#10790: Too many joins in a comparison for NULL.
-------------------------------------+-------------------------------------
Reporter: mtredinnick | Owner:
Type: Bug | mtredinnick
Component: Database layer | Status: new
(models, ORM) | Version: master
Severity: Normal | Resolution:
Keywords: | Triage Stage: Accepted
Has patch: 1 | Needs documentation: 0
Needs tests: 0 | Patch needs improvement: 1
Easy pickings: 0 | UI/UX: 0
-------------------------------------+-------------------------------------
Comment (by akaariai):
I have now a working version of join trimming beforehand. The approach
taken in setup_joins is this:
1. Walk the lookup path, and turn it into "JoinPath" structures. The
structures basically tell from which model we are coming from, and to
which model we are going to, and using which fields. All different join
types are converted to these simple join structures (m2m, generic relation
joins etc).
2. Trim non-necessary joins from the path. This is easy to do at this
stage.
3. Generate the actual joins. This again is pretty trivial, just call
self.join() for the generated joins.
I think it would be possible to get totally rid of trim_joins(), but it
seems to do some special trimming which the trim in setup_joins() can't do
currently. The trimming logic of setup_joins() could be improved easily.
There were some complications related to generic relations. When fixing
this I ended up removing the whole process_extra hack. Now the generic
relations extra join restriction is appended directly into the join
clause:
{{{
JOIN a on (a.id = b.object_id and b.content_type_id = xxx)
}}}
This should result in correctly working joins in .exclude() and LEFT OUTER
JOIN cases, though I haven't added any tests.
In addition I had to make the m2m through models pickleable. I don't know
if the approach used was correct, but at least it works.
Apart of the above there is one test failure. It is related to combining
ORed queries. In short the combined query now uses LEFT JOIN for one join
where INNER JOIN would be correct, and the reason for this is that pre-
patch one join was added, then trimmed (so it was with refcount of 0 in
the query), but now it is never even added to the query. I think I know
how to fix this one, but haven't done so yet.
In total I think the patch is a major cleanup to the join logic, and it
should allow further improvements if the names_to_path call would be done
already in add_filter (or even add_q) stage.
However, there is very little chance this one can ever end up in 1.4. This
is just too big of a patch. For 1.4 the correct approach will be to teach
the trim_joins to just trim the extra joins.
The new patch can be found from:
https://github.com/akaariai/django/tree/new_10790
--
Ticket URL: <https://code.djangoproject.com/ticket/10790#comment:34>
Django <https://code.djangoproject.com/>
The Web framework for perfectionists with deadlines.
--
You received this message because you are subscribed to the Google Groups
"Django updates" group.
To post to this group, send email to django-updates@googlegroups.com.
To unsubscribe from this group, send email to
django-updates+unsubscr...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.
