Re: [PATCH] build fix when source tree is fully read-only

["Jan Beulich" <jbeulich at novell dot com>]

>The problem here seems odd to me.  Why are these directories and their
>contents being created as read-only?  If that itself is a build issue,
>should it not be corrected rather than its result (giving an
>unnecessary chmod on all runs)?  If the tree has been changed manually
>by the user to read only, then, as Andrew says, we shouldn't override
>that.

When I take a snapshot from the repository (from any repository I build
from, not just gcc), I change the source tree with 'chmod -R a-w' to as
much as possible prevent accidental modification (I retain ownership of
the files, though, to not overly complicate eventual fixing of the
snapshot itself, i.e. in order to allow a reference build - without any
patches applied - to succeed, which is exactly the case here).

This generally doesn't represent a problem anywhere (though there
are occasional breakages introduced which usually face no resistance
to fix). As described in an earlier mail, what the patch changes (and
what needs to change in some way, even if the patch is unacceptable
in its current shape for whatever reason) is not the (intentional by the
user) attributes of the source tree, but the attributes of the copy of it
made in the build tree. That is, if it is necessary to use 'cp -pR' to
create that copy, then deleting the copy must account for the
potential effect this has.

But as I also explained, the make rule in question is flawed anyway,
as it isn't properly dealing with being interrupted (e.g. user pressing
Ctrl-C, but also, as in the case given, failure of part of the rule) in
the middle. So perhaps one of you wants to fix the entire issue
rather than just the immediate issue of the build breaking?

Jan