[Bug]: Meson build fails due to qhull link issue. #27159
Comments
|
Hi @Jacob-Stevens-Haas, since the move to meson, we need a different command to install. This is shown in the development version of the guide: |
|
Your initial error also looks like a (likely temporary) networking error. Perhaps the qhull download servers were offline for a bit. |
|
Thanks both! Correcting the Command
When using the new command from the development guide, I get the Error when building with: `python -m pip install --verbose --no-build-isolation --editable .[dev]`The same error occurs even if I replace the wrap file url/filename/hash with the one from github. Using system librariesUsing the correct command with system Qhull works. Thanks! Details: upgraded from pip 22.2 to pip 23.3, Still not able to download qhull manually
They're not offline, as |
|
if you do |
|
I'm not too concerned about https, because we do checksum the download (sha256) so having ssl during download doesn't provide much additional safety, actually. I am able to download the file (though admittedly firefox gives me a warning about https when I click the link from github) (I was also able to do |
|
Oh, that's a good point about the checksum. Today it did download (with the mentioned security risk). It might have just been an issue over the weekend. |
This comment was marked as outdated.
This comment was marked as outdated.
|
Meson does the downloading now, and that directory is not used any more. Tarballs are cached in |
|
oh, good to know! Hid my incorrect comment to not confuse people in the future. |
|
While I had been able to get around the issue in the past by using system libraries, I came back to work on a a new laptop at a new workplace and ran into the same issue of being unable to download from qhull servers. A desk neighbor mentioned that they too had had the same problem building matplotlib to make a PR in the past (#27158). I had assumed I was the only developer trying to work on matplotlib that ran into the qhull issue. But randomly meeting someone else, in-person, only a month later makes me think that a lot of would-be developers on academic/corporate networks could be getting stopped or burdened by this issue. My colleague believed it was because IT's firewall felt something qhull, perhaps because it was HTTP and not HTTPS. OTOH, the authors of qhull mentioned that the qhull server didn't support range requests, in case that had been the issue. I have a branch moving the build to qhull's github-hosted 8.0.2 tarball. My feeling is that github is generally a more stable place for code than a project like qhull's website. Would that be a welcome PR? |
|
I think moving to use the GH hosted tarball is reasonable (so we stop hammering the qhull server). When we were doing the downloading I think the tarballs were included in the directories that we had CI restore from cache. |
Bug summary
Building matplotlib fails due to "Connection reset by peer" when meson is building qhull. Perhaps relevant: qhull/qhull#132
Code for reproduction
Actual outcome
[Errno 104] Connection reset by peer
Expected outcome
builds successfully
Additional information
It could be an http vs https issue: the link (and all of qhull's page) is http. Chaning the link/file/hash on qhull.wrap to the equivalent tarball on qhull's github page at least advances the error to
I also tried installing libqhull8.0 and using system libraries, but got the same issue
Operating system
Linux 6.2.0-34-generic; Ubuntu 22.04.1
Matplotlib Version
masterfrom today: dfdb37aMatplotlib Backend
N/A
Python version
3.10.12
Jupyter version
N/A
Installation
git checkout
The text was updated successfully, but these errors were encountered: