updated asgi handler for disconnects does not process ROOT_PATH

[Michael Smith]

as a result of this PR, ​https://github.com/django/django/pull/16603, a root path, in this case, set via uvicornm is not handled in the response.

I have created a small PR to fix this issue at ​https://github.com/msmitherdc/django/pull/1.

[Michael Smith]

Replying to Michael Smith:

as a result of this PR, ​https://github.com/django/django/pull/16603, a root path, in this case, set via uvicornm is not handled in the response.

I have created a small PR to fix this issue at ​https://github.com/msmitherdc/django/pull/1.

django pr is ​https://github.com/django/django/pull/17828

[Mariusz Felisiak]

Thanks for the report, however, I'm puzzled, process_request() has never accepted or passed a scope (even before 1d1ddffc27cd55c011298cd09bfa4de3fa73cf7a), so how this can be a regression? Also, scope is an element of request so why you need to pass it separately and call set_script_prefix(). This should be already handled by ​ASGIHandler.

[Carlton Gibson]

Root path handling was adjusted in Uvicorn just recently. ​https://github.com/encode/uvicorn/pull/2213

[Michael Smith]

all i can say is that when we upgraded to django 5 from 4.2.9, all the template url references outside of static files lost their root_path prefixes. By making these changes I was able to get the root_path passed back to the templates. If you have another way to resolve this, please but I have not been able to get the ROOT_PATH from uvicorn asgi server passed back to our application without this. I used git bisect to identify the specific commit that caused this. And by adding the scope, it restored the django 4 behavior.

[Mariusz Felisiak]

all i can say is that when we upgraded to django 5 from 4.2.9

Are you sure that it's not related with the uvicorn update?

I used git bisect to identify the specific commit that caused this.

Can you add a regression test? You had to have one to actually bisect.

[Michael Smith]

I will add a regression test

[Michael Smith]

It may be that process_request does not need scope but just run_get_response. I will determine what exactly is needed with the regression test

[Carlton Gibson]

@Michael. set_script_prefix targets a Local, so if we change thread maybe the value gets lost, but I'm not seeing yet how that can be happening (in anything related to 1d1ddffc27cd55c011298cd09bfa4de3fa73cf7a) — the create_task call uses the current event loop, so the same thread... 🤔 — Happy to have a look once you've got a reproduce. Thanks!

[AJ Slater]

I've run into the same problem, except with hypercorn.
script prefixes are no longer prefixed from hypercorn root_path with Django 5.0.0+ They are prefixed with Django 4.2.11 and the same hypercorn version.

My workaround right now is to manually call set_script_prefix() in my urls.py:

from django.urls import set_script_prefix
from myapp.settings import MANUALLY_PARSED_HYPERCORN_SETTINGS

set_script_prefix(MANUALLY_PARSED_HYPERCORN_SETTINGS.root_path) 

urlpatterns = [ ... ]

[Carlton Gibson]

I've created a ​PR adding a test for this here. The test passes (on both main and stable/4.2.x branches).

Comment from there:

Possible change may be in 041b0a3 @sarahboyce: maybe folks have FORCE_SCRIPT_NAME set to a non-None value. That's the sort of thing one might have done when experimenting with mounting under a root_path.
If it's not that, then I'm out of ideas without a reproduce.

[Sarah Boyce <42296566+sarahboyce@…>]

In 4d2ef9b:

Refs #35169 -- Added test for ASGIRequest root_path handling.