This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: final install problems
- To: Joe Buck <jbuck at synopsys dot COM>
- Subject: Re: final install problems
- From: Zack Weinberg <zack at codesourcery dot com>
- Date: Fri, 2 Nov 2001 22:50:46 -0800
- Cc: David Edelsohn <dje at watson dot ibm dot com>,Phil Edwards <pedwards at disaster dot jaj dot com>,David <david_fly113 at yahoo dot com>, gcc at gcc dot gnu dot org
- References: <200111022207.RAA14712@makai.watson.ibm.com> <200111022324.PAA06468@atrus.synopsys.com>
On Fri, Nov 02, 2001 at 03:24:12PM -0800, Joe Buck wrote:
...
> Perhaps there's a difference if -m is not specified? (I can see two
> reasonable behaviors: preserve permission or use the umask of the
> person running the command).
According to the NetBSD install(1) man page, without -m the file
permissions are set to 0755. It looks like we rely on this, so I'd
better test it.
Does this pseudocode look complete to you?
for each 'install' found along $PATH
create scratch file S
install -c S T1
install -c -m 644 S T2
test that S still exists
test that T1 exists, has mode 0755, and is identical to S
test that T2 exists, has mode 0644, and is identical to S
> OK, but on systems like Solaris and AIX, I think we need to look
> for /usr/ucb/install even if /usr/ucb is not on the path.
I'm not sure - the user may have deliberately taken it out to prevent
our using it. Autoconf scripts generally trust the user to get their
path right, except where the binary is kept in a strange location
(/usr/lib/sendmail, for instance). We can always fall back to
install-sh.
zw