Mark,
On 2004 Jul 15 , at 16.39, Mark Taylor wrote:
> I haven't looked closely at what you're doing here, but my first
> reaction is that constructing the whole tree from / in temporary
> and copying it to / is going to be more precarious than only
> constructing the bit that goes under prefix and copying that to
> the prefix directory. But I expect you've got some good reason
> for it (which I don't require you to explain...).
It does seem a bit odd, indeed, but the motivation is really that it
helps feed the `real' install target's delusion that it's installing
into the real location, on a general principle of least surprise.
Having said that, and since the real install has no business touching
anything _above_ $prefix, it would probably be better to go down to
that point in the tree, as you suggest, and copy from there.
pax is tempting but problematic: POSIX doesn't mandate symlinks, and so
the pax specification mentions them only conditionally, so some pax
implementations have a tendency to stomp on them (you wouldn't want
/mbt/starauto suddenly being turned into a directory containing just
tcl, for example). You can't _really_ rely on tar being present (it
wasn't by default on OSX until 10.2, I think, and the pax was a
stomp-on-the-link one). File-by-file install is tempting (if
time-consuming), but probably has problems if there are any file types
outside file/directory/symlink. I think the semantics of POSIX cp are
adequate, and adequately portably implemented, that that's the way to
go, but it looks like it needs substantially more smartness in how it
chooses its targets.
I probably oughtn't to care or worry about special files, but a fully
general solution would be good, if only to solve someone's headache
when they do install such an application. Hmm: this doesn't have to be
fully general, and only really has to cope with the applications in our
tree.
This isn't an explanation as much as an out-loud agonise. I'm taking
tomorrow and Monday off, so it'll be Tuesday before I get back to this,
I'm afraid.
See you,
Norman
--
----------------------------------------------------------------------
Norman Gray : Physics & Astronomy, Glasgow University, UK
http://www.astro.gla.ac.uk/users/norman/ : www.starlink.ac.uk
|