uv fails to install jupyter_contrib_nbextensions==0.7.0 while pip succeeds

[leodevian]

Hello,

I'm using uv 0.4.30 (61ed2a2 2024-11-04) on Windows and I'm trying to install jupyter_contrib_nbextensions==0.7.0. It seems that uv fails to install the package as it cannot be built with the latest versions of setuptools, due to some deprecations.

uv pip install jupyter_contrib_nbextensions==0.7.0

However, the latest version of pip (24.3.1) succeeds to build and install the package and I would like to understand this difference in behavior.

pip install jupyter_contrib_nbextensions==0.7.0

I read the documentation on build failures and I don't understand if I can force a version of setuptools and wheel for jupyter_contrib_nbextensions.

Also, can I manage this package with uv if it has been installed using pip?

Thank you for your awesome work!

[FishAlchemist]

I'm a Windows user and I couldn't reproduce the problem you mentioned.
Furthermore, according to your description, the package installed with uv is not the same as the one installed with pip. (Corrected)

[FishAlchemist]

@leodevian
I suggest you provide more complete information. After all, if we can't reproduce the issue, we won't know what the error message is. As for the method, at least add --verbose to get complete information.

[leodevian]

Thanks for your quick feedback! Indeed, I did mispell the package name in the description (now edited).

The following command should reproduce the issue:

mkdir issue
cd issue
uv venv -p 3.11 *> log.txt
uv pip install jupyter_contrib_nbextensions==0.7.0 -vvv *>> log.txt

I'm using cpython-3.11.10-windows-x86_64-none, which has been installed using uv.

Here are the logs with --verbose (I'm not used to redirection and struggled to get them).

log.txt

[FishAlchemist]

@leodevian Could you provide the pyproject.toml that can reproduce this issue? Because when I tried your problem, I used a default pyproject.toml.

Edit:
But it has a lot of warnings.

[leodevian]

@FishAlchemist I don't have pyproject.toml in my directory.

I tried the following commands and it still failed with the same error:

mkdir issue
cd issue
uv init -p 3.11
uv add jupyter_contrib_nbextensions==0.7.0

It seems that pip succeeds to build a wheel while uv fails?

EDIT: I don't understand the difference in behavior. Does pip support deprecated methods that have not been implemented in uv?

[charliermarsh]

Are you perhaps hitting the filename length limit on Windows?

[leodevian]

The directory is located in %USERPROFILE%, which is 18 characters long, so I doubt it.

I will try to reproduce the issue on my personal computer later, to investigate further.

[FishAlchemist]

Are you perhaps hitting the filename length limit on Windows?

It seems possible, as my computer enable Long Paths.

To check if it's enabled, you can simply open PowerShell to query. Reading the registry doesn't require administrative privileges.

Get-ItemProperty -Path "HKLM:\SYSTEM\CurrentControlSet\Control\FileSystem" `
∙ -Name "LongPathsEnabled"

[charliermarsh]

Yes but the error you shared is for build\lib\jupyter_contrib_nbextensions\nbextensions\code_prettify\README_code_prettify.md which will already be nested inside the build cache directory.

[leodevian]

I don't have it and I have no admin rights on this computer. I will give your more information once I tried to reproduce the issue on my personal computer (with admin rights).

Could it be that building universal wheels is not supported by uv? To me, it seems that this feature has been deprecated for quite some time...

From the logs:

With Python 2.7 end-of-life, support for building universal wheels
        (i.e., wheels that support both Python 2 and Python 3)
        is being obviated.

[FishAlchemist]

Yes but the error you shared is for build\lib\jupyter_contrib_nbextensions\nbextensions\code_prettify\README_code_prettify.md which will already be nested inside the build cache directory.

@charliermarsh Yes, I just closed the long path, and then the problem occurred

Edit:
Can UV detect long paths that are not enabled during the build phase?

---

@leodevian This is the configuration method provided by Microsoft.
https://learn.microsoft.com/en-us/windows/win32/fileio/maximum-file-path-limitation?tabs=powershell#registry-setting-to-enable-long-paths

[leodevian]

@FishAlchemist Thanks a lot! I guess that's good news.

It pains me because I'm promoting uv within my organization and no one will be able to disable the maximum file path limitation because of internal policies...

[leodevian]

Is it possible to bypass the limitation using \\?\?

Here is how I would do it using Python.

import os

longpath = "..."

longpath = os.path.realpath(longpath)

unc = "\\\\?\\" if len(longpath) >= 260 else ""

os.mkdir(unc + longpath)

with open(unc + os.path.join(longpath, "hello.txt"), "w") as f:
    f.write("Hello world!")

[charliermarsh]

I don't believe so. Using those UNC paths breaks a lot of things -- for example, any relative paths in the build will be interpreted relative to the drive: #1277

[leodevian]

Well, I had mitigated hopes on this... I hope that I won't encounter this issue too many times. Meanwhile, I'll look for a workaround.

Is it possible to install a package using pip and still manage it using uv (so it won't be uninstalled with uv sync or uv add)?

[charliermarsh]

I'll look into reducing the length of the cache path.

[charliermarsh]

I actually don't really understand why this succeeds with pip but not uv. Their build cache prefix isn't any shorter, and even when I set --cache-dir to something short, uv still fails.

[FishAlchemist]

@charliermarsh I've tracked program's behavior, and it should serve as a reference.
The nesting of the UV cache seems quite deep, but pip seems to have only the nesting of the source code.

UV (uv 0.4.30 (61ed2a2 2024-11-04))

( maximum 233 without user home path)

~\AppData\Local\Temp\.tmpyBcss8\sdists-v5\pypi\jupyter-contrib-nbextensions\0.7.0\Jk8_nRpxZF8Zdq-pjfEJr\jupyter_contrib_nbextensions-0.7.0.tar.gz\build\lib\jupyter_contrib_nbextensions\nbextensions\code_prettify\README_code_prettify.md

PIP (pip 24.3.1)

(maximum 209 without user home path)

~\AppData\Local\Temp\pip-install-7pgurthd\jupyter-contrib-nbextensions_0d7b5c639ab74a4fb664a5235eb7c74d\src\jupyter_contrib_nbextensions\nbextensions\code_prettify\README_code_prettify.md
~\AppData\Local\Temp\pip-install-7pgurthd\jupyter-contrib-nbextensions_0d7b5c639ab74a4fb664a5235eb7c74d\build\lib\jupyter_contrib_nbextensions\nbextensions\code_prettify\README_code_prettify.md
~\AppData\Local\Temp\pip-install-7pgurthd\jupyter-contrib-nbextensions_0d7b5c639ab74a4fb664a5235eb7c74d\build\bdist.win-amd64\wheel\jupyter_contrib_nbextensions\nbextensions\code_prettify\README_code_prettify.md

[konstin]

Collecting clues:

With pip install jupyter_contrib_nbextensions==0.7.0:

  copying build\lib\jupyter_contrib_nbextensions\nbextensions\codefolding\main.js -> build\bdist.win-amd64\wheel\.\jupyter_contrib_nbextensions\nbextensions\codefolding
  copying build\lib\jupyter_contrib_nbextensions\nbextensions\codefolding\readme.md -> build\bdist.win-amd64\wheel\.\jupyter_contrib_nbextensions\nbextensions\codefolding
  creating build\bdist.win-amd64\wheel\jupyter_contrib_nbextensions\nbextensions\codemirror_mode_extensions
  copying build\lib\jupyter_contrib_nbextensions\nbextensions\codemirror_mode_extensions\codemirror_mode_extensions.yaml -> build\bdist.win-amd64\wheel\.\jupyter_contrib_nbextensions\nbextensions\codemirror_mode_extensions

With uv pip install jupyter_contrib_nbextensions==0.7.0, but also with uv pip install jupyter_contrib_nbextensions==0.7.0 --cache-dir C:\Users\Konsti\a:

copying src\jupyter_contrib_nbextensions\nbextensions\codefolding\main.js -> build\lib\jupyter_contrib_nbextensions\nbextensions\codefolding
copying src\jupyter_contrib_nbextensions\nbextensions\codefolding\readme.md -> build\lib\jupyter_contrib_nbextensions\nbextensions\codefolding
creating build\lib\jupyter_contrib_nbextensions\nbextensions\codemirror_mode_extensions
copying src\jupyter_contrib_nbextensions\nbextensions\codemirror_mode_extensions\codemirror_mode_extensions.yaml -> build\lib\jupyter_contrib_nbextensions\nbextensions\codemirror_mode_extensions

With uv pip install jupyter_contrib_nbextensions==0.7.0 --cache-dir asdf in C:\Users\Konsti\projects\uv\:

copying src\jupyter_contrib_nbextensions\nbextensions\codefolding\main.js -> build\lib\jupyter_contrib_nbextensions\nbextensions\codefolding
copying src\jupyter_contrib_nbextensions\nbextensions\codefolding\readme.md -> build\lib\jupyter_contrib_nbextensions\nbextensions\codefolding
creating build\lib\jupyter_contrib_nbextensions\nbextensions\codemirror_mode_extensions
copying src\jupyter_contrib_nbextensions\nbextensions\codemirror_mode_extensions\codemirror_mode_extensions.yaml -> build\lib\jupyter_contrib_nbextensions\nbextensions\codemirror_mode_extensions
copying src\jupyter_contrib_nbextensions\nbextensions\codemirror_mode_extensions\main.js -> build\lib\jupyter_contrib_nbextensions\nbextensions\codemirror_mode_extensions
creating build\lib\jupyter_contrib_nbextensions\nbextensions\collapsible_headings
copying src\jupyter_contrib_nbextensions\nbextensions\collapsible_headings\collapsible_headings.yaml -> build\lib\jupyter_contrib_nbextensions\nbextensions\collapsible_headings
copying src\jupyter_contrib_nbextensions\nbextensions\collapsible_headings\icon.png -> build\lib\jupyter_contrib_nbextensions\nbextensions\collapsible_headings
copying src\jupyter_contrib_nbextensions\nbextensions\collapsible_headings\main.css -> build\lib\jupyter_contrib_nbextensions\nbextensions\collapsible_headings
copying src\jupyter_contrib_nbextensions\nbextensions\collapsible_headings\main.js -> build\lib\jupyter_contrib_nbextensions\nbextensions\collapsible_headings
copying src\jupyter_contrib_nbextensions\nbextensions\collapsible_headings\readme.md -> build\lib\jupyter_contrib_nbextensions\nbextensions\collapsible_headings
copying src\jupyter_contrib_nbextensions\nbextensions\collapsible_headings\screenshot.png -> build\lib\jupyter_contrib_nbextensions\nbextensions\collapsible_headings
creating build\lib\jupyter_contrib_nbextensions\nbextensions\comment-uncomment
copying src\jupyter_contrib_nbextensions\nbextensions\comment-uncomment\comment-uncomment.yaml -> build\lib\jupyter_contrib_nbextensions\nbextensions\comment-uncomment
copying src\jupyter_contrib_nbextensions\nbextensions\comment-uncomment\icon.png -> build\lib\jupyter_contrib_nbextensions\nbextensions\comment-uncomment
copying src\jupyter_contrib_nbextensions\nbextensions\comment-uncomment\main.js -> build\lib\jupyter_contrib_nbextensions\nbextensions\comment-uncomment
copying src\jupyter_contrib_nbextensions\nbextensions\comment-uncomment\readme.md -> build\lib\jupyter_contrib_nbextensions\nbextensions\comment-uncomment
creating build\lib\jupyter_contrib_nbextensions\nbextensions\contrib_nbextensions_help_item
copying src\jupyter_contrib_nbextensions\nbextensions\contrib_nbextensions_help_item\README.md -> build\lib\jupyter_contrib_nbextensions\nbextensions\contrib_nbextensions_help_item
copying src\jupyter_contrib_nbextensions\nbextensions\contrib_nbextensions_help_item\contrib_nbextensions_help_item.yaml -> build\lib\jupyter_contrib_nbextensions\nbextensions\contrib_nbextensions_help_item

I can also uv build in the repo

[charliermarsh]

@konstin -- You may need to use --use-pep517 in pip install (it still failed for me, but I think the behavior may be different).

[charliermarsh]

Oh my guess is the problem is this huge prefix? sdists-v5\pypi\jupyter-contrib-nbextensions\0.7.0\Jk8_nRpxZF8Zdq-pjfEJr\jupyter_contrib_nbextensions-0.7.0.tar. I need to trace this.

[charliermarsh]

So sketching this out... When we go to install that wheel:

1. We create a new revision in the cache, like ${UV_CACHE_DIR}/sdists-v5/pypi/jupyter-contrib-nbextensions/0.7.0/Jk8_nRpxZF8Zdq-pjfEJr.
2. We stream-unzip the contents into a temporary directory within ${UV_CACHE_DIR}/sdists-v5, like ${UV_CACHE_DIR}/sdists-v5/.tmpyBcss8.
3. We move the temporary directory to into the sdists-v5 bucket, e.g., we move it to ${UV_CACHE_DIR}/sdists-v5/pypi/jupyter-contrib-nbextensions/0.7.0/Jk8_nRpxZF8Zdq-pjfEJr/jupyter_contrib_nbextensions-0.7.0.tar.gz (this is a directory, despite the filename).
4. We then build the source that has been extracted to ${UV_CACHE_DIR}/sdists-v5/pypi/jupyter-contrib-nbextensions/0.7.0/Jk8_nRpxZF8Zdq-pjfEJr/jupyter_contrib_nbextensions-0.7.0.tar.gz.
5. To build, we create a virtual environment in a temporary directory in ${UV_CACHE_DIR}/builds-v0, like ${UV_CACHE_DIR}/builds-v0/.tmay5cs91.
6. To build, we create a temporary directory within the target directory... So, like: ${UV_CACHE_DIR}/sdists-v5/pypi/jupyter-contrib-nbextensions/0.7.0/Jk8_nRpxZF8Zdq-pjfEJr/.tmpQWxTfb. Then we build the wheel using and into that directory (i.e., it's the output directory passed to the PEP 517 hooks).
7. We then rename the built wheel them to, e.g., ${UV_CACHE_DIR}/sdists-v5/pypi/jupyter-contrib-nbextensions/0.7.0/Jk8_nRpxZF8Zdq-pjfEJr/interpreters_pep_734-0.4.0-py3-none-any.whl (this is a file).
8. We then unzip the built distribution:
   • First, we create a temporary direcotry in the cache root (e.g., ${UV_CACHE_DIR}/.tmay5cs91).
   • Then we unzip into that directory.
   • Then we persist it in the archive bucket (e.g., we rename it to ${UV_CACHE_DIR}/archive-v0/5ZQyJUPRJyOe8jCGM3gql).
   • Then we create a symlink from the archive bucket to the sdists-v5 bucket (e.g., ${UV_CACHE_DIR}/sdists-v5/pypi/jupyter-contrib-nbextensions/0.7.0/Jk8_nRpxZF8Zdq-pjfEJr/interpreters_pep_734-0.4.0-py3-none-any).

[charliermarsh]

I think one thing we should definitely change is that we shouldn't build the wheel into ${UV_CACHE_DIR}/sdists-v5/pypi/jupyter-contrib-nbextensions/0.7.0/Jk8_nRpxZF8Zdq-pjfEJr/.tmpQWxTfb. That's just a huge path and there's no reason for it? I'll do that first.

[charliermarsh]

We may also want to consider shortening ${UV_CACHE_DIR}/sdists-v5/pypi/jupyter-contrib-nbextensions/0.7.0/Jk8_nRpxZF8Zdq-pjfEJr. It's just a huge path. And if you use a non-PyPI index, it's even longer because the pypi segment gets expanded to, e.g., index/6934cf8789bf06c0.

[charliermarsh]

Okay cool, I just got this package to build on my machine. It'll be a series of small PRs.