Re: [PATCH v2 1/2] Allow deriving the current user's home directory via the HOME variable

[Johannes Schindelin <Johannes.Schindelin@gmx.de>]

Hi Corinna,

On Mon, 24 Oct 2022, Corinna Vinschen wrote:

> On Oct 23 23:04, Johannes Schindelin wrote:
> > On Tue, 18 Oct 2022, Corinna Vinschen wrote:
> > [...]
> > > That means, the results from the "env" method is equivalent to the
> > > "windows" method, just after checking $HOME.  That's a bit of a downer.
> > >
> > > Assuming the "env" method would *only* check for $HOME, the user would
> > > have the same result by simply setting nsswitch.conf accordingly:
> > >
> > >   home: env windows
> >
> > Except when the domain controller is (temporarily) unreachable, e.g. when
> > sitting in a train with poor or no internet connection. Then that latter
> > approach would have the "benefit" of having to wait 10-15 seconds before
> > the network call says "nope".
> >
> > This particular issue has hit enough Git for Windows users that I found
> > myself being forced to implement these patches and run with them for the
> > past seven years.
> >
> > Given the scenario of an unreachable domain controller, I hope you agree
> > that the `env` support added in the proposed patches _has_ merit.
>
> Yes, I don't doubt an `env' method checking for $HOME even a bit.

Cool!

> I'm just not sure as far as HOMEDRIVE/HOMEPATH/USERPROFILE are
> concerned.  Those vars should be left alone, but we can't control that,
> so reading them from genuine sources is preferred.

I do not recall the exact reasons because it has been a good while since I
worked on these patches. But I do remember that we had to have a fall-back
for the many scenarios in Git for Windows where `HOME` is not even set,
and we specifically had to add HOMEDRIVE/HOMEPATH handling because
USERPROFILE alone would lead to problems (IIRC there were plenty of
corporate setups where USERPROFILE pointed to a potentially-disconnected
network drive).

> Sure, the downside in terms of the LDAP server is clear to me
>
> So I guess it's ok to allow the env method to read the values of those
> vars from the env.  I would just feel better if we urge the
> user to set $HOME and read that exclusively.

I would feel better about that, too, if it was practical.

But I cannot ask millions of Git for Windows users to please go ahead and
first configure their `HOME` variable correctly, it took much less time to
implement the patch we're discussing than asking all users individually
;-)

And since there is nothing specific about Git for Windows here, I expect
Cygwin users to benefit from this feature, too.

With this context in mind, I would like to ask to integrate the patch
as-is, including the HOMEDRIVE/HOMEPATH and USERPROFILE fall-backs.

Thanks,
Dscho