On Sun, Jul 30, 2017 at 6:35 PM, Martin Fl=C3=B6ser wr= ote: > Am 2017-07-29 22:18, schrieb Ben Cooksley: >> >> On Sun, Jul 30, 2017 at 3:33 AM, Olaf Schmidt-Wischh=C3=B6fer >> wrote: >>> >>> Ben Cooksley: >>>> >>>> I've checked and it appears that only a small handful of applications >>>> still use newstuff.kde.org: >>>> - KBlocks >>>> - KDiamond >>>> - KGoldRunner >>>> - Kigo >>>> - KSirk >>>> - KSnakeDuel >>>> - KSysguard >>>> >>>> These applications should all be ported to use store.kde.org. >>> >>> >>> How long will it take for users to get the updates? And are we planning >>> to ignore users on, for example, Debian? >> >> >> That is up to distributions. Unfortunately we need to break things >> from time to time on the server side. >> >> As I mentioned, the web server for newstuff.kde.org has been down for >> several months, so this just makes the current arrangements permanent. >> >> If we had to wait until all users were able to receive an update to >> change something like this we would never be able to make any changes >> to our systems. To give an example, the original GetHotNewStuff >> implementation as used by many KDE 3 and early KDE 4 applications >> still receives a large number of requests. > > > But that is a point for not breaking it! If we still have users, we > shouldn't break it! Then I can tell you that in some cases making any change would be absolutely impossible. I have to choose between maintainability of our infrastructure and compatibility, and in this case maintainability wins out. To give some examples... we would never have been able to upgrade to Bugzilla 4 if your above requirement had to remain true. DrKonqi until our conversion to Bugzilla 4 (which introduced the REST API) was reliant on scraping the web pages. The content and layout of those pages were changed by the upgrade. We kept compatibility measures in place within Bugzilla 4 for a good 2 years or so to avoid that breakage. There is a cost to that (modifications to the Bugzilla core code, along with tweaks to the theme - all of which had to be merged on each update including security updates). As another example, moving resources fetched by applications has caused us issues in the past. I can cause at least one application to crash simply by removing one line from an Apache config. In another case we had to put a cronjob in place to copy metadata files from one system (with lots of disk space) to the old location which older versions of an application reference to keep them working. These changes have to be maintained, taken into account when making changes to other systems, and carried over when we perform system migrations and upgrades. Both of those issues were due to developers using QNetworkAccessManager, which by default, doesn't enable redirects. The Qt Installer Framework uses QNetworkAccessManager to make HTTP requests, and crashes whenever it gets hit with a redirect (at least in it's Qt 4 variant). The reason the legacy gethotnewstuff system still works is because it uses static files (which have no maintenance cost) and it uses KIO on the client side - which handles redirects correctly. The maintenance cost to Sysadmin of it is basically nil. What you're essentially asking here is for Sysadmin to takeover maintenance responsibility for a set of scripts which we don't fully understand, which have been hardcoded to use Postgres (a system which isn't really used by us) and which may have other unknown dependencies. That seems a bit unreasonable to me. > > Seriously that is one of the points where we can significantly differ fro= m > proprietary providers where random stuff breaks because services get shut > down. > > Is there no way to redirect it in a way that it can continue to work? Unfortunately no. From what I can tell, newstuff.kde.org speaks the OCS protocol, which in applications is handled by the Attica library (through knewstuff). Unfortunately Attica uses QNetworkAccessManager and redirect handling was only added sometime last year, and only to the Qt 5 variant. KWin 4.11 is forever stuck using a Qt 4 version of Attica, which can't handle redirects. It might even crash if we try to do redirects. We can't even do upgrades to HTTPS (which is enforced on all the servers Sysadmin maintains with few exceptions). > > Cheers > Martin Regards, Ben