[dbl9]

In the past the GTK3 code path was tested only by Red Hat as the primary developer of all things GTK and GNOME. At that time, there was only a GTK2 build from Mozilla, and now instead of doing the same as Windows 32-bit/64-bit we're presented with just GTK3 support with the promise to obsolete the GTK2 code, while seemingly not considering the regressions of GTK3.

Why do we use Mozilla binaries of Firefox on linux distro where there's Firefox builds in the package repository? Many reasons:

1. fast access to security fixes

2. access to EMEfree builds

3. access to different channels

While it is easy to build with cairo-gtk2 as the backend, that code path, as I wrote in a sibling comment, reliably crashes for me anytime I try to do File-Open. I'll try to find out if that AUR recipe or Gentoo ebuild do something different that makes it stable, but the fact remains that GTK2 is about to be unsupported by Mozilla while GTK3 hasn't gotten stable yet, which will make life for anyone that tries to build a GTK2 variant very hard. Therefore, I wouldn't really say it's an easy choice.

[kbenson]

> While it is easy to build with cairo-gtk2 as the backend, that code path, as I wrote in a sibling comment, reliably crashes for me anytime I try to do File-Open.

That points towards GTK2 support possibly not being functional, which could be a good reason why they don't provide a build for it, even if they wanted to.

It sounds like the real problem here is that Mozilla is changing stuff that some people don't want changed. I don't think the solution is to ask Mozilla to produce a GTK2 build, but to ask them to fix anything they've broken. If they've decided to move towards GTK3, then that's their choice, and presumably they have reasons for that choice. I don't think it's out of line for them to expect to support a single working build for an architecture/OS combo. That said, they can and should be notified and pressured to fix any regressions. Additionally, if the change is something people don't want, Mozilla should be notified of that (although I suspect there are technical reasons for the change that most people are ignoring).

[dbl9]

Good points. Just finished building firefox 50.0 with for cairo-gtk2 and trying it out now. Hope there are no crashes.

I agree with what you say and want to add that there two issues here. First is the GTK3 port of Firefox not working as a native Wayland GTK3 window. Second is the themeing issues with GTK3 that people ran into, which is something you can live with. Finally, it's the serious regressions of GTK3 itself which go unnoticed or ignored as evident in the gnome.org bug-tracker and the tickets I've linked to in a sibling comment.

The most important argument for Mozilla providing GTK2 builds is that otherwise the likelihood of the code bitrotting is very high as it happened with the Qt port. The interesting aspect of the Qt port was that at that point in time GTK2 was the best choice, but now for substantiated reasons major applications chose to rather port to Qt than GTK3. Therefore, if the Qt port were to be revived it's much more likely to be of interest and maintained than back when the GTK2 port was good enough that nobody cared about Qt.

I know X11 and Wayland are not Mozilla's main platforms of interest and I'm grateful that they do support with the feature set they do. I'm surprised at the seemingly isolated echo chamber perspective of the GTK3 devs. It's an interesting behavior to observe since without non-GNOME users it could just as well be rolled into GNOME itself. Interesting times, having lived through the times when jwz in 1998 was debating whether GTK1 was any good for Linux.

[kbenson]

> First is the GTK3 port of Firefox not working as a native Wayland GTK3 window.

Is that officially supported yet? I see that back in June/July experimental support was released. Possibly is the current work in an effort to get that working, but it's not ready yet? I just tracked down what looks like the feature tracking bug[1], and it doesn't appear to be ready, but I could be reading it wrong.

> Second is the themeing issues with GTK3 that people ran into, which is something you can live with.

Sure, but from my own problems with themes in Chrome, it's not something you want to live with, especially if you used themes to signify instance information and they stopped working, so I feel people's pain there. :/ Preferably Mozilla would have kept this gated in a branch or experimental build until these issues were worked out. That said, earlier today I did see something about plugins built using GTK2 being loaded into Firefox with GTK3 and the symbols loading wrong, so perhaps it's a much harder problem than it seems, and if it requires theme's to rebuild, then perhaps the quickest and easiest way to make that happen is to force a little breakage. Then again, the distro build probably makes sure this isn't a problem.

> I'm surprised at the seemingly isolated echo chamber perspective of the GTK3 devs.

I imagine that's somewhat to do with where their focus lies. I'm under the impression that a lot of the funding comes from Red Hat, so there are likely complex motives at multiple levels from the enterprise to the funded developers (who might want to justify their paycheck).

In the end, it's one of those things that's hard to accurately critique as an outsider, because there's a lot of specific information that goes into a decision like that. Is a GTK2 to GTK3 migration easier than a GTK2 to QT (or some other toolkit) migration? Probably, but by how much? If this was happening a few years ago, we might be complaining that Firefox uses a toolkit (that is, underneath their own toolkit) instead of X directly. Now we have X and Wayland, so that wouldn't have presented a similar situation anyway. In the end, unless you have some upstream vendor willing to put lots of time and money into making sure you have a performant, backwards compatible API to call (e.g. Microsoft), you'll probably want to make changes at some point in any long-lived project.

1: https://bugzilla.mozilla.org/show_bug.cgi?id=635134

[hsivonen]

The "EME-free" build differs only by the default value for the "Play DRM content" checkbox. You can put a normal build in the equivalent state by unchecking the checkbox.

[dbl9]

Good to know. I thought it doesn't include and prevent the download of some DRM blobs.

[cpeterso]

The "Play DRM content" checkbox controls both the EME API and the automatic download and installation of the DRM blob (the Google Widevine CDM). So you are still correct. :)