You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I've been following the situation and not really sure where things are going. Like they highlighted in your link, who knows how long their fork will survive and I've been looking at niquests as well, which looks to solve almost all of the other problems I've run into with httpx. Specifically the issue about connection errors under load hit my team very hard and we had to switch to other libraries that didn't need gasapi.
Maybe the best solution is to fork this repo and do a find & replace for httpxyz in the meantime? Doesn't really scale that well though if projects come and go.
@JacobHenner are you still actively using this package? You've previously been on top of new HTTPX releases, so I'm curious what your current plans are.
@JacobHenner are you still actively using this package? You've previously been on top of new HTTPX releases, so I'm curious what your current plans are.
Yes, my organization is still using this package.
I've been following the situation and not really sure where things are going. Like they highlighted in your link, who knows how long their fork will survive and I've been looking at niquests as well, which looks to solve almost all of the other problems I've run into with httpx. Specifically the issue about connection errors under load hit my team very hard and we had to switch to other libraries that didn't need gasapi.
This is similar to our experience. We've been following the situation with httpx, and my personal interpretation is not optimistic. While httpnext was still around I noted parallels between requests/requests3 and httpx/httpnext, and it does appear like httpnext met the same fate and was abandoned. We also were affected by connection errors under load, but we still very much require gssapi support.
Looks like the community is moving towards httpx2. What do you think about a breaking change and switch to httpx2 for httpx-gssapi?
We initially swapped our projects over to httpxyz, but this blog post from the httpxyz maintainer suggests the httpx2 fork should be viewed as the 'blessed' one. I agree with the thoughts expressed in that post.
https://pydantic.dev/docs is a solid home for httpx2 (it is difficult to think of a better place), so are there places where you need help to establish and test compatibility with it?
Sorry I haven't been actively pushing this forward.
I guess the important decision is should this package be renamed to httpx2-gssapi to ensure a clean break? While it's possible the APIs will be compatible between the two libraries for at least a while, it could get a bit nasty trying to figure out breakage for users who don't switch their clients down the line.
I'm also fine with forking this repo and archiving it, leaving the existing package and source frozen. Though we should push one last release to PyPI to add a notice to the README.
Changing the package name sounds like a lot of work for maintainers of this repo but also for each project that has httpx-gssapi as a dependency. My vote (I am not a maintainer) would be that you keep the current package name and test the two side by side and then when the time is right drop support for legacy httpx.
Changing the package name sounds like a lot of work for maintainers of this repo but also for each project that has httpx-gssapi as a dependency. My vote (I am not a maintainer) would be that you keep the current package name and test the two side by side and then when the time is right drop support for legacy httpx.
I'm currently the sole maintainer and it doesn't really matter to me, I'm just looking to avoid headaches for users that could come with dropping the current dependency.
The biggest problem is that when initializing a client, httpxexplicitly checks isinstance(auth, httpx.Auth), which means we can't just inherit from httpx2.Auth and have it be compatible with an httpx.Client.
So IMO, any way you cut it, you're introducing some sort of backwards incompatible change that downstream has to be aware of:
An environment variable to downgrade the backing library like in the proposed PR
This requires the user to know to set it and runs into conflicts in complex deployments where dependencies might have differing opinions on which library to use
Requiring a user to select the older library when initializing the class
Now we're requiring a code change and hot swapping self.__class__ in Auth__init__
Dropping httpx support altogether without changing the name
Those who aren't pinning the package version will be in for a rude awakening
Requiring adoption of a new package name via some find and replace
Requires migration but preserves existing behavior for those not pinning packages
I figure the last option is the most explicit, which does fit with the Zen.
Regardless, I'm willing to just roll with whatever is most popular/convenient for those actually using this package. So please provide feedback!
Also, once we've got this ironed out, I plan to jump straight to v2.0 to match httpx2. I was only waiting for v1.0 to match httpx since there were still the occasional breaking changes from upstream.
That only works if only one is installed... Most users of this package are going to be also listing httpx as a dependency and then if we swap to httpx2 as the required dependency, both will be installed.
Thanks for the feedback everyone. Once #67 is merged, I'm going to fork this repo to httpx2-gssapi in the same org and continue development from there. Then we'll do one last release here to ensure the docs are updated with a note about the migration.
Activity
aiudirog commentedon May 11, 2026
Yeah this is a tricky one.
I've been following the situation and not really sure where things are going. Like they highlighted in your link, who knows how long their fork will survive and I've been looking at niquests as well, which looks to solve almost all of the other problems I've run into with httpx. Specifically the issue about connection errors under load hit my team very hard and we had to switch to other libraries that didn't need gasapi.
Maybe the best solution is to fork this repo and do a find & replace for httpxyz in the meantime? Doesn't really scale that well though if projects come and go.
aiudirog commentedon May 31, 2026
@JacobHenner are you still actively using this package? You've previously been on top of new HTTPX releases, so I'm curious what your current plans are.
chassing commentedon Jun 3, 2026
Looks like the community is moving towards httpx2. What do you think about a breaking change and switch to httpx2 for
httpx-gssapi?[-]httpxyz support[/-][+]httpxyz/httpx2 support[/+]JacobHenner commentedon Jun 3, 2026
Yes, my organization is still using this package.
This is similar to our experience. We've been following the situation with
httpx, and my personal interpretation is not optimistic. Whilehttpnextwas still around I noted parallels between requests/requests3 and httpx/httpnext, and it does appear like httpnext met the same fate and was abandoned. We also were affected by connection errors under load, but we still very much require gssapi support.We initially swapped our projects over to httpxyz, but this blog post from the httpxyz maintainer suggests the httpx2 fork should be viewed as the 'blessed' one. I agree with the thoughts expressed in that post.
cclauss commentedon Sep 17, 2026
https://pydantic.dev/docs is a solid home for httpx2 (it is difficult to think of a better place), so are there places where you need help to establish and test compatibility with it?
aiudirog commentedon Sep 18, 2026
Sorry I haven't been actively pushing this forward.
I guess the important decision is should this package be renamed to
httpx2-gssapito ensure a clean break? While it's possible the APIs will be compatible between the two libraries for at least a while, it could get a bit nasty trying to figure out breakage for users who don't switch their clients down the line.I'm also fine with forking this repo and archiving it, leaving the existing package and source frozen. Though we should push one last release to PyPI to add a notice to the README.
cclauss commentedon Sep 18, 2026
Changing the package name sounds like a lot of work for maintainers of this repo but also for each project that has httpx-gssapi as a dependency. My vote (I am not a maintainer) would be that you keep the current package name and test the two side by side and then when the time is right drop support for legacy httpx.
aiudirog commentedon Sep 18, 2026
I'm currently the sole maintainer and it doesn't really matter to me, I'm just looking to avoid headaches for users that could come with dropping the current dependency.
The biggest problem is that when initializing a client,
httpxexplicitly checksisinstance(auth, httpx.Auth), which means we can't just inherit fromhttpx2.Authand have it be compatible with anhttpx.Client.So IMO, any way you cut it, you're introducing some sort of backwards incompatible change that downstream has to be aware of:
self.__class__inAuth__init__httpxsupport altogether without changing the nameI figure the last option is the most explicit, which does fit with the Zen.
Regardless, I'm willing to just roll with whatever is most popular/convenient for those actually using this package. So please provide feedback!
Also, once we've got this ironed out, I plan to jump straight to v2.0 to match
httpx2. I was only waiting for v1.0 to matchhttpxsince there were still the occasional breaking changes from upstream.cclauss commentedon Sep 18, 2026
aiudirog commentedon Sep 18, 2026
That only works if only one is installed... Most users of this package are going to be also listing
httpxas a dependency and then if we swap tohttpx2as the required dependency, both will be installed.cclauss commentedon Sep 18, 2026
Now I understand. Users would need to bring their own httpx which might break some projects.
sim0nx commentedon Sep 23, 2026
I would suggest to rename/fork this repo to httpx2-gssapi for the above mentioned reasons. It also seems cleaner.
aiudirog commentedon Sep 28, 2026
Thanks for the feedback everyone. Once #67 is merged, I'm going to fork this repo to
httpx2-gssapiin the same org and continue development from there. Then we'll do one last release here to ensure the docs are updated with a note about the migration.