Support included blocks override

[paradoxxxzero]

This commit fix inheritance with included blocks.
Consider these templates:

A.jinja2

I am A
{% include 'I.jinja2' %}

I.jinja2

Who are you ?
{% block whoami %}
    I am I
{% endblock %}

B.jinja2

{% extends 'A.jinja2' %}
{% block whoami %}
    I am B
{% endblock %}

As of now, rendering B.jinja2 would output:

I am A
Who are you ?
I am I

This patch allow B to override I block, the output will be instead:

I am A
Who are you ?
I am B

[mitsuhiko]

This changes behavior and was never intended to be supported. However since the whole scoping is something I want to overhaul I will decide in neither way right now and try to find a solution that is compatible with the current behavior and would support blocks be defined in included templates.

[SimonSapin]

How is the proposed patch not compatible with the current behavior?

[equake]

I would love to be able to override blocks like this.
In my case, i have a base template that is extended by a lot of templates.
Sometimes those templates have to include "components" (for example: an slideshow).
This include need to add some <script> tag in the page header to work properly. It would be nice if it could override/extends the "scripts" block defined in the base template.
Is there any alternative way to accomplish this with jinja?

[mitsuhiko]

I merged this now. I don't see how this breaks anything currently.

[navilan]

@mitsuhiko - This change breaks a few websites that use hyde. Here is a simple demo:
https://github.com/hyde/j2includebug.

---

Pushing parent blocks along with other context variables is surprising IMO. I'd like to propose:

{% include "xxx" with context and blocks %}

or something similar instead if its not too late.

Edit:

Here is (hopefully) a better statement of the bug:

Given a container C and an include I, this patch really includes container C with the include I injected at the top of the extension hierarchy of C. As long as I and C have completely different set of blocks this is okay. However, when there is an overlap between the blocks of I and C, the results will likely be surprising(bad).

[navilan]

I think its safer to completely pull this change out and reintroduce it only if a block resolution order can be provided alongside.

[ripperdoc]

To add to this, this update seem to break my site badly (unless I'm doing something else wrong of course...)

I have three template files, A, B and Base.

Base
{% block content %}{% endblock}

A
{% extends Base %}
{% block content %}
lorem ipsum {% include B %}
{% endblock}

B
{% block content %}
dolor sit
{% enblock %}

This throws me in maximum recursion, as (if I understand it correctly), in A.content we include B.content which was defined in A.content and so on? I had the same structure which was not an issue before, I presume before this update the block content in the included template was simply ignored?

I can include without context but that will break other things, e.g. not being able to pass my model objects down to the included templates.

[SuryaSankar]

Will this feature be revived? Would be very happy to see this. The lack of this feature is the only issue I have had so far with Jinja2.