#27936 closed Cleanup/optimization (fixed)
Add some clarifications to "Spanning multi-valued relationships"
Reported by: | Thomas Güttler | Owned by: | Jacob Walls |
---|---|---|---|
Component: | Documentation | Version: | dev |
Severity: | Normal | Keywords: | |
Cc: | tzanke@…, Simon Charette, Zach Borboa | Triage Stage: | Ready for checkin |
Has patch: | yes | Needs documentation: | no |
Needs tests: | no | Patch needs improvement: | no |
Easy pickings: | no | UI/UX: | no |
Description
First of all: Thank you for the great docs.
Since it took some time until we got the difference:
filter(entry__headline__contains='Lennon').filter(entry__pub_date__year=2008) vs filter(entry__headline__contains='Lennon', entry__pub_date__year=2008)
The docs are great:
https://docs.djangoproject.com/en/dev/topics/db/queries/#spanning-multi-valued-relationships
But maybe the ascii art below helps to understand it better?
What do you think?
+--------------------+ | Lennon | - | Entry 1 | +---------------+ / | 2008 | | | / +--------------------+ | Blog 1 |/ filter(entry__headline__contains='Lennon', entry__pub_date__year=2008) | |\ - +---------------+ \ - \ +--------------------+ \ | | -| Entry 2 | | | +--------------------+ +--------------------+ | Lennon | - | Entry 3 | +---------------+ / | | | | / +--------------------+ | Blog 2 |/ filter(entry__headline__contains='Lennon').filter(entry__pub_date__year=2008) | |\ +---------------+ \ - \ +--------------------+ \ | 2008 | -| Entry 4 | | | +--------------------+
Ascii Art: https://textik.com/#100375b764993664
The ascii art could get improved, I wanted to ask you first before polishing it.
Change History (23)
follow-ups: 4 19 comment:1 by , 8 years ago
comment:2 by , 8 years ago
Cc: | added |
---|
comment:3 by , 8 years ago
Summary: | ASCII Art for docs "Spanning multi-valued relationships" → Add some clarifications to "Spanning multi-valued relationships" |
---|---|
Triage Stage: | Unreviewed → Accepted |
Type: | Uncategorized → Cleanup/optimization |
comment:4 by , 8 years ago
Replying to Josh Smeaton:
I think there's definitely scope to improve the docs around multi valued relationships, but I don't think ASCII art (or that diagram) is really the right way of doing it. For some added confusion, the docs fail to mention that duplicates are possible with the second query (with multiple filters) if there is an entry with the headline Lennon AND it was posted in 2008. It seems the docs go through a lot of effort to avoid mentioning that a second filter causes the query to create a second join to the same table.
Hi Josh,
I think your proposal to change the docs are valid.
This issue is about the ascii art.
Why not open a new issue for your proposal?
comment:5 by , 8 years ago
I don't think there would be consensus to use ASCII art in the Django documentation. If you think some diagram might be helpful (even though Josh said he didn't think the diagram is the right way to clarify the situation), please follow the pattern used by existing images. For simplicity, I retitled this ticket rather than closing it and creating a new one.
comment:6 by , 8 years ago
Yes, you are right. the ASCII art is not a perfect solution.
If I would provide a SVG diagram instead of the ascii art, would you include it into the docs?
comment:7 by , 8 years ago
I agree with Josh that a diagram probably isn't the best way to clarify things. The " To select all blogs" sentences seem clear to me but perhaps you can help state them more clearly if you found them confusing.
comment:8 by , 8 years ago
Experts like you are, don't need a diagram.
Last weekend I taught 32 people the joy of python programming who had few or no experience with this language. I do this yearly since about 13 years.
Trust me, this helps to see the IT world from a different perspective.
My goal is to make software development newbee friendly.
I think a diagram like this would help.
comment:9 by , 8 years ago
Diagrams are used sparsely because they are more difficult to maintain. I don't think the diagram offers advantages compared to an example shell session that creates objects, runs queries, and shows the results.
comment:10 by , 8 years ago
Yes, now I see. The real problem is that diagrams are hard to maintain. If maintaining them would be easier, then there would be more.
There are several extensions for sphinx which could be used (I needed to remove a link to the sphinx plugin overview page, otherwise trac thinks my post is spam)
But this gets off topic for this particular issue.
Tim, what do you think?
comment:11 by , 8 years ago
I'm not convinced that a diagram is advantageous for this but if you want to create one, you can get other opinions on the DevelopersMailingList.
comment:12 by , 8 years ago
Yes, know I think I understood you. The real problem is that diagrams are hard to maintain.
If maintaining them would be easier, then there would be more diagrams. The problem is the media break between human editable ascii (which we all love) and some svg which looks like some binary randomly encoded to xml.
There are several extensions for sphinx which could be used: http://www.sphinx-doc.org/en/stable/develop.html
But this gets off topic for this particular issue.
Tim, what do you think?
comment:13 by , 7 years ago
Cc: | added |
---|---|
Version: | 1.10 → master |
Given how often the multi-valued filter()
chaining behavior is reported as a bug I think this might be worth a shot.
I'm not a big fan of diagrams either and I think a simplified shell session would be a good step forward. I guess mixing both is also an option.
If we really want to go with graphs I'd suggest we use the the Graphiz extension which makes it easy to maintain, generate SVGs, and should be flexible enough to express the previously mentioned ASCII graph.
comment:14 by , 7 years ago
I personnaly prefer the ascii art (https://textik.com/#100375b764993664) but with graphiz you can do much more. But on the other hand the
ascii art is straight forward. It is WYSIWYG :-)
But I think diagramms help. I don't care if it gets created with tool x or tool y.
comment:15 by , 6 years ago
My objection was not to any diagram, but to that one specifically. I think the right diagram definitely would help, but it should clearly show items that match and items that don't match for each query type, perhaps with colors. Bonus points if it's able to call out the duplicates problem, but that could be documented separately.
I don't personally mind which technology choice is used.
comment:16 by , 6 years ago
Coming here from django-developers post
I side with Josh that the proposed diagram probably isn't the best, and if we were to include one, another could explain the problem better. I find the existing text fairly clear - I've forgotten and re-learned this distinction from the docs a few times.
On technology, graphviz isn't the most flexible as it can only do graphs. Other diagrams we might add to the docs in future might need more flexibility, so I think SVG is a better choice. There's nothing stopping the first version of an SVG diagram being generated with graphviz.
comment:17 by , 6 years ago
WRT the format, +1 for SVG, but *clean* SVG if possible for better maintainership. The SVG syntax of the OmniGraffle-generated ones is awful.
comment:18 by , 6 years ago
Cc: | added |
---|
comment:19 by , 3 years ago
Owner: | changed from | to
---|---|
Status: | new → assigned |
I linked to this doc in the release note I wrote for #16063. I think the main flaw of the doc is it's a bit wordy. It starts with examples, but sort of hypothetically, then discusses the implementation, then does concrete examples. Could be streamlined. Use the saved space to discuss duplicates and joins (that's where I could see the shell session coming in). I'll try to give it a go.
comment:21 by , 3 years ago
Triage Stage: | Accepted → Ready for checkin |
---|
I think there's definitely scope to improve the docs around multi valued relationships, but I don't think ASCII art (or that diagram) is really the right way of doing it. For some added confusion, the docs fail to mention that duplicates are possible with the second query (with multiple filters) if there is an entry with the headline Lennon AND it was posted in 2008. It seems the docs go through a lot of effort to avoid mentioning that a second filter causes the query to create a second join to the same table.
I'm not proposing new language for this myself, but I've seen many an experienced developer get caught by this without understanding what was actually happening with the underlying query. I'd very much like to see these docs improved in some way.
Here's some shell output for those curious about what's happening:
And the queries: