Hi all,
If anyone followed the e-mail thread on yt-users between me and Carla
Bernhardt, they'd see that we had a lot of difficultly dealing with the
INST_CONDA=0 option in the install script.
This option reverts the install script to "legacy" mode, installing yt and
all of yt's dependencies from source rather than using miniconda and conda
packages to bootstrap a python environment.
The problem is that the packages that get installed must be manually and
laboriously updated. In practice this means packages are basically never
updated. This also means the script installs relatively old versions of
libraries (currently, matplotlib 1.5.1, numpy 1.11) simply because no one
has taken the time to go through the list of libraries installed and update
them. This means anyone who uses this option isn't benefiting from upstream
bugfixes. It also means we risk bitrot.
In practice, I don't think I've recently heard of a situation where
INST_CONDA=1 failed but INST_CONDA=0 did not.
The other option of course is for someone to step up and take
responsibility for updating the INST_CONDA=0 section of the install script.
I'd be happy to help out with that if anyone wants to volunteer.
However, without a maintainer, I think we are doing our users a disservice
by leaving this option available.
If we *do* eliminate this option, we immediately reap the following
benefits:
* The install script can be trimmed down by several hundred lines
* We can simplify the install-script based install instructions in the docs
* It becomes easier to test modifications to the install script because
there are fewer permutations to test.
Please let me know what you think. I'd especially like to hear from people
who are fans of INST_CONDA=0 or who recently used it with no issues.
-Nathan
