I'm pretty sure I saw some related work less that 3.5 years ago, so it can have been moved somewhere else. Not sure about this, it could be me mixing it up with something else.
Good arguments are always appreciated.
T107119#4684407: The naming schema used for Lua is pretty weird and should be revised. As it is now it creates way to long names without any real need. Such names are only necessary in a system where code can be loaded at will from external sites. Within a controlled system it is not necessary.
Not quite sure if it is related, but the path used in the templatecall seems weird. It does not fail on the path though.
This lib still has the problem that it only makes sense when both Lua and Templatedata exist in the environment. It is a fundamental flaw for several extensions, they add some functionality together with some more or less hack-ish code to figure out whether they shall run in the current environment.
Does this still happen?
This is closed for a wrong reason, and blocks proper localization by using json files. It also blocks use of json-based configuration files, forcing such files into other non-related namespaces.
Sat, Jan 12
Fri, Jan 11
The original bug is not about pinning the php-luasandbox, that is a later chang by @Reedy
Thu, Jan 10
I have double checked with a clean install. I can also test on alternate machines if necessary.
Vagrant works for all my instances, also instances with Mediawiki, except for those cases where I try to include php-luasandbox. Simplest explanation (Occams razor) ; something is wrong with php-luasandbox.
Double checked with a clean install of Ubuntu 18.04 LTS, with Vagrant 2.2.3 64bit Debian (https://www.vagrantup.com/downloads.html) and Virtualbox 5.2.22 Ubuntu 18.04 (https://www.virtualbox.org/wiki/Download_Old_Builds_5_2).
A virtual environment is isolated from the surroundings. If an error emerge inside the environment it is most likely due to whatever is running inside that environment. I have tried debian/contrib-stretch64, debian/stretch64, and generic/debian9. I have also tried vagrant 2.0.2 and 2.2.2, and VirtualBox 5.2.18 (the sluggish one), 5.2.22, and 6.0.
Mon, Jan 7
How could it be anything special with this machine? A newly installed Ubu 18.04? It is booringly standard.
Ouch, made the change no 7 from https://gerrit.wikimedia.org/r/482394 manually, and got a segfault.
Got the luasandbox not present.
Should try again with Reedys changes.
Strange thing; I'm pretty sure I have already tested debian/stretch64…
Got an idea reading T212950#4857183 and changed the box to debian/stretch64 and built a new instance. Only about 3-4 000 tests has run, but it seems to work… The tests for Lua is towards the end of the test run, so it will be some time before I can say it do work for the full test.
Sun, Jan 6
Perhaps $wgScribuntoDefaultEngine should be set by an argument. It is not given that the luasandbox is always wanted or needed. Default value could be luasandbox to make minimal diruption. Other entries could be adjusted accordingly.
Changed the role to use the standalone to check if that would be usable
Commented out codeeditor and syntaxhighlight, but did not change anything.
Note T212950#4856571, I have done a vagrant box update a week ago.
Sat, Jan 5
Seems like the segfault happen at nearly random places.
If I run the complete tests before enabling Scribunto they pass. If I enable Scribunto then they segfaults on seemingly random places, and so early I doubt it reaches the extensions. If phpunit reshuffles the tests it could come from extensions.
Smoke testing indicates that the solution works, but running the full test suite as
Tested the solution by copy-pasting the changes to scribunto.pp and it seems to work as expected. Have not checked the complete set of tests yet.
I will try this, but it seems like I have to make a new instance, that takes time on my sluggish machine. ;)
Ok. Not sure what more I can provide in this case?
Seems like T212950#4856570 gives the same as here. (Gee, its a Vagrant instance!)
In case it matter
Did run vagrant provision twice and the previous weird behavior is gone.
I do not fully recall but I believe I might have set up a Vagrant instance around 20.–22. December and it worked, then later on I got errors and used an older Vagrant instance instead. Then I tried to check if everything worked on 4. January, and I found that the extension failed.
I have not tried to symlink the whole directory, only copied over the file. That does not solve the issue.
Only strange thing I've found is `/etc/php/$version/mods-available' only having an entry in 7.0, while the wiki reports 7.2.
Fri, Jan 4
Thu, Jan 3
Tue, Jan 1
Sun, Dec 30
Due to T209089#4791149 and the additional failures I refactored the code, but the new code is not committed.
There are community consensus for changing the sidebar, a very weak consensus for changing the entry in the site matrix, and no consensus for any changes in the prefix. The later discussion is very confusing with people arueing against their own voting. My proposal was to delay any changes because I doubt it was possible to get consensus for anything else. It triggered fierce resistance, which was opposite of the recommendation, but it seems like the users read it as an opposite recommendation.
Please mark hooks as deprecated before such changes, and update the documentation accordingly.
This (in combo with a change in PHPunit) trigger some very strange bugs.
Tue, Dec 25
Dec 19 2018
- ontology.owl (should probably not be changed)
Dec 18 2018
I really dislike the whole idea of removing functionality the editors expect to work. It is a very "technical" approach, "fix it" and leave the actual users with more markup. The editors, the actual users, want to write their stuff without learning a whole lot of markup.
I'll await what happen with Change 480480 before I upload any more.
Dec 15 2018
Dec 13 2018
Dec 12 2018
Link is visible at nowiki!
Dec 11 2018
The behavior comes from JsonConfig\JCDefaultObjContentView::getDefault() which is, quite frankly, complete crackpot! Someone has created a default behavior that is a BUG. If you want to prefill a page, use preload as a parameter to index.php, use the placeholder attribute, or put the example in a message placed outside the edit frame.
Dec 10 2018
Verified coexistence with Wikibase, which is ok.
Have not verified whether there were changes in reported errors from tests.
Dec 9 2018
- Feel free to figure out how to reproduce, I can't even get this behavior to go away.
- An empty edit window, there are no previous edit done.
- There is a screendump above.
Dec 8 2018
Dec 1 2018
Unit tests gives
OK (24 tests, 84 assertions)
Nov 30 2018
Another pretty kewl thing to do is to calculate which articles are trending in the morning. Because the whole pageview-mix is pretty noisy, you must first try to create a model for how the mix of articles changes through the day and week, and then try to figure out whether the observed change in the morning is for real or just ordinary noise.
There are several cases where a daily article resolution for pageviews could make sense, or even hourly or per minute. This is not so much for the usual article, but for special marker articles. Such statistics only (?) makes sense if it can be given with a geolocation.