El dissabte, 9 de novembre de 2019, a les 22:03:37 CET, Ben Cooksley va escriure: > On Sun, Nov 10, 2019 at 9:29 AM Alexander Potashev <[email protected]> > wrote: > > > > сб, 9 нояб. 2019 г. в 21:50, Ben Cooksley <[email protected]>: > > > > > Over the past number of years one of the tasks the Sysadmin team has > > > > > worked on has been improving the overall maintainability of our > > > > > systems, with a significant number of specialised cronjobs, exceptions > > > > > and hidden linkages being eliminated. > > > > > > > > > > That is with one great exception: api.kde.org and ebn.kde.org. > > > > > > > > > > Both of these are suffering from an extreme amount of digital bitrot > > > > > and special casing and in general are now in a condition where I > > > > > cannot say for certain whether it would be possible to replicate the > > > > > setup on a new system without us experiencing some degree of breakage > > > > > (some of which we may not discover until weeks/months afterwards). > > > > > > > > > > In addition, the current setup relies on an old-fashioned overnight > > > > > reprocessing of all repositories, which is inefficient and resource > > > > > expensive. A more modern approach would have the various projects api > > > > > documentation generated on a delayed cycle from relevant branches as > > > > > part of something like a CI job (but not part of the actual CI > > > > > workflow itself). > > > > Hi Ben, > > > > I can't discuss this topic before we understand what exactly is wrong > > with api.kde.org and ebn.kde.org and why they are hard to manage. > > Could you please describe the current situation (where to find source > > code, how many servers, etc) in new Phabricator tasks like you did for > > identity.kde.org in https://phabricator.kde.org/T8449 ? > > > > P.S. Complicated legacy systems can be easy to replicate once you > > automate their deployment by using Docker containers and/or > > configuration management software like Ansible. > > The problem isn't just the complicated legacy nature of them, but also > how fragile and impossible to maintain they are. > > There are currently to my knowledge the following ways of generating > documentation within the system as it currently stands: > - Legacy KDE 4 era generation tooling, still in use for large parts of > KF5 based applications (which needs periodic fixes, see > https://cgit.kde.org/kdelibs.git/commit/?id=12a8bb503351b98869f722ba932f822e3c495883) > - DoxyQML, which is in part reliant on the above KDE 4 era generation tooling > - Specialist handling for 'cmakedocs' and 'mkdocs' based projects > - New era KApiDox tooling > > On top of all of this, the entire process can only be run as one > single monolithic piece, which makes debugging and investigating > faults difficult. > > To use an example of this, back in February 2018 we received a request > (Sysadmin ticket) from someone reporting that their project's API > documentation wasn't being generated. > Despite investigation by Allen and others, we were unable to resolve > the issue, and recommended that the project in question be switched > from the KDE 4 era tooling to the newer KApiDox tooling, as it wasn't > possible to identify the fault. > > Earlier this year there was a fault with API generation in general > which broke a number of projects due to encoding of filenames/folders > (fixed in > https://cgit.kde.org/websites/quality-kde-org.git/commit/?id=f71c68bc80cce68aa3bcc6a51626c8f22b7c73c3).
I volunteer to fix it. Cheers, Albert > > > > > -- > > Alexander Potashev > > Cheers, > Ben >
