Re: configure scripts rely on "diff"; could use "cmp" instead if available

[Eric Blake]

[re-adding the list]

On 05/25/2011 01:58 PM, John Hawkinson wrote:
> Eric Blake <address@hidden> wrote on Wed, 25 May 2011
> at 08:47:56 -0600 in <address@hidden>:
> 
>> Autoconf stated goal is to generate configure scripts that run on all
>> platforms that comply with the GCS set of minimum prerequisites - it so
> 
> Wow, this is a strange change. It used to be that autoconf's goal
> was to try to work on as many platforms as possible? When did that happen.

This is not a change.  It has _always_ been the case that autoconf tries
to work on as many platforms as possible.  But we have to draw the 'as
many as possible' line somewhere, and the GCS set of minimum apps is a
good place to draw it.  I'm perfectly willing to accept a patch that
changes autoconf to not use diff, if it is easy to maintain and will
allow us to go beyond the GCS bounds to support this busybox setup.
That is, a platform's non-compliance with GCS is not a reason to reject
a patch.

At the same time, the reason that I will not be writing such a patch
myself (unless GCS is changed), but am instead encouraging the busybox
setup to provide diff, is because that complying with the GCS has bigger
benefits to the busybox community than doing a one-off patch in
autoconf, only to discover that more one-off patches are needed
everywhere else.  And since I don't personally use this broken busybox
setup, I won't be the one arguing for a GCS change (at least, not any
more so than the fact that I've already cc'd bug-standards earlier in
this thread).

> 
> Since the whole point of busybox is to get by with as little as possible,
> asking it to add diff when autoconf doesn't really need it seems
> perverse.

Then make that argument to rms when you propose changing GCS to not
mandate diff in the minimum set of required apps.  It's not just
autoconf, but _all_ GNU applications that have thought (for years now)
that diff would blindly be available.  And if we get consensus that we
shouldn't be relying on diff, then I'll gladly help audit for places
that violate that assumption - but we have to get consensus that such
work will be worth the effort, first.  In my mind, it's cheaper to fix
the root cause, then to paper over the symptoms everywhere else.

-- 
Eric Blake   address@hidden    +1-801-349-2682
Libvirt virtualization library http://libvirt.org

signature.asc
Description: OpenPGP digital signature