Closed Bug 579078 Opened 14 years ago Closed 14 years ago

Decide & declare a minimum Windows version required for Firefox 4

Categories

(Firefox :: General, defect)

Product:
Firefox
Component:
General
Type:
defect
Priority:
Not set
Severity:
normal

Tracking

()

Status:
RESOLVED FIXED
Tracking Flags:
Tracking Status
blocking2.0 --- betaN+

People

(Reporter: beltzner, Unassigned)

Details

(Whiteboard: see comment 31 for conclusion) 

This bug is to track the product decision around minimum requirements for Firefox 4.

Firefox 3.6 was supported on Windows 2000, Windows XP, Windows Server 2003, Windows Vista and Windows 7.

There are a few issues popping up which are predicated on a decision to be made about the minimum version of Windows that will be required for Firefox 4 (ex: bug 563318, bug 577486)

The proposal is to raise the Windows system requirements to:

 Windows XP (SP2 or later)
 Windows Server 2003
 Windows Vista
 Windows 7

Pros: makes things around Breakpad (crash reporting) easier, allows us to use VC++ 2010 compiler which will likely carry some small performance improvements

Cons: limits audience (some evidence that schools still use Windows 2000)
OS: Mac OS X → All
Hardware: x86 → All
OS: All → Mac OS X
Hardware: All → x86
OS: Mac OS X → All
Hardware: x86 → All
Looping in a couple more people that should be in this discussion.

I talked with joduinn about this a bit after the Platform meeting this week and have come up with a solution that I think will satisfy everyone.

For Firefox 4/Gecko 2.0, Win2k should become a Tier 2 platform.  We will not intentionally break it, and will take patches to fix the breakages that do occur.  Releng and QA will continue to provide the same level of automated assurance that they have until this point (which is none whatsoever).

The two immediate problems with Win2k are the breakpad bustage and moving to MSVC 2010.  The breakpad bustage is being addressed in that bug, and the MSVC 2010 switch is not going to happen before we branch for Firefox 4 (it requires, among other things, switching to a completely different method of jemalloc integration).  It's unlikely that anything else will come up that would require us to break Win2k, but we can always revisit this decision if that becomes the case.

Switching Win2k to be officially Tier 2 changes almost nothing in practice (since we don't do automated builds or test suites on it ATM).  We'll continue to put forward the minimum effort needed to ensure that it runs (which basically means fielding bug reports from community members about missing symbols at load time) and other than that people will be on their own, as they more or less are now.
And in case it's not clear, Win2k will be officially obsoleted on trunk when we do switch to MSVC 2010.
Poking around stats, I see a breakdown by os, but not by os version. I wonder if we could pull win2k usage data out of our daily user stats? It would be interesting to see how many of our users actually run each os version.
I strongly support Kyle's plan; keeping the Win2K status quo doesn't hurt anything and changing it for Firefox 4 wouldn't really gain anything (it just adds more complexity with MSVC 2010 between betas), but for any version after that, it's worth it in my opinion.
blocking2.0: --- → ?
(In reply to comment #4)
> Poking around stats, I see a breakdown by os, but not by os version. I wonder
> if we could pull win2k usage data out of our daily user stats? It would be
> interesting to see how many of our users actually run each os version.

Dveditz pulled numbers in #planning a couple days ago.  He said it was about 800K daily users on Win2k.
(In reply to comment #6)
> (In reply to comment #4)
> > Poking around stats, I see a breakdown by os, but not by os version. I wonder
> > if we could pull win2k usage data out of our daily user stats? It would be
> > interesting to see how many of our users actually run each os version.
> 
> Dveditz pulled numbers in #planning a couple days ago.  He said it was about
> 800K daily users on Win2k.

If that was 50K, releasing something that was broken on 2K wouldn't matter. But shunning ~1 million users is bad. While I don't think we should waste any time fixing minor bugs or back porting new features (oopp, for example, is disabled on 2K), issues like bug 577486 should still be considered release blockers.
Someone mentioned yesterday that we had 2,000,000 users on 10.4 when we dropped support for that, fwiw.
Hi everyone. Let's keep this bug to proposed plans and strong data, not heresay. We have abandoned far larger audiences than 1M ADU.

Kyle: so your proposal is to continue to support W2K on Firefox 4, but obsolete it on trunk as soon as we branch so we can take MSVC 2010?
(In reply to comment #9)
> Kyle: so your proposal is to continue to support W2K on Firefox 4, but obsolete
> it on trunk as soon as we branch so we can take MSVC 2010?

Yes, with the caveat that we probably don't want to have releng and others sink effort into running unit tests on Win2k.  Basically, let's change the status quo as little as possible.
(In reply to comment #9)
> Hi everyone. Let's keep this bug to proposed plans and strong data, not
> heresay.

Good point. I just dug into the data myself, and for the record we have ~950,000 users on Windows 2000.
If we are not going to test Win2K we shouldn't continue to support it.  We've been in that precarious position too many times already, and if users actually have a Win2K-specific problem, we are in no position to fix it.

I think we should require one of the 4 in beltzner's righteous comment #0, and SSE2, for Windows.
(In reply to comment #12)
> If we are not going to test Win2K we shouldn't continue to support it.  We've
> been in that precarious position too many times already, and if users actually
> have a Win2K-specific problem, we are in no position to fix it.

Ok, are we going to block installing on Windows 2000 or can we just give a "use at your own risk" warning and let them continue?

> I think we should require one of the 4 in beltzner's righteous comment #0, and
> SSE2, for Windows.

I think there should be a separate numbers-backed discussion for the sse2 requirements.
(In reply to comment #13)
> (In reply to comment #12)
> > I think we should require one of the 4 in beltzner's righteous comment #0, and
> > SSE2, for Windows.
> 
> I think there should be a separate numbers-backed discussion for the sse2
> requirements.

Yes let's handle that elsewhere please.
Adding Patrick Finch who did some more digging on the numbers (notes pasted below)

From patrick_
The OS has been EOL since March 2004, end of service life as of Tuesday this week.  Windows 2000 users really should be planning a migration from the platform if they are not already.

Metrics dashboard is down, which is annoying, but here is what I can get from NetApplications: Windows 2000 share of Fx usage:

July 2008       1.79%
June 2009       0.93%
June 2010       0.52% (0.45% market overall).


I'd say there is a strong case not to support this platform for Fx 4, -a supported application on an unsupported platform is still an unsupported solution. But we do need to understand how long we expect to build 3.6.x for Windows 2000, and be able to communicate that.

I would suggest that if we decide not to proceed with Win 2k support, we prepare a blog post with this data on it.
Comment 0 also mentioned requiring SP2+ for WinXP. Is a decision on Win2k support independent of a min WinXP SP or does keeping Win2k support imply keeping old WinXP support? What are the benefits of dropping WinXP SP1 and older?
(In reply to comment #16)
> Comment 0 also mentioned requiring SP2+ for WinXP. Is a decision on Win2k
> support independent of a min WinXP SP or does keeping Win2k support imply
> keeping old WinXP support? What are the benefits of dropping WinXP SP1 and
> older?
MSVC10 doesn't support WinXP SP1 either.
The issue with K12 schools in the US (and to some extent Canada) is that they
still have a large number of machines running Win2K.  e.g. New York City
schools have about 500,000 machines on their LANs and about 100,000 of those
are running Win2K.  Most of the Win2K machines in K12 still run IE6 (and can't
run later versions of IE).  Moving to Firefox is a big win from a security
standpoint, and I've been lobbying district superintendents and tech directors
to move in that direction.  

Budgets being what they are, schools are unlikely to spend money upgrading
those machines to later versions of Windows as long as the machines still work.
 Available budget will more likely be spent on adding machines. 

That said K12 users are still a small percentage of the total FF user base and
Win2K compatibility can't be allowed to slow the progress of Firefox.  The
worst-case situation for these users is that FF 3.6.x provides a significant
security improvement over IE-6.
(In reply to comment #11)
> (In reply to comment #9)
> > Hi everyone. Let's keep this bug to proposed plans and strong data, not
> > heresay.
> 
> Good point. I just dug into the data myself, and for the record we have
> ~950,000 users on Windows 2000.

Here's some trending data:

NetApplications - global browsing usage on Win2k:

July 2008  2.05% *
June 2009  1.05%
June 2010  0.45% 

*not available for June 2008.
Win2k usage appears to be > halving every year.


Win2k users Firefox, second Tuesday in July

2008 - 1,095,147
2009 -   933,516
2010 -   778,268

Win2k users appear to drop by 150k / year
(In reply to comment #11)
> (In reply to comment #9)
> > Hi everyone. Let's keep this bug to proposed plans and strong data, not
> > heresay.
> 
> Good point. I just dug into the data myself, and for the record we have
> ~950,000 users on Windows 2000.

Here's some trending data:

NetApplications - global browsing usage on Win2k:

July 2008  2.05% *
June 2009  1.05%
June 2010  0.45% 

*not available for June 2008.
Win2k usage appears to be > halving every year.


Win2k users Firefox, second Tuesday in July

2008 - 1,095,147
2009 -   933,516
2010 -   778,268

Worth also noting that the Fx user base doubled in the period July 2008-July 2010.

Win2k users appear to drop by 150k / year, so we might expect to be at around 700k average daily users by the end of 2010, assuming the end-of-service-lifing on Win2k does not accelerate this trend.
I find the question to be whether we need to intentionally break Windows 2000 over the breakpad update.  The only benefit of doing so is the ability to use MSVC10 for Firefox 4.  According to Kyle, to use MSVC10, we need to redo jemalloc integration, and Ted is concerned about doing this between betas.  Are we certainly planning to move to MSVC10 for Firefox 4 and are the benefits of that change sizable enough to merit dropping an OS for Firefox 4?
That's not the only question.  Another question, to me more important, is whether we're comfortable "supporting" Win2k in any way if we're not going to have test coverage, or a meaningful ability to fix problems.  I personally think we should end win2k support with the 3.6.x product, rather than commit, even half-heartedly, to supporting a fully unsupported OS through the full Fx4 lifecycle, which almost certainly runs through the end of 2011, and likely longer.

Also, FWIW, I don't think tier 2 is meaningful within OS revs, unless our installer blocks win2k installs, and someone else is providing those builds.  If you can download and run it from our site, it looks supported, whether we admit it or not.  A lower tier platform like FreeBSD/Solaris/etc are all builds from their package maintainers, and users on those platforms are used to lower levels of quality/support.
Is it fair to say that because there are no test boxes running on Windows 2000 that it is not known to what extent Firefox is broken there beyond what gets reported in Bugzilla? If running a Windows 2000 test box is more effort than it's worth, could somebody do a one-off test run of everything? Then it would at least be known how much needed fixing to pass all of the tests.

I have Windows 2000 somewhere, I can run the tests if someone points me in the right direction of what to run and how to run them.
Kev gave a talk about Fx in Enterprise at the summit. I'll alert him to the existence of this bug, as he had numbers on the size of the possible enterprise market and the portion of said market that runs Win2k.

It suggests that while our Win2k user base is small, it might actually be *growing* in the near-ish term.

That said, I agree with that other Milewski :) that 3.6.x may be sufficient to address that market, and perhaps extending the EOL for 3.6 may be a better use of resources.
Tanner: this bug is not about whether we should intentionally break Windows 2000 - that's being tracked in bug 577486.

Based on the data and the opportunity cost here, I'm leaning towards dropping Windows 2000 from the supported list of OSes for Firefox 4, but keeping it for Firefox 3.6.x of course.

cc'ing kev as per comment 24, but I'm not sure that we'd change this decision if enterprise support demanded that. It hasn't been a huge driver for us in the past.
(In reply to comment #24)
> extending the EOL for 3.6 may be a better use of resources.

How far could it be reasonably extended?
(In reply to comment #22)
> Also, FWIW, I don't think tier 2 is meaningful within OS revs, unless our
> installer blocks win2k installs, and someone else is providing those builds. 
> If you can download and run it from our site, it looks supported, whether we
> admit it or not.  A lower tier platform like FreeBSD/Solaris/etc are all builds
> from their package maintainers, and users on those platforms are used to lower
> levels of quality/support.

You can run a Mozilla built version of Firefox on WINE and nobody expects us to support that, so I don't think the comparison holds.  Making the installer artificially error out on Windows 2000 is just about the worst solution here IMO.
Currently the build doesn't run on win2k. We shouldn't spend any time fixing bug 577486 (especially because of the costs related to upstreaming the change) and should make the installer fail precisely because we don't work on win2k.

This is what we've been saying for a year or more: when win2k breaks, we won't spend time fixing it. We've reached that point, let's stop dragging deadweight around with us. A natural reaction to the build not running on win2k is that the installer should refuse to install on that platform.

FWIW, whichever version of Firefox has a security sandbox around content processes will require WinXP SP3.
I am no daily user on W2K, but I have regularly used W2K in a virtual machine to test localized daily builds. W2K is clearly the least resource intensive OP for virtual machines. 

Therefore I recommend to fix bug 577486 instead of dropping Windows 2000.
This is *not* an advocacy bug. People should feel free to bring *data* to this discussion, but not *opinions*. I'm sure that some people feel like we should support W2K 4evr! and others feel like we should drop it immediately. That's why there's a bug about making a *decision*.

(FWIW, bug 577486 will be fixed, most likely for Firefox 4 Beta 3)
For Firefox 4, the minimum (earliest) supported version of Windows required will be Windows 2000
Status: NEW → RESOLVED
blocking2.0: ? → betaN+
Closed: 14 years ago
Resolution: --- → FIXED
Whiteboard: [pm] → see comment 31 for conclusion
(In reply to comment #31)
> For Firefox 4, the minimum (earliest) supported version of Windows required
> will be Windows 2000

For the record, I just want to point out that this is NOT consistent with previous practice to drop support for OS versions that Microsoft is no longer providing any security patches for.
You need to log in before you can comment on or make changes to this bug.