ara818 / django-autocompleter Goto Github PK

Currently, there is a pre-save signal that removes an object from the autocompleter, then a post save adds it back. This is to account for term changes on the object that would affect the autocompleter. However, if for some reason the save() on the object fails then the post-save never happens and the object is lost from the autocompleter.

We should update so that there is no pre-save() and all the data needed to remove an object from the AC is in redis and so the remove/add in the AC both happen on post_save.

this term: Macedonia GDP

with these aliases:
'Macedonia Denar' => 'MKD'
'MKD' => 'Macedonia Denar'
'Macedonia' => 'MKD'
'MKD' => 'Macedonia'

Will cause a endless loop.

Easier to maintain / more standard.

It is not used by anyone, not worth maintaining right now.

Objects:
California Unemployment Rate
Cananda Unemployment Rate

Aliases:
California: CA
Canada: CA

Autocompleter returns: "canada unemployment rate" as option when searching for california

why? "calfiornia ur" which gets aliased to "ca ur" which gets aliased into "canada ur"

Hi, I'm the author of a similar project django-completion (I see you have a fork of it in your projects list)! I was wondering if there were any big differences in our implementations?

We do all these extra copies to handle intersection of multi-word search queries, even if we have 1 word. Need to devise a fast path for this common case that ignore unnecessary work

We do the facet intersection last after we do the initial search.

This makes it so a very broad search that returns a ton of results but is then narrowed down significantly via a facet is quite slow.

Instead, we should figure out how to better integrate the facet intersection with the core search so we don't do all this extra work copying around giant search result sets.

remove_all() has 2 steps.

1. remove things using their ID
2. remove anything left over

Part 2 is there really only as a precaution, in case some weird ID changes cause us to lose track of an item.

However, It appears step 1) may not be doing anything due to an encoding issue, and so we effectively try to delete each item twice and succeed the second only with step 2)

We should fix this, but also we should create a way to turn off 2) for testing that step 1) actually works. As it is, even though we test remove_all() the bug the step 1) code is effectively hidden by the catch step 2) deletion.

This project needs a sample project to show how it is used and/or documentation.