1603956 - Racing for the same prototype document causes others to receive an untranslated prototype and then cache it

Status: REOPENED

(Core :: XUL, defect, P3)

Description

[Matthieu S.]

Attachment: example of this issue

User Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.15; rv:71.0) Gecko/20100101 Firefox/71.0

Steps to reproduce:

the issue often happens when I click on a link from my email client (Mail or Outlook)

Actual results:

sometimes the menus text is not displayed. (an example of this issue with the screen shot) Note: I have this issue on macOs only

Comment 1

[Clara Guerrero ( Need Info Brindusa Tot Please)]

Hi,

Thanks for submitting this bug to us!

I was unable to reproduce this issue on my end, I tried on MacOS 10.14.5 with Firefox Nightly version 73.0a1 (2019-12-12) (64-bit) nor Firefox Release v71.0

Does this issue occur with a fresh profile as well? You can find the steps here: https://support.mozilla.org/en-US/kb/profile-manager-create-and-remove-firefox-profiles?redirectlocale=en-US&redirectslug=Managing-profiles#w_starting-the-profile-manager

You can also try testing if the issue is reproducible in safe mode, here is a link that can help you:
https://support.mozilla.org/en-US/kb/troubleshoot-firefox-issues-using-safe-mode

Or you can also download Firefox Nightly from here: https://nightly.mozilla.org/ , to see if the issue still occurs there as well.

Thanks in advance.

Best regardas, Clara.

Comment 2

[:Gijs (he/him)]

Zibi, any idea what could be causing this? Perhaps the strings are not yet there when we pass the menu info to cocoa and we never update it, or something? Or something to do with langpacks or missing strings, or something?

Reporter: also, if you open the browser console (not the regular devtools console; use cmd-shift-j as a shortcut to open it), are there any errors there? Also, from an affected Firefox instance, can you open about:support, click "copy raw data to clipboard" and attach as a file (by pasting) at https://bugzilla.mozilla.org/attachment.cgi?bugid=1603956&action=enter ? Thank you!

Comment 3

[Zibi Braniecki [:zbraniecki][:gandalf]]

Nothing obvious comes to mind. If it was persistent, I'd expect langpack misalignment, but if it's intermittent, I'd rather assume race and since macos menus update on change (we can switch languages on fly and macos menu picks up updated DOM), I'd expect that to happen here.

I can try to reproduce once I get macos machine, but if Clara was unable to, it's likely not trivial to reproduce.

Comment 4

[Matthieu S.]

a supplement, my problem occurs when firefox is not started beforehand. I tried with version 73.0a1 (2019-12-19) the problem is the same on the page displaying the link. while the second page page is correct I hope this can help you.

Comment 5

[Matthieu S.]

Attachment: Capture d’écran 2019-12-19 à 22.19.16.png

Comment 6

[Matthieu S.]

Attachment: with issue

Comment 7

[Matthieu S.]

Capture d’écran 2019-12-19 à 22.19.16.png => is my second page (started automatically ?) and "with issue" is my first page

Comment 9

[Francesco Lodolo [:flod]]

Amended subject, since this seems to be the case.

Comment 10

[Zibi Braniecki [:zbraniecki][:gandalf]]

I would love to see the browser console errors, I suspect it's something about I/O for FTL files, but I don't have a macos around so I can't reproduce it :(

Comment 11

[:Gijs (he/him)]

Mathieu / Crudo, can you both please go to about:support, click the "copy raw data" button, and attach the result at https://bugzilla.mozilla.org/attachment.cgi?bugid=1603956&action=enter (using paste in the "File" box) ?

Also, can you press cmd+shift+j after Firefox has opened in this way, and then take a screenshot of the console that pops up and provide it here? Thank you!

Tentatively moving to Core::Intl as this seems fluent-related, but it may be that actual patching needs to happen in cocoa-land. Also, bug 1568518 seems related, but we explicitly disallowed the early blank window option on mac from 70 onwards, so there must be some other issue - still, perhaps that can provide clues as to what's going on here...

Comment 12

[Zibi Braniecki [:zbraniecki][:gandalf]]

I'd also appreciate browser console log, esp if there's anything related to L10nRegistry or Fluent.

Comment 13

[Matthieu S.]

Attachment: RawData about:support

Comment 14

[Matthieu S.]

Attachment: console : first page

Comment 15

[Matthieu S.]

Attachment: console : last page

Comment 16

[Matthieu S.]

Thanks,

Comment 17

[Zibi Braniecki [:zbraniecki][:gandalf]]

Hmm, the errors might be same as bug 1605489, or from some missing menubar.ftl?

Comment 18

[Francesco Lodolo [:flod]]

I tried to reproduce this morning and I can't (from Nightly to Release), the build always starts normally.

Based on the supportURL, I would assume Matthieu is using a full localized build, not a language pack. At that point, the only difference would be the OS (mine is still Mojave 10.14.6).

Comment 19

[crudo.daniele]

Attachment: raw data

Comment 20

[crudo.daniele]

Hi all,
I have another mac with firefox 71.0 with mojave 10.14.6 18G2022 and the top menu is ok, so I think this is os specific (bug can be replicated in Catalina 10.15.2 19C57)

Comment 21

[crudo.daniele]

Attachment: console

Comment 22

[:Gijs (he/him)]

:spohl, I don't suppose you have any ideas about what might have affected this behaviour on catalina?

Comment 23

[crudo.daniele]

Unfortunately not, :(

Comment 24

[crudo.daniele]

If it can help this bug doesn't happen in 10.15.2 when firefox is started in safe mode (i.e. option+double click on the link shortcut --> start in safe mode).
Even if already two windows are opened, both have a consistent top menu.

Comment 25

[crudo.daniele]

I also noticed a difference when firefox is started in safe mode vs normal mode.
In normal mode a first window is opened, then a second one: the second window (foreground) is the window with the link, the first one is the firefox homepage (background).
In safe mode it is the opposite: the second window is the firefox homepage (foreground), the first one is the window with the link (background).

Comment 27

[Eric[:ericswpark]]

Same problem on macOS 10.15.2. Menu bar is completely gone when launching URLs from other applications (like Thunderbird or Microsoft Excel).

Interestingly, opening a new window correctly renders the Menu bar. I attached a screencapture on this bug report (closed as duplicate): https://bugzilla.mozilla.org/show_bug.cgi?id=1605853

Comment 28

[Stephen A Pohl [:spohl]]

Could someone who can reproduce this verify that this isn't a regression? Unfortunately, since this has to do with launching the application, we can't use mozregression in this case. There is an archive with our builds here: https://ftp.mozilla.org/pub/firefox/releases/

Comment 29

[Eric[:ericswpark]]

(In reply to Stephen A Pohl [:spohl] from comment #28)

Could someone who can reproduce this verify that this isn't a regression? Unfortunately, since this has to do with launching the application, we can't use mozregression in this case. There is an archive with our builds here: https://ftp.mozilla.org/pub/firefox/releases/

Tried build 70.0, bug is not present on that build. So it's something that changed between 70.0 and 71.0.

(Side note for testing: if you want to set the default browser to old Firefox, quit all instances of Firefox, launch Safari, set as default browser, then launch the old Firefox and click on set as default.)

Comment 30

[crudo.daniele]

Confirmed, this is a regression, I started with v. 69.0 and there's no bug.

Comment 31

[crudo.daniele]

(In reply to crudo.daniele from comment #30)

Confirmed, this is a regression, I started with v. 69.0 and there's no bug.

Damn, I'm typing too fast..sorry for spamming: this is NOT a regression

Comment 32

[:Gijs (he/him)]

(In reply to crudo.daniele from comment #31)

(In reply to crudo.daniele from comment #30)

Confirmed, this is a regression, I started with v. 69.0 and there's no bug.

Damn, I'm typing too fast..sorry for spamming: this is NOT a regression

A regression means the behaviour used to be correct, and in more recent versions of Firefox it is not correct anymore. It seems like you're saying that 69 didn't have the bug, and 71 does, which would mean it is a regression. Or are you saying you're seeing the same problem in 69?

Comment 33

[crudo.daniele]

Sorry :) this means I don't know what a regression is...
v.69.0 works ok, there is no bug

Comment 34

[Matthieu S.]

for your information, I've the same issue with firefox73

Comment 35

[Matthieu S.]

Attachment: about:support on firefox73

Comment 36

[Matthieu S.]

Attachment: Console log on firefox73

Comment 38

[Liz Henry (:lizzard) (relman/hg->git project)]

Too late for a fix in 71 and likely in 72 though I'll keep this affected in case we come up with a fix that's upliftable for an RC2 or later dot release

Comment 41

[Jim Mathies [:jimm]]

Any ideas here Haik?

Comment 42

[Haik Aftandilian [:haik]]

We could check for Mac Console.app errors being logged and also check if the problem reproduces with the sandbox disabled.

Matthieu or anyone else that can reproduce this issue, could you try the following tests (they are two different things)?

1. Check for errors on the Mac Console.app application (/Applications/Utilities/Console).

   1. Quit Firefox
   2. Open /Application/Utilities/Console
   3. Enter Firefox in the search box (We can also try filtering with plugin-container or with no search term at all, but there are a lot of messages streamed to the Console application.)
   4. Click the Clear button
   5. Reproduce the problem (click the Clear button before each attempt at reproducing)
   6. If the problem reproduces, collect the logged entries in Console. You can do this in the Console program with Edit->Select All, Copy, then Paste into a text file and attach the file to bug.

2. Another test to try is to check if this problem is caused by sandboxing:

   1. Use Nightly
   2. Open about:config in the browser
   3. Set security.sandbox.content.level to 0
   4. Quit Nightly (restart is required for the change to take effect).

Note: disabling the sandbox does affect the security of the browser. The change will be tied to the profile used and the sandbox should be re-enabled after the tests. After testing, remember to set security.sandbox.content.level back to 3.

Comment 44

[Haik Aftandilian [:haik]]

I have a reproducible test case on 10.15.1 (not updated to latest) with Nightly 74 (2020-01-17) where I open a link from Apple Mail.

The problem also reproduces with Firefox Beta 73.0b6 and Firefox 72.0.1.

The problem reproduces with the sandbox disabled. I don't see any messages in the Mac Console.app. Some errors in the Browser Console-- see attached image on comment 45.

Comment 45

[Haik Aftandilian [:haik]]

Attachment: Browser console screenshot collected after problem occurred.

Comment 46

[Francesco Lodolo [:flod]]

Note: this is very easy to replicate, at least for me (had a chance to test a loaner with Mojave Catalina), without setting up any external apps.

With Firefox set as default (I used Nightly), open a terminal and run open https://www.mozilla.org. This will result in two windows being open. One has all the menus, the one with the new link has none.

For what it's worth, this doesn't happen if I open an URL running Firefox from the command line with --new-tab or --new-window and a URL.

Comment 47

[Cristian Fogel [:cfogel]]

Tried to reproduce the issue with Nightly 74 (2020-01-21) on macOS 10.15 & 10.15.2 but to no success.
Attempted with en-us and fr locales.
Not even steps from Francesco's previous comment were of any help.

Comment 48

[Francesco Lodolo [:flod]]

That's surprising. I've just updated the Macbook I use at work, and I can replicate it with Nightly. Note that the browser needs to be completely closed.

I've also realized that it doesn't happen if you have the profile manager showing up (i.e. you uncheck the "Use the selected profile without asking at startup" option in the profile manager).

Comment 49

[:glob ✱]

Attachment: console log

Comment 50

[Cristian Fogel [:cfogel]]

Bingo, the profile manager was it.
10.15 also suffers from this issue.

Comment 51

[Haik Aftandilian [:haik]]

Now that we can easily reproduce this with the US locale, does that rule out internationalization?

Comment 52

[:Gijs (he/him)]

(In reply to Francesco Lodolo [:flod] from comment #46)

Note: this is very easy to replicate, at least for me (had a chance to test a loaner with Mojave), without setting up any external apps.

This bug blocks the catalina metabug, but this seems to imply this reproduces on 10.14 as well for you?

With Firefox set as default (I used Nightly), open a terminal and run open https://www.mozilla.org. This will result in two windows being open. One has all the menus, the one with the new link has none.

Does it also repro with something like open -a /path/to/FirefoxNightly.app/ --args -P profilename https://www.mozilla.org/ (ie when we're not the default browser) ?

Comment 53

[Francesco Lodolo [:flod]]

(In reply to :Gijs (he/him) from comment #52)

(In reply to Francesco Lodolo [:flod] from comment #46)

Note: this is very easy to replicate, at least for me (had a chance to test a loaner with Mojave), without setting up any external apps.

This bug blocks the catalina metabug, but this seems to imply this reproduces on 10.14 as well for you?

No, only Catalina. That's just me being very bad at remembering version names of macOS (corrected to avoid further confusion).

With Firefox set as default (I used Nightly), open a terminal and run open https://www.mozilla.org. This will result in two windows being open. One has all the menus, the one with the new link has none.

Does it also repro with something like open -a /path/to/FirefoxNightly.app/ --args -P profilename https://www.mozilla.org/ (ie when we're not the default browser) ?

/Applications/Firefox\ Nightly.app/Contents/MacOS/firefox -P test https://www.mozilla.org

This works correctly, it opens the URL in a tab. That's consistent with comment 46 (I can't reproduce if I call the Firefox executable, only when the OS opens the browser).

Comment 54

[Francesco Lodolo [:flod]]

Missed the open part of the command.

open -a /Applications/Firefox\ Nightly.app/Contents/MacOS/firefox --args -P test https://www.mozilla.org

This works too correctly: one window open with the URL, and there's a menu.

Comment 56

[Cristian Fogel [:cfogel]]

In an attempt to get the regression range, only manual "bisection" was possible with a fresh profile created for each check.
This goes all the way back to 63.0a1.

• Last good: 2018-07-31-22-02-08
• First bad: 2018-08-01-10-01-16

Comment 57

[:Gijs (he/him)]

(In reply to Cristian Fogel, QA [:cfogel] from comment #56)

In an attempt to get the regression range, only manual "bisection" was possible with a fresh profile created for each check.
This goes all the way back to 63.0a1.

• Last good: 2018-07-31-22-02-08
• First bad: 2018-08-01-10-01-16

Pushlog: https://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=d57a89840dbb4ae0611d0d9a1e6d27e3d0a99e00&tochange=af6a7edf0069549543f2fba6a8ee3ea251b20829

Not sure off-hand what in there could have tripped this... Perhaps the nsTArray refactor makes a difference for some l10n or cocoa API? Stab in the dark though...

Comment 58

[Zibi Braniecki [:zbraniecki][:gandalf]]

It feels unlikely to be bug 1479606 but I see no other explanation. Can someone who can reproduce it try to back it out and try to repro?

Ooh, and if it's this - did anyone reproduce it on debug build? It would trip on MOZ_ASSERT in debug!

Comment 59

[:Gijs (he/him)]

(In reply to Zibi Braniecki [:zbraniecki][:gandalf] from comment #58)

It feels unlikely to be bug 1479606 but I see no other explanation. Can someone who can reproduce it try to back it out and try to repro?

That bug seemed to be content-process only? And the menus are only in the parent process, so I don't think that's possible. Am I missing something?

Comment 60

[Zibi Braniecki [:zbraniecki][:gandalf]]

Am I missing something?

I don't think you are. I'm just scanning the range and can't see anything that would otherwise I could connect to that, so started thinking of some scenario where the parent process registers wrong locales because content process gets spinned up differently and triggers some negotiation that leads to different outcome? I'm really blind here without being able to reproduce :(

I doubt it's that, but I also don't see any better candidate in that range.

Comment 61

[Haik Aftandilian [:haik]]

In case this helps, it might be a clue that some menu items are always displayed while others are not (at least in my case). For example, in the File menu, "Close Tab" is always displayed. In the Bookmarks menu, "Bookmark This Page" is always displayed.

Comment 62

[Haik Aftandilian [:haik]]

Attachment: Menu Screenshot With Some Text Displayed

Comment 64

[:Gijs (he/him)]

(In reply to Haik Aftandilian [:haik] from comment #61)

In case this helps, it might be a clue that some menu items are always displayed while others are not (at least in my case). For example, in the File menu, "Close Tab" is always displayed. In the Bookmarks menu, "Bookmark This Page" is always displayed.

These are the items that have not yet been converted to fluent. I think we know the issue has something to do with fluent, but we don't know why or how.

Comment 65

[Sean Voisen (:svoisen)]

Given the number of duplicates and severity of the problem (for instance bug 1600153 and bug 1580799 seems to be related) can we bump the priority of this one?

Comment 66

[Makoto Kato [:m_kato]]

(In reply to Sean Voisen (:svoisen) from comment #65)

Given the number of duplicates and severity of the problem (for instance bug 1600153 and bug 1580799 seems to be related) can we bump the priority of this one?

OK, we should raise to P1 (we should fix on 74. 73 is too late). I am interesting that first Windows sets menu label, but 2nd windows doesn't use data-l10n-id string even if English. (New Fission Window doesn't use fluent, so it is shown correctly.)

Comment 67

[Makoto Kato [:m_kato]]

(In reply to Makoto Kato [:m_kato] from comment #66)

(In reply to Sean Voisen (:svoisen) from comment #65)

Given the number of duplicates and severity of the problem (for instance bug 1600153 and bug 1580799 seems to be related) can we bump the priority of this one?

OK, we should raise to P1 (we should fix on 74. 73 is too late). I am interesting that first Window sets menu label, but 2nd window doesn't use data-l10n-id string even if English. (New Fission Window doesn't use fluent, so it is shown correctly.)

Comment 68

[Makoto Kato [:m_kato]]

Zibi, as long as I look this, Gecko sets empty text to menu item that FTL resource. When this occurs, DOMLocalization::TranslateElements seems to return promise (mIsSync is false). Could you investigate this? I guess that this promise isn't resolved.

Comment 69

[Zibi Braniecki [:zbraniecki][:gandalf]]

I'll take a look at this tomorrow. From comment 13 and comment 14 it seems that FTL files are not loaded maybe? Some Resource protocol I/O issue during startup in this scenario?

Comment 70

[Zibi Braniecki [:zbraniecki][:gandalf]]

Ok, I did repro using steps from comment 46. It now looks more like some race. Maybe XUL Cache?

Comment 71

[Jens Stutte [:jstutte]]

Zibi, would you want to assign yourself to this bug (it's a P1 regression for 74) ?

Comment 72

[Zibi Braniecki [:zbraniecki][:gandalf]]

Yeah, I'll give it a shot.

Comment 73

[Zibi Braniecki [:zbraniecki][:gandalf]]

I spent today looking at this, and narrowed it down to the XUL Cache.

Basically, if I switch this if to be always false - https://searchfox.org/mozilla-central/rev/7e92a667e3829831c31e8d46aefe7ef67ad5be1c/dom/l10n/DocumentL10n.cpp#125 - and then open the window with

open -a /Users/zbraniecki/projects/mozilla-unified/objdir/dist/Firefox.app/MacOS/firefox https://www.mozilla.org

both windows are fully localized.

Mossop - I'm sorry to bother you but I assume you may be more familiar with MacOS than I am and in particular how the open command works in relation to our cache. Also, since you designed the original cache for Fluent, you may have an easier time reasoning on why would it happen and how to adjust it.

My best guess is that we open two windows "in parallel" and when the second window is localized it already "has proto", but the proto is not localized, and yet because it does have a proto, we skip localization of it.

One workaround would be to check if we're in this "open" mode on Mac and always force-localize the document, but I hope there's a better way to address it.

Comment 74

[Alexandra Martin]

Attachment: Raw data Mac 10.15.2, FF Nightly 75.0a1 (2020-02-13)

Could reproduce the issue on Mac 10.15.2 with Firefox Nightly 75.0a1 (2020-02-13). Will add the according flag.

Comment 75

[Alexandra Martin]

Attachment: Browser Console, Mac 10.15.2, FF Nightly 75.0a1 (2020-02-13).png

Comment 76

[Dave Townsend [:mossop]]

(In reply to Zibi Braniecki [:zbraniecki][:gandalf] from comment #73)

Mossop - I'm sorry to bother you but I assume you may be more familiar with MacOS than I am and in particular how the open command works in relation to our cache. Also, since you designed the original cache for Fluent, you may have an easier time reasoning on why would it happen and how to adjust it.

My best guess is that we open two windows "in parallel" and when the second window is localized it already "has proto", but the proto is not localized, and yet because it does have a proto, we skip localization of it.

The way this is supposed to work is that the first window starts creating the prototype and any other windows for the same document then wait for the first window to finish loading and then use the created prototype. Last I looked we had applied the localization to the prototype before we notify other windows to start loading (https://searchfox.org/mozilla-central/source/dom/xul/nsXULPrototypeDocument.cpp#411) but it from debugging it looks like that has changed now.

Comment 77

[Zibi Braniecki [:zbraniecki][:gandalf]]

Thank you!

Olli - is that accurate? Did something changed within the last year that could cause this race? If so, can you recommend a way for fluent xul cache to avoid loading a window with no translation? (I think it should be as easy as adjusting the if to not use cache if we're not done caching)

Comment 78

[Olli Pettay [:smaug][[e-mail]]]

So which bug regressed this? We have a range, but did anyone go through that range and compile patches possibly one by one to see which patch regressed this? That would be the first thing to do with a regression bug ;)

Last year we went through XUL -> XHTML conversion, so lots of code moved around.

Comment 79

[Zibi Braniecki [:zbraniecki][:gandalf]]

Oh, hmmm.. I initially thought that it was the issue since the beginning, so bug 1517880. But now I think that maybe it used to work and then regressed?

Comment 80

[Zibi Braniecki [:zbraniecki][:gandalf]]

Ah, @cfogel pointed out that we already have a regression window in comment 57: https://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=d57a89840dbb4ae0611d0d9a1e6d27e3d0a99e00&tochange=af6a7edf0069549543f2fba6a8ee3ea251b20829

Olli - does anything in that list look likely to cause such race regression?

Comment 81

[Olli Pettay [:smaug][[e-mail]]]

I did look at that yesterday, and no, nothing was obvious. So someone should go through the patches one by one.

Comment 82

[Zibi Braniecki [:zbraniecki][:gandalf]]

@cfogel confirmed that the QA team cannot further bisect the range because builds from that range are not available on treeherder unfortunately.

I tried to bisect it manually, but it seems like the revision d57a89840dbb (last known good one) cannot be build on mac os with the recent toolchain anymore.

First, there was an error in build/moz.configure/init.configure which I was able to fix by applying the change that happen within this range: https://paste.mozilla.org/O2NwPAYZ#L3

Then, the errors kept sprinkling:

 0:03.24 In file included from /Users/zbraniecki/projects/mozilla-unified/obj-x86_64-apple-darwin19.3.0-opt/modules/libmar/verify/Unified_cpp_libmar_verify0.cpp:2:
 0:03.24 /Users/zbraniecki/projects/mozilla-unified/modules/libmar/verify/MacVerifyCrypto.cpp:19:18: error: typedef 'OpaqueSecKeyRef' cannot be referenced with a struct specifier
 0:03.24   typedef struct OpaqueSecKeyRef* SecKeyRef;
 0:03.24                  ^
 0:03.24 /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX.sdk/System/Library/Frameworks/Security.framework/Headers/SecBase.h:114:25: note: declared here
 0:03.24 typedef struct __SecKey OpaqueSecKeyRef;
 0:03.24                         ^
 0:03.24 1 error generated.

0:10.38 error[E0713]: borrow may still be in use when destructor runs
 0:10.38    --> /Users/zbraniecki/projects/mozilla-unified/third_party/rust/url/src/form_urlencoded.rs:261:40
 0:10.38     |
 0:10.38 259 | impl<'a> Target for ::UrlQuery<'a> {
 0:10.38     |      -- lifetime `'a` defined here
 0:10.38 260 |     fn as_mut_string(&mut self) -> &mut String { &mut self.url.serialization }
 0:10.38 261 |     fn finish(self) -> &'a mut ::Url { self.url }
 0:10.38     |                                        ^^^^^^^^ - here, drop of `self` needs exclusive access to `*self.url`, because the type `UrlQuery<'_>` implements the `Drop` trait
 0:10.38     |                                        |
 0:10.38     |                                        returning this value requires that `*self.url` is borrowed for `'a`
 0:10.38 error: aborting due to previous error
 0:10.38 For more information about this error, try `rustc --explain E0713`.

I don't think I can cherrypick the fixes or invest time to set up a macos system with XCode appropriate for that revisions time to be able to bisect the range and find the revision that caused it :(

Comment 83

[Zibi Braniecki [:zbraniecki][:gandalf]]

I'm unassigning myself to indicate that this bug needs an owner.

I also am moving it to DOM::Core because it is not related to intl/l10n or intl/locale directories, but to dom/xul or even dom/base directories based on comment 76 and comment 78.

Comment 84

[Haik Aftandilian [:haik]]

(In reply to Zibi Braniecki [:zbraniecki][:gandalf] from comment #82)

I don't think I can cherrypick the fixes or invest time to set up a macos system with XCode appropriate for that revisions time to be able to bisect the range and find the revision that caused it :(

It looks like you might just need an older version of rust. You can temporarily downgrade your rust version for a directory/repo using the rustup override command. See rustup help override.

Comment 85

[Olli Pettay [:smaug][[e-mail]]]

Not DOM core. More like XUL, or localization.

Comment 86

[Brian Grinstead [:bgrins]]

(In reply to Zibi Braniecki [:zbraniecki][:gandalf] from comment #73)

I spent today looking at this, and narrowed it down to the XUL Cache.

Basically, if I switch this if to be always false - https://searchfox.org/mozilla-central/rev/7e92a667e3829831c31e8d46aefe7ef67ad5be1c/dom/l10n/DocumentL10n.cpp#125 - and then open the window with

open -a /Users/zbraniecki/projects/mozilla-unified/objdir/dist/Firefox.app/MacOS/firefox https://www.mozilla.org

both windows are fully localized.

I'm catching up on backscroll here so I might be missing something. But is it possible that this regressed at some point in 2018, was fixed, and then regressed again in 2019 (and possibly from Bug 1517880)? A few things make me think that:

1. The regression range from Comment 57 is Last good: 2018-07-31-22-02-08, First bad: 2018-08-01-10-01-16
2. Comment 33 says this was working fine in Firefox 69 (which had a merge date of 2019-05-20)
3. The line of code you referenced in Comment 73 was added in Bug 1517880 (on 2019-07-26)

If so, then chasing down the regressor from the range in Comment 57 won't help us since whatever broke it then was already fixed, and then something different broke it subsequently. We'd want a new regression range starting with Firefox 69 as a known good.

Comment 87

[Zibi Braniecki [:zbraniecki][:gandalf]]

Thank you Brian! Requesting a new regression range starting with Firefox 69 as a known good!

Comment 88

[Zibi Braniecki [:zbraniecki][:gandalf]]

It looks like you might just need an older version of rust. You can temporarily downgrade your rust version for a directory/repo using the rustup override command. See rustup help override.

Yeah, that fixed the rust, but not the C++ errors. I probably would also need to downgrade my XCode SDK. But as Brian pointed out, maybe we can avoid having to go through that?

Comment 89

[Stephen A Pohl [:spohl]]

(In reply to Zibi Braniecki [:zbraniecki][:gandalf] from comment #88)

It looks like you might just need an older version of rust. You can temporarily downgrade your rust version for a directory/repo using the rustup override command. See rustup help override.

Yeah, that fixed the rust, but not the C++ errors. I probably would also need to downgrade my XCode SDK. But as Brian pointed out, maybe we can avoid having to go through that?

I'll point out that the only SDK version that we officially support for building Firefox at the moment is 10.11 [1].

[1] https://wiki.developer.mozilla.org/en-US/docs/Mozilla/Developer_guide/Build_Instructions/Mac_OS_X_Prerequisites#macOSSDK

Comment 90

[Cristian Fogel [:cfogel]]

Rechecked after the previous comments, and the note in comment 33.
Confirmed with 69.0a1 as being a good build.
Last Good: 2019-10-13-21-36-50
First Bad: 2019-10-14-09-52-34

Comment 91

[Stephen A Pohl [:spohl]]

Could you use mozregression, with 2019-10-13-21-36-50 as last good, to narrow this down further? Thank you

Comment 92

[Dave Townsend [:mossop]]

I'm not sure chasing a regression range is necessarily helpful as it may only reveal what caused this particular case to start hitting this race condition. I think it will be easier to just fix the race condition itself.

Comment 93

[Cristian Fogel [:cfogel]]

Manage to find a workaround and get mozregression usable.

Got to the following results:
Pushlog URL

Last good revision: 3bdfb7bc00a06a84c676d4d72ce47ea16b4b0042
First bad revision: 585fe45563356a10a9d53cddfeda2bd699e46dd5

Comment 94

[Dave Townsend [:mossop]]

What's going on here is that the OSX open request is causing us to open a second window (ideally we wouldn't do this but that is a separate bug). The two windows open at basically the same time and so this happens:

1. Window 1 starts opening and attempts to get browser.xhtml from the startup cache.
2. The document isn't cached and so we start to parse the file from disk into a prototype document.
3. Window 2 starts opening and attempts to get browser.xhtml from the startup cache.
4. The cache reports there is a pending prototype so it waits for the parse to complete.
5. Parsing of the prototype completes and both windows are given the built prototype.
6. Both windows construct a DOM document from the prototypes and each sees that their document needs a localization pass and so fires up fluent.
7. Whichever finishes first applies the translation to the prototype.

I'm not entirely sure why the empty menus are happening but it's most likely a result of both windows doing the same translation and attempting to apply it to the same prototype. Fixing the race will solve that and so I'm finishing up a patch that makes window 2 not get the prototype until after window 1 has translated it.

But there is something odd. If browser.xhtml is in the startup cache then none of this should happen. Both windows should just get the translated prototype directly at step 1 and 3 and there should be no race so I was pretty confused for a while about how this was reproducible normally and not just when the cache had been purged.

Well it turns out that the XUL prototype cache has been broken since early October and so right now we're parsing and translating browser.xhtml on every startup!

Fixing bug 1617092 will make this bug much less of an issue since it would only occur in cases where the startup cache is empty (every Firefox upgrade basically).

The range in comment 93 doesn't look right, nothing in there is a likely candidate but it is around a week away from when the startup cache broke so my suspicion is that that is what actually revealed this. It is possible that this bug has existed for longer, maybe even since we started caching fluent passes, but it's just been hidden by the working startup cache.

Comment 95

[Zibi Braniecki [:zbraniecki][:gandalf]]

Oh wow! Thank you Mossop! This seems like a fun investigation and maybe it also will explain why am I seeing tspaint wins in bug 1613705!

Comment 96

[:Gijs (he/him)]

The second window when opening a URL thing is probably effectively bug 1580799 (see particularly comments 4 through 13); there's some kind of race in how we collect URLs we were asked to open by the OS on mac, and the exact effects vary depending on what wins that race.

Comment 97

[Pascal Chevrel:pascalc]

The mitigation in bug 1617092 should be enough to stop tracking this bug, marking 74 as wontfix given where we are in the cycle.

Comment 98

[Francesco Lodolo [:flod]]

This is actually not happening anymore for me (I only get one window, and the menu is localized). Is there a point in keeping this bug open?

Comment 99

[:Gijs (he/him)]

(In reply to Francesco Lodolo [:flod] from comment #98)

This is actually not happening anymore for me (I only get one window, and the menu is localized). Is there a point in keeping this bug open?

Based on Dave's comment #94 (see penultimate paragraph) this could still happen if you stage an update, close Firefox, and then the next time it starts you do so from an external link, ie if the external link case is combined with having to apply updates (thus blowing away the cache).

Comment 100

[Dave Townsend [:mossop]]

(In reply to :Gijs (he/him) from comment #99)

(In reply to Francesco Lodolo [:flod] from comment #98)

This is actually not happening anymore for me (I only get one window, and the menu is localized). Is there a point in keeping this bug open?

Based on Dave's comment #94 (see penultimate paragraph) this could still happen if you stage an update, close Firefox, and then the next time it starts you do so from an external link, ie if the external link case is combined with having to apply updates (thus blowing away the cache).

Yes. If the startup cache is missing or gets purged when you start in this way you'll still hit it. Much less of an issue (gonna argue P3, could even be P5) but there is definitely still a bug in how we handle races to parse the same XUL/XHTML document right now.

Comment 102

[ethan.beyer]

Apologies for posting a duplicate bug. Read the whole thread, looks like it's fixed in 75?

Comment 103

[scott.feier]

Attachment: Context menu with blank entries

Just noticed an interesting secondary effect -- in the primary window opened with this issue, right clicking produces a context menu with blank entries as well.

Comment 104

[Dave Townsend [:mossop]]

(In reply to scott.feier from comment #103)

Just noticed an interesting secondary effect -- in the primary window opened with this issue, right clicking produces a context menu with blank entries as well.

This is to be expected. Basically all static strings will be missing as the UI has not had any localization applied.

Comment 105

[JC]

(In reply to ethan.beyer from comment #102)

Apologies for posting a duplicate bug. Read the whole thread, looks like it's fixed in 75?

It has been fixed as far as I'm concerned.

Comment 106

[JC]

Sorry, I meant it has NOT been fixed

Comment 107

[Francesco Lodolo [:flod]]

(In reply to JC from comment #106)

Sorry, I meant it has NOT been fixed

It is fixed in Nightly and Beta, minus specific (rare) situations described in comment 100. Are you still seeing the same issue?

Comment 108

[JC]

My apologies, it has been fixed in the Beta version.

When testing I did note a minor bug, however. On first access of the "Bookmarks" sub-menu of the Top Menu, the option to "Add bookmark" is missing. Performing any operation after this causes the Add bookmark option to appear.

Thanks.

Comment 109

[Dave Townsend [:mossop]]

(In reply to JC from comment #108)

When testing I did note a minor bug, however. On first access of the "Bookmarks" sub-menu of the Top Menu, the option to "Add bookmark" is missing. Performing any operation after this causes the Add bookmark option to appear.

Please file a new bug for this issue.

Comment 110

[Zibi Braniecki [:zbraniecki][:gandalf]]

That's fixed already in https://bugzilla.mozilla.org/show_bug.cgi?id=1612653

Comment 112

[[e-mail]]

I am still experiencing this issue with v73.0.1 (64-bit) on Catalina.

The original launch page will load the menus correctly and all are populated. However, clicking on a link in another page or from an email will launch a second window and all menu text is missing. Clicking on where the menus are SUPPOSED to be will create their corresponding drop down menus, but there is no text, only the short cuts as shown in the above posting by Haik Aftandilian.

I have disabled all plugins, extensions, etc. but nothing has resolved the problems.

Thanks for following up with this issue.

-DON-

Comment 113

[Francesco Lodolo [:flod]]

(In reply to [e-mail] from comment #112)

I am still experiencing this issue with v73.0.1 (64-bit) on Catalina.

Bug 1617092 should fix most of the cases, and will be in 74 next Tuesday.

Comment 116

[Matthieu S.]

II no longer have worries since the update of OsX (version 10.15.4), you confirm?
take care

Comment 117

[[e-mail]]

(In reply to Matthieu S. from comment #116)

II no longer have worries since the update of OsX (version 10.15.4), you confirm?
take care

I am on FF ver. 74.0 (64 bit) OSx and have just run the OS update. All seems OK for the moment.

Thanks for your help.

Comment 118

[Frank]

(In reply to [e-mail] from comment #117)

(In reply to Matthieu S. from comment #116)

II no longer have worries since the update of OsX (version 10.15.4), you confirm?
take care

I am on FF ver. 74.0 (64 bit) OSx and have just run the OS update. All seems OK for the moment.

Thanks for your help.

Yes, things seem to suddenly work the way they were intended now. How weird. ;)

Comment 119

[Stephen A Pohl [:spohl]]

Thanks for verifying. Closing as resolved, presumably fixed by Apple in the latest dot release.

Comment 120

[Frank]

It's back - in the new release of Firefox 74.0.1 (64-bit): no menus, Forward/Back button hover tooltips are missing. Keyboard shortcuts such as Cmd-D don't work as well. Mozilla really needs to get in better synch with Catalina changes.

Comment 121

[Frank]

The menus are missing both when opening Firefox 74.0.1 from the Mac dock, and when opening windows from e-mail links.

Comment 122

[:Gijs (he/him)]

I'm pretty sure this has never been gone, but it now only happens if you update Firefox as part of the restart. See comment #100. It should be fine after a Firefox restart (for Firefox 74 and above). If you're still seeing it then, please provide more details about your system.

Comment 123

[Frank]

(In reply to :Gijs (he/him) from comment #122)

I'm pretty sure this has never been gone, but it now only happens if you update Firefox as part of the restart. See comment #100. It should be fine after a Firefox restart (for Firefox 74 and above). If you're still seeing it then, please provide more details about your system.

I'm not sure what you mean by a "restart". I updated Firefox last evening, and the computer was shutdown at the end of the day. After booting the Mac today the problem remains.

Comment 124

[:Gijs (he/him)]

(In reply to Frank from comment #123)

(In reply to :Gijs (he/him) from comment #122)

I'm pretty sure this has never been gone, but it now only happens if you update Firefox as part of the restart. See comment #100. It should be fine after a Firefox restart (for Firefox 74 and above). If you're still seeing it then, please provide more details about your system.

I'm not sure what you mean by a "restart". I updated Firefox last evening, and the computer was shutdown at the end of the day. After booting the Mac today the problem remains.

If you create a fresh dummy profile ( https://support.mozilla.org/en-US/kb/profile-manager-create-and-remove-firefox-profiles ), set it as the default, quit Firefox, and then start Firefox with that from the dock, does the dummy profile have the same problem? (you can then set your "normal" profile back to being the default and restart again, then delete the dummy profile)

Comment 125

[Frank]

If you create a fresh dummy profile ( https://support.mozilla.org/en-US/kb/profile-manager-create-and-remove-firefox-profiles ), set it as the default, quit Firefox, and then start Firefox with that from the dock, does the dummy profile have the same problem? (you can then set your "normal" profile back to being the default and restart again, then delete the dummy profile)

I created a new profile, quit Firefox, and then restarted. The window displayed with menus in place and working (context menu too). Forward/Back buttons tooltips displayed normally and keyboard shortcuts behaved normally. I then switched back to the older "default" profile and that appears to work normally now - both cold starting and opening windows from e-mail links.

Thanks for the tip.

Comment 127

[Frank]

The problem has returned.
After cold starting Firefox, and in browser windows opened by clicking on email links - all active windows are missing navigation menus. Right-click context menu text is missing. Keyboard shortcuts not working. Forward/Back button tooltips missing. Changing profile (I have 3 existing profiles) doesn't change the results.

Closing all windows sometimes leaves me with a normal Firefox menu bar, but opening a new window to browse gives me a window with no menus, no keyboard shortcuts, etc. Firefox is pretty much unusable.

Mac OS X Catalina Version 10.15.4
Firefox 75.0
iMac Retina 5K, 27-inch, 2019

Comment 130

[ethan.beyer]

In Firefox 80.0.1, I'm still seeing this issue. Is there a fix intended? I don't see any tracking status beyond firefox76.

Comment 131

[Salvatore]

Same issue, double firefox opened when I click on external link

firefox 94 - macos monteray

Comment 132

[:Gijs (he/him)]

(In reply to Dave Townsend [:mossop] from comment #94)

What's going on here is that the OSX open request is causing us to open a second window (ideally we wouldn't do this but that is a separate bug). The two windows open at basically the same time and so this happens:

1. Window 1 starts opening and attempts to get browser.xhtml from the startup cache.
2. The document isn't cached and so we start to parse the file from disk into a prototype document.
3. Window 2 starts opening and attempts to get browser.xhtml from the startup cache.
4. The cache reports there is a pending prototype so it waits for the parse to complete.
5. Parsing of the prototype completes and both windows are given the built prototype.
6. Both windows construct a DOM document from the prototypes and each sees that their document needs a localization pass and so fires up fluent.
7. Whichever finishes first applies the translation to the prototype.

I'm not entirely sure why the empty menus are happening but it's most likely a result of both windows doing the same translation and attempting to apply it to the same prototype. Fixing the race will solve that and so I'm finishing up a patch that makes window 2 not get the prototype until after window 1 has translated it.

Did you end up writing this patch, or part of this patch?

(In reply to Salvatore from comment #131)

Same issue, double firefox opened when I click on external link

firefox 94 - macos monteray

Every time? That seems like it might be a separate issue, given that per Dave's comment this should now only happen if you start Firefox externally when it needed to upgrade.

Comment 133

[Dave Townsend [:mossop]]

Attachment: WIP: Bug 1603956: Racing loads of the same prototype should wait until any necessary localiation pass has completed.

Comment 134

[Dave Townsend [:mossop]]

(In reply to :Gijs (he/him) from comment #132)

(In reply to Dave Townsend [:mossop] from comment #94)

What's going on here is that the OSX open request is causing us to open a second window (ideally we wouldn't do this but that is a separate bug). The two windows open at basically the same time and so this happens:

1. Window 1 starts opening and attempts to get browser.xhtml from the startup cache.
2. The document isn't cached and so we start to parse the file from disk into a prototype document.
3. Window 2 starts opening and attempts to get browser.xhtml from the startup cache.
4. The cache reports there is a pending prototype so it waits for the parse to complete.
5. Parsing of the prototype completes and both windows are given the built prototype.
6. Both windows construct a DOM document from the prototypes and each sees that their document needs a localization pass and so fires up fluent.
7. Whichever finishes first applies the translation to the prototype.

I'm not entirely sure why the empty menus are happening but it's most likely a result of both windows doing the same translation and attempting to apply it to the same prototype. Fixing the race will solve that and so I'm finishing up a patch that makes window 2 not get the prototype until after window 1 has translated it.

Did you end up writing this patch, or part of this patch?

See attached. I can't really remember where I left off though, I think I was having problems testing it. Also I attempted to rebase this and resolve the conflicts but haven't even attempted to build it in its current state.

Comment 135

[BugBot [:suhaib / :marco/ :calixte]]

The severity field for this bug is relatively low, S3. However, the bug has 14 duplicates and 5 See Also bugs.
:enndeakin, could you consider increasing the bug severity?

For more information, please visit auto_nag documentation.

Comment 136

[BugBot [:suhaib / :marco/ :calixte]]

Clear a needinfo that is pending on an inactive user.

Inactive users most likely will not respond; if the missing information is essential and cannot be collected another way, the bug maybe should be closed as INCOMPLETE.

For more information, please visit auto_nag documentation.