Open Graph images (social images)

[Andytizer 269]

So we used to have a social tags extension custom-written which we had to disable due to some performance issues (https://github.com/PCGamingWiki/socialtags). Right now if you share an article, the social preview won't necessarily grab the correct thumbnail (we normally want the game article 'cover' image).

I was browsing around and saw that there's an extension that does OpenGraph images, it is also bundled with MW 1.34 which must mean it is stable: https://www.mediawiki.org/wiki/Extension:PageImages - should we install this?

[Aemony 142]

I don’t see why we shouldn’t try it out. I never did understand why we used a custom extension when there were multiple ones available.

But perhaps it was a different time back in the day.

[Garrett 214]

Going by the documentation, $wgPageImagesLeadSectionOnly should be enabled to ensure only images above the first section (in this case Availability) will be considered. It might also be necessary to change the resolution values to ensure any valid image will qualify (the extension page is a bit vague regarding the non-Wikimedia default values).

A ParserFunction-style way of tagging the exact image to use would still be preferable to trusting any sort of algorithm, but with a bit of tinkering this should work reliably for at least game pages and probably the company and engine pages as well.

7 minutes ago, Aemony said:

I don’t see why we shouldn’t try it out. I never did understand why we used a custom extension when there were multiple ones available.

But perhaps it was a different time back in the day.

The PageImages extension added OpenGraph support in MediaWiki 1.29, so this feature probably didn't exist at that time.

[Aemony 142]

2 hours ago, Garrett said:

A ParserFunction-style way of tagging the exact image to use would still be preferable to trusting any sort of algorithm, but with a bit of tinkering this should work reliably for at least game pages and probably the company and engine pages as well.

Wasn't that what broke the old plugin to begin with? It had a very simple parserfunction-style call that somehow was enough to break stuff that weren't even supposed to run at the same time.