Subject: Re: pkg/20978: mail/deliver is not compatible with recent NetBSD libc
To: None <email@example.com>
From: Ben Harris <firstname.lastname@example.org>
Date: 04/07/2003 18:46:14
In article <Pine.NEB.email@example.com> you write:
>On Mon, 7 Apr 2003, David Laight wrote:
>: > This is a severe problem. I've seen other code which depends on basename
>: > returning an in-situ pointer, and the change to libc (without versioning the
>: > call because of the ABI change) is alarming.
>: Such code is non-portable (aka broken).
>However, it was relying on a historic behavior of basename() that is *not*
>uncommon -- and *was* the behavior in NetBSD prior to the change.
Has anyone come up with any rationale _for_ the change? "It's permitted by
X/Open" is hardly much of a reason on its own.
>(See what I said about versioning the call because of the ABI change. This
>is about compatibiltiy with a precompiled executable, not compiling anew.)
That could be said for any change to the system's behaviour. For instance,
the change to top-down allocation of mmap'ed space might break a
particularly silly application, but that didn't mean that we needed to
version the ABI for it. Versioning the ABI is only sensible when an
application compiled before a change would break after it, but would work if
recompiled after the change.