We often see people confused about which version of Cygwin in the
mailing list even when there is no "cygwin version mismatch" error.
Since the proposed change removes the error, I think that we'll now
likely see more confusion like "I installed the latest version to fix
problem X but it is still there!" And, we'll likely have issues where
the cygwin from directory A creates a file with permissions slightly
different from the cygwin in directory B. Or the pipes used by one
version of cygwin differ slightly from the pipes used by another.
But, anway, the meta issue that really troubles me is that, AFAIK,
Corinna has always agreed with the
one-Cygwin-per-system-unless-you-know-what-you're-doing rule. It was
complete news to me that she thought this was a good idea when I first
heard about it. If I had known she thought it was a good idea I would
have certainly thought twice about the policy.
The ability to run multiple versions of Cygwin has been in the DLL for a
while. I put it there. Corinna added a lot in her patch to make it
more solid but her change is no technological breakthrough (I'm not
saying that she is claiming that it is). We could have always have been
doing this. As far as I know, the only reason that we weren't doing
this is because we had a policy against it.
The proposed change is coming about because someone wanted to pay for
it. So, it seems like, for the second time, we're talking about
allowing a Red Hat customer to set policy for the Cygwin open source
project. I think that maybe everyone but me takes it as a given that
this should be allowed but I think it is a slippery slope. It certainly
does make it hard for me to be a co-project lead if I can't come to a
decision and know that it will stick.