Purge the search tables directly

[aschempp]

Q	A
Fixed issues	Fixes #2162

It does not make sense to delete from the search indexer(s), because we query the tl_search table directly. This can only work with the default indexer anyway.

[leofeyer]

Makes sense to me. 👍🏻 Let's wait for @Toflar's approval though.

[Toflar]

Fine for me, however, it was @ausi who built that when implementing the new search :)

[ausi]

Shouldn’t we instead not query the tl_search table?
How should this work with a non-default indexer?

it was @ausi who built that when implementing the new search :)

I think I mainly switched to the indexer to remove the duplicated SQL code in 6fe7dc9 (#1679)

[leofeyer]

Shouldn’t we instead not query the tl_search table?

How would we find the URL that belongs to the page ID then?

[aschempp]

It doesn't work with other indexes. The (might) need to register their own callbacks.

[ausi]

How would we find the URL that belongs to the page ID then?

We could let the indexers themself decide what to delete?
The page ID can be passed via JSON LD?

$this->searchIndexer->delete(new Document(
	new Uri(''), 
	404, 
	[], 
	'<script type="application/ld+json">' . json_encode([
		'@context' => 'https://contao.org/',
		'@type' => 'Page',
		'pageId' => (int) $pageId,
	]) . '</script>'
));

[leofeyer]

That is a clever idea for sure. 🦊

[Toflar]

I propsed the same idea to @aschempp yesterday but there's just so much more that could possibly be added there. I can add my own JSON-LD data in the FE and use that to index. It would be missing here again.

[ausi]

The page ID is the only information we have in this case. If you added your own JSON-LD data, you can create a callback and call searchIndexer->delete() with your data the same way.

But this way we have the benefit, that the indexer can decide if and how it wants to handle the deletion.

[Toflar]

I tend to agree, yes.

[aschempp]

It might work. I'm just not sure the (massive) overhead is really worth the effort of theoretically have someone implement their own search index.

@Toflar and me also discussed that search results should use the same tags the reverse proxy does. This would allow to correctly purge the search from everywhere (e.g. when changing a news item).

[leofeyer]

Ok, so how do we proceed? Are we going with @ausi's suggestion above or do I merge this PR?

search results should use the same tags the reverse proxy does. This would allow to correctly purge the search from everywhere

I assume that this would be a second step in a separate feature PR.

[aschempp]

🤷 I would opt for my (simple) solution and hope for tags support in a future release …

[ausi]

Ok, so how do we proceed?

I think we should merge this PR for now and create a new issue for the suggestion from #2265 (comment)
If someone finds the time (and/or needs the feature) they can implement it.