WMF is upgrading to 14.04 and switching to HHVM.
Although we can't do it yet, we should look at bumping the minimum PHP version we support (considering 5.3 is deaded as far as PHP are concerned), and clean up any hacks/conditionals in core
I guess we should use 5.5.9 as that is what 14.04 brings us... http://packages.ubuntu.com/trusty/php5
We've still got some misc servers and such we're using that are still using zend PHP
We are not at 100%. Just apache stuff. Misc boxes, image scalers, etc all are still pending.
We're not at 100% HHVM yet. imagescalers, job runners, and other servers like terbium and tin. Though I think we'll still be able to do this for 1.25.
Ah, my mistake. Have created a blocking task as nothing was already around that I could find.
Though I think we'll still be able to do this for 1.25.
Cool.
Bumping from 1.25-release to 1.26-release as that looks more realistic.
If someone still would like to push this for 1.25: Feel free to reset.
A couple of points:
Please don't make a WMF-centric decision without considering the many thousands of other MediaWiki installations out there, and how they will be impacted.
I agree that one should consider third party users of MediaWiki. However, I disagree that bumping the required version is equivalent to abandoning users that still use 5.3. MediaWiki makes patch releases and has LTS versions. Furthermore, most of the people still running PHP 5.3 are running out of date MediaWiki versions anyway. Not all of them of course, but the numbers are not as clear cut as "40% of users will get screwed over".
This is a good non-WMF-centric post on the topic: http://blog.ircmaxell.com/2014/12/on-php-version-requirements.html
Hi - yes, I know that post. I agree with some of it.
The problem, as I see it, is that I have never yet encountered a shared-hosting provider where "upgrading PHP" != "migrating servers" and for a lot of people that turns what could be a small job into quite a big scary job. End result? User stays way they are, but stops upgrading MediaWiki.
My point wasn't "40% of users get screwed over" but rather I was asking the question "what percentage of users get screwed over?". The best figure I can find is 40%, but is there a more relevant figure? We should answer that question before dropping 5.3 support. If it turns out that it is about 40%, then we're in a very different place from if it turns out to be 5%.
Wikiapiary has stats about PHP usage on wikis, although it would be useful if filtered by wikis using a supported MediaWiki version (1.25, 1.24 and maybe 1.23). Note that those graphs are not accurate, since they are sampling only the first 500 wikis. Someone which knows how to query SMW data properly may come with a good overview of PHP 5.3 usage
Given that PHP 5.4 will stop receiving security updates in 25 days, and I don't expect Wikimedia to have migrated all 8 of the remaining 5.3 hosts (snapshot100[1-4], tmh100[12], tin and terbium) by then, we should probably be thinking about moving the requirement to PHP 5.5 instead (no longer actively supported, but will be getting security updates for another 10 months).
Change 266932 had a related patch set uploaded (by Jforrester):
Bump required PHP version to 5.5.9
Change 266931 had a related patch set uploaded (by Jforrester):
WIP: Bump required PHP version to 5.5.9
Change 266941 had a related patch set uploaded (by Jforrester):
WIP: Test MW-core against PHP55 for master/REL1_27
Many hosts use php 5.3 and wont upgrade manly the ones that are free that a lot of users like using.
Tough.
They should be upgrading for security reasons, never mind a whole host of other reasons.
Could we do this change in 1.28 instead of 1.27 if it is a LTS release since we changed the requirements after 1.19 release which was a LTS release and it would buy users more time to upgrade there server.
Please, stop flogging the dead horse.
We have an LTS that supports PHP 5.3 - 1.23
The decision has already been made.
Making this LTS support PHP 5.3, would mean we're supporting PHP 5.3 till 2019, effectively.
PHP themselves stopped supporting 5.3 mid 2014, so we were still doing so well over a year after they did
No. Decisions have been made. Deadlines for decisions have been announced widely beforehand (e.g. on wikitech-l).