, it usually doesn't load properly on a refresh.  I tried clearing storage, cache and reinstalling Nightly but it doesn't change the behaviour.

Comment 33

[Chris Peterson [:cpeterson]]

(In reply to Denis Palmeiro [:denispal] from comment #32)
> Not sure how to interpret my experience.  It works using Focus 8.0.3 that
> has a 64.0.20181214004633 build of Geckoview.   However, it seems
> intermittent when using Fennec Nightly using a 2018-12-26 build.   Sometimes
> it loads, and other times it doesn't.  And when it does load, it usually
> doesn't load properly on a refresh.  I tried clearing storage, cache and
> reinstalling Nightly but it doesn't change the behaviour.

Denis, can you please file a new bug about the Fennec Nightly issue? This bug report has a lot of history about the JS error messages. Tracking the Fennec Nightly issue will be easier in a fresh bug. :)

Comment 34

[Denis Palmeiro [:denispal]]

(In reply to Chris Peterson [:cpeterson] from comment #33)

(In reply to Denis Palmeiro [:denispal] from comment #32)

Not sure how to interpret my experience. It works using Focus 8.0.3 that
has a 64.0.20181214004633 build of Geckoview. However, it seems
intermittent when using Fennec Nightly using a 2018-12-26 build. Sometimes
it loads, and other times it doesn't. And when it does load, it usually
doesn't load properly on a refresh. I tried clearing storage, cache and
reinstalling Nightly but it doesn't change the behaviour.

Denis, can you please file a new bug about the Fennec Nightly issue? This
bug report has a lot of history about the JS error messages. Tracking the
Fennec Nightly issue will be easier in a fresh bug. :)

Filed bug 1518484.

Comment 35

[Firefox Bug Husbandry Bot]

Please specify a root cause for this bug. See :tmaity for more information.