Autoescaping variable input in template tags

[Andrew Badr]

(revised description)

When variables are used as input to template tags, we should handle autoescaping a bit better.

Original description was even shorter:

I'd expect firstof to escape variables.

[Andrew Badr]

Fix, with test

[Andrew Badr]

My expected behavior for Django is: The results of all template tags should be escaped unless marked safe. For builtin templatetags that return safe strings, this should include escaping any variables involved in rendering that template tag.

The included patch solves locally what I see as part of a bigger problem.

[Malcolm Tredinnick]

Hmm. This is a bit of a slight change in policy, but not necessarily a bad one (right now, intentionally, only filters autoescape). However,

1. Import _render_change_in_context() smells and not of roses. Needs to be renamed or not imported or something. That leading underscore is a big clue.
2. We need something similar to the autoescape parameter that is passed to filters, so that template tags do not autoescape results when they are being used inside a block that does not do auto-escaping.

I guess template tags know whether they're putting the results into the context -- which means the output shouldn't be autoescaped, since it will be autoescaped on output and you don't know at template tag usage time whether output will involve auto-escaping or not, so it must be delayed until rendering -- or directly displaying the results? I suspect so, but that's something to think about and confirm.

What are the backwards compatibility issues here? In particular, right now, a template tag doesn't have to care about calling mark_safe() on the result, since it just inserts it into the template. If you add the auto-escaping behaviour at the wrong place, you'll undo that and a lot of existing template tags will break. That's going to be a little fiddly to work around.

I'm unenthusiastic about fixing just this single tag, since it does change the currently consistent behaviour across all template tags. I'm more enthusiastic about addressing the general issue, but it's a lot more tricky. I'm really enthusiastic about you doing it, Andrew, since I know you guys work in the template space a lot, so the results will have some validity and it means I don't have to work on this (that latter point is important to me; maybe less so to you).

[krystal]

Just to mention that I just discovered a flaw on my website due to this non-escaping-feature when using firstof tags ; It should really be at least explicitly documented so people get aware of this while reading the firstof (or cycle) doc.

A little paragraph with a one line example should be nice until a good way to do is found.

[krystal]

I move this to milestone 1.1 so the big-boss-team give his opinion ; default django security is involved and the doc clearly say that "firstof" is equivalent to {%if bla %}{{varx}} for now, it's very missleading.

I can wrote a patch for the documentation if needed, just tell me if I have to open a new ticket.

[Andrew Badr]

I'm not big-boss-team, but this feature is definitely not making it into 1.1, so I'm removing the milestone. I think your concern about documentation is valid, but I would make a new ticket.

[Russell Keith-Magee]

Promoting this to ensure it is looked at in 1.3

[Chris Beaven]

Guess it didn't get looked at in 1.3... I'm going to bump severity.

Since the accepted was 2 years ago, I'm also going to push back to a design decision.

[kenth]

There are two template tags which generate output that is not autoescaped: firstof & cycle. This behavior might be unexpected, but it appears to be well documented even back in the version 1.0 documents. There is a policy statement in the documentation stating output is not escaped "... because template tags do not escape their content."

The code appears to work as documented & the documentation appears to be clear. As to if the behavior is as it should be, that's clearly a design decision...

[Adrian Holovaty]

In [17176]:

Added a test that 'firstof' template filter doesn't auto-escape. cycle tag already has such a test (cycle20). Refs #10912

[Adrian Holovaty]

In [17177]:

Tweaked templates/builtins.txt to make it clearer that cycle and firstof filters don't auto-escape. Refs #10912

[Adrian Holovaty]

OK, I'm marking this as fixed because the documentation is clear about the fact that firstof and cycle don't auto-escape. I also added a test for firstof (we already had one for cycle).

Just for future reference, we added some original docs in [11163].