Re: setup revisit

[Christopher Faylor <cgf at redhat dot com>]

On Sat, Mar 24, 2001 at 05:19:02PM +1100, Robert Collins wrote:
>Less messy than having the distribution channel completely external to
>the software package... (example: Change mount styles -> rebuild
>setup.exe). And I'm not suggesting a special case _beyond the
>bootstrap_. I'm suggesting that the bootstrap _be the special case_.

If we change the mount style then many existing cygwin applications
break.

>Installshield uses a bootstrap system. MS's Active Setup uses an
>external system. RedHat's rpm falls into A) (How do you install the
>kernel the 1st time? Not via rpm :] ). Windows update falls into A). MS
>Visual studio falls into B). Most end-user software falls into B). Some
>software (a growing number) fall into A) because they're including
>"updaters".

I like the consistency of setup.exe.  There is one method that updates
everything.  You can do this because we rely on an underlying Windows
OS being there.  Red Hat setup doesn't have that luxury.

We have discussed the issue of mount tables, etc., but given the pledge
of backwards compatibility that we have, I don't think this is an issue.

Maybe this really just argues for splitting the functionality that
setup.exe needs into a separately loadable module that is used by both
cygwin1.dll and setup.exe.

>Now I'll admin that in unix based environments you can upgrade the
>kernel and the packaing system in-place, without trickery... but at the
>moment we don't have that luxury. It's something I can and will attack.
>If I succeed with that, we should be able to update cygwin1.dll from
>within cygwin1.dll. Note: Like you, I'm not worried about the special
>cases: users of snapshots & developers. Cygwin B20 is an issue though..
>hmm.

snapshots != developers.  We ask end users to try snapshots.  We can't
penalize them for doing this.

cgf