Use urllib downloads instead of curl

loulawrence commented 3 years ago

urllib must manually decode responses to utf-8, which may cause errors on special characters, e.g. ñ, ö, ß . Can be ignored to pass tests, but may have unforeseen consequences.

BetsyMcPhail commented 3 years ago

Resolved by https://github.com/spack/spack/pull/21398

BetsyMcPhail commented 3 years ago

I have a WIP branch where all 'url_fetch' tests are passing on Windows --> https://github.com/BetsyMcPhail/spack/tree/urllib-windows

A few notes:

This almost certainly breaks Linux, there's no switching on OS.
For the most part, the changes deal with handling Windows filepaths
The changes are just good enough to get the tests passing, I have not thought about how robust they will be in other situations

Command to run tests: spack unit-test lib/spack/spack/test/url_fetch.py

FYI @loulawrence if you want to give the branch a try on your system

BetsyMcPhail commented 3 years ago

Thanks for testing @loulawrence, hopefully all of the tests should be passing on Windows now!

BetsyMcPhail commented 3 years ago

PR #21594 has the changes necessary for url_fetch test to pass on Windows.

To test, the Windows system must have 1. curl available and 2. #20412

Command to run tests: spack unit-test lib/spack/spack/test/url_fetch.py

spack / spack-bootstrap-installer

Use urllib downloads instead of curl #7