> > +INSTALL="install"
> > +printf "Checking for working \"install\" program... "
> > +mkdir _tmp_
> 
> This doesn't feel like a hot idea.

Out of curiosity, why not?  An "install" that behaves as expected is
one of the first things that an autoconf-generated "configure" looks
for.  Now, autoconf-configure implements that check using some
assumptions about where things are on different operating systems,
but that sort of check runs the risk of becoming stale (see below).

> Don't tell me you'd need to create a compatibility script for using
> mktemp --tmpdir too...

Yes I would, if it were used; not all the world's a GNU.  But in the
case of mktemp, the widely available "-p" and "-t" options seem to get
you most of the way there.  The SVR4 "install" in Solaris' /usr/sbin is
different enough from the BSD/GNU versions that I wouldn't want to try
to emulate it with a wrapper.

> Or how about just always using ginstall on Solaris?

I'd rather not do that.  With the old UCB tools having been EOL'ed [1],
/usr/ucb/install (which would have worked) will be going away.  There is
an open bug in the Sun bug tracking system about moving GNU install to
/usr/bin/install, specifically motivated by this change.  So while I
don't know if/when that bug will be fixed, I would guess [2] that a
future release of Solaris may have a BSD/GNU-compatible version of
"install" in the default $PATH.

Blake

[1] http://www.oracle.com/technetwork/systems/end-of-notices/eonsolaris11-392732.html
[2] Caveat: I work for Oracle in the Solaris kernel group, but I am not
    speaking for my employer.