3 ms·

What reading articles on how bad ORMs are has taught me: More people need to use Django. Django does a lot and does it well.

collyw 10y ago
I wouldn't say the ORM is Djangos strongest point. Its great for simple stuff, but not great for any moderately complex queries (e.g. extra conditions on a join, subqueries, conditional aggregates - though I think that last one may be solved in the most recent version). I don't think they have done a bad job with the ORM, but you really ought to know SQL as well.
- jsmeaton 10y ago
  I think the ORM is Djangos strongest point. Take the ORM away and then ModelForms and Admin go with it. I think Django makes little sense if you're not using the ORM. That said, you're absolutely right that it is really bad at certain types of queries. There's no way to really influence joins. Subqueries are coming (https://github.com/django/django/pull/6478 https://github.com/django/django/pull/6478), and conditional aggregates became possible in Django 1.8 with Expressions support (which I authored). Queries that map over many to many relations that have .exclude() or .annotate() generally produce incorrect results. Most of the problems with the ORM are as a result of Django handling joins for the user - which is also one of its biggest strengths. You should absolutely know SQL if you're using an ORM. If you're doing anything other than basic CRUD, you have to know SQL. An ORM should give you an escape hatch to write that custom SQL. Django's ORM does not save you from all the problems mentioned in the article.
- oliwarner 10y ago
  I'm not arguing that people shouldn't learn SQL, I'm just saying that a lot of people who bemoan all ORMs should try using a really good one —like Django's— and see if all their complaints still ring true. Your examples, for example, might have been true a while ago but: - Extra conditions on a join (eg a prefetch filter): https://docs.djangoproject.com/en/1.9/ref/models/querysets/#django.db.models.Prefetch https://docs.djangoproject.com/en/1.9/ref/models/querysets/#... - I'm not sure what you think you'd want explicit subqueries for, at least in a context that annotations, prefetch, etc don't already do. If it's that weird, just use `.extra(..)` - Conditional aggregates: https://docs.djangoproject.com/en/1.9/ref/models/conditional-expressions/ https://docs.djangoproject.com/en/1.9/ref/models/conditional... - Not that you asked but also have a look at the new database functions stuff. http://stackoverflow.com/questions/38017076/annotate-a-comma-separated-list-of-related-items-onto-django-queryset http://stackoverflow.com/questions/38017076/annotate-a-comma... And it keeps getting better. The list of "ORMs are rubbish because I can't do X" is evaporating. Yeah there's probably some stuff that never comes up but you can still run raw SQL through the ORM. When you consider the development time saved and performance improvements (caching, good query writing, etc) an ORM delivers, I personally think it's hard to argue against them.

collyw 10y ago

I wouldn't say the ORM is Djangos strongest point. Its great for simple stuff, but not great for any moderately complex queries (e.g. extra conditions on a join, subqueries, conditional aggregates - though I think that last one may be solved in the most recent version). I don't think they have done a bad job with the ORM, but you really ought to know SQL as well.

jsmeaton 10y ago
I think the ORM is Djangos strongest point. Take the ORM away and then ModelForms and Admin go with it. I think Django makes little sense if you're not using the ORM. That said, you're absolutely right that it is really bad at certain types of queries. There's no way to really influence joins. Subqueries are coming (https://github.com/django/django/pull/6478 https://github.com/django/django/pull/6478), and conditional aggregates became possible in Django 1.8 with Expressions support (which I authored). Queries that map over many to many relations that have .exclude() or .annotate() generally produce incorrect results. Most of the problems with the ORM are as a result of Django handling joins for the user - which is also one of its biggest strengths. You should absolutely know SQL if you're using an ORM. If you're doing anything other than basic CRUD, you have to know SQL. An ORM should give you an escape hatch to write that custom SQL. Django's ORM does not save you from all the problems mentioned in the article.
oliwarner 10y ago
I'm not arguing that people shouldn't learn SQL, I'm just saying that a lot of people who bemoan all ORMs should try using a really good one —like Django's— and see if all their complaints still ring true. Your examples, for example, might have been true a while ago but: - Extra conditions on a join (eg a prefetch filter): https://docs.djangoproject.com/en/1.9/ref/models/querysets/#django.db.models.Prefetch https://docs.djangoproject.com/en/1.9/ref/models/querysets/#... - I'm not sure what you think you'd want explicit subqueries for, at least in a context that annotations, prefetch, etc don't already do. If it's that weird, just use `.extra(..)` - Conditional aggregates: https://docs.djangoproject.com/en/1.9/ref/models/conditional-expressions/ https://docs.djangoproject.com/en/1.9/ref/models/conditional... - Not that you asked but also have a look at the new database functions stuff. http://stackoverflow.com/questions/38017076/annotate-a-comma-separated-list-of-related-items-onto-django-queryset http://stackoverflow.com/questions/38017076/annotate-a-comma... And it keeps getting better. The list of "ORMs are rubbish because I can't do X" is evaporating. Yeah there's probably some stuff that never comes up but you can still run raw SQL through the ORM. When you consider the development time saved and performance improvements (caching, good query writing, etc) an ORM delivers, I personally think it's hard to argue against them.