Fix mamba init --no-user

[xylar]

This merge makes changes similar to
conda/conda#11949
to fix mamba init so that the --no-user flag actually has the intended effect of not changing a user's .bashrc or equivalent. Without this change, --no-user has no effect as was previously the case in conda, see:
conda/conda#11948

[xylar]

I tested this locally by making a fresh install of mambaforge, then doing:

python -m pip install -e mamba
mamba init --no-user
mamba init --user
mamba init

As expected, mamba init and mamba init --user both lead to a modified .bashrc, whereas mamba init --no-user does not modify my .bashrc.

This behavior is needed in situations where we do not wish to have mambaforge activate automatically in a user's shell (e.g. because we have multiple conda or mambaforge installations available on an HPC system and shell scripts that call python scripts may revert back to the wrong conda base environment).

A workaround has been to copy a user's .bashrc (etc.) to a temporary file, then copy it back after calling mamba init but this shouldn't be necessary if --no-user functioned properly.

[jonashaag]

Will this break backwards compat with older Conda versions?

[xylar]

@jonashaag, I can test that but I don't believe so beyond the fact that the --no-user option didn't work for any version prior to the latest release.

[xylar]

@jonashaag, yes, unfortunately with conda<23, this appears to have the undesirable effect of making the behavior of mamba init equivalent to mamba init --no-user, whereas with conda>=23, mamba init behaves like mamba init --user (as it is intended to).

I don't see an obvious way to fix this for all conda versions. I will give it some thought but would appreciate suggestions. It is broken as written now so I don't think leaving it as it is would be a very desirable option, especially because I put the work into fixing conda only to discover that mamba is bypassing my fix.

[jonashaag]

Thanks for checking! I think just adding a Conda version check will be fine:

if conda version is before removal of no_user:
  use old code (before #2249)
else:
  use your new impl

[xylar]

@jonashaag, do you have code somewhere for detecting the conda version? I didn't want to add any new dependencies for comparing versions but if you already have such a dependency,

For now, I'm using the presence of args.no_user as the way to detect this.

[jonashaag]

@jonashaag, do you have code somewhere for detecting the conda version? I didn't want to add any new dependencies for comparing versions but if you already have such a dependency,

For now, I'm using the presence of args.no_user as the way to detect this.

This is fine IMO, actually it’s better than checking for version: checking for availability of a feature. Version numbers are just proxies for that.

[xylar]

Okay, this is the block of code before #2249. So we'll behave as before if args.no_user is present but things will work properly (including the --no-user flag for newer versions where the argument is no longer there.

[xylar]

Thanks @jonashaag. I think we're on the same page. Let me know what you think at this point.

[jonashaag]

Awesome thanks so much!