Re: [Gnu-arch-users] library inode signature detection

[Tom Lord]

    > From: Robert Collins <address@hidden>

    > Heh, and uncovered a wackyness.

    > The addition of a new library revision breaks the inode signature on all
    > previous revisions with files in common. The ctime is altered.

Oh... of course.


    > Tom, is ctime needed in the inode signature at all? I'd have though
    > mtime was the key for detecting alterations.


Hmm.   ctime is not, strictly speaking, necessary -- as far as I can
tell at a quick glance.

However, I'd like to keep it.   It isn't currently used as a signature
that means "this file has/hasn't changed at all, in any respecct" --
it's only used as "the _contents_ of this file might have changed."
But the former is a clearer semantic, I think.   So, leave ctime.

I would say the problem is really that libraries are built with
`build_link_tree'.  It would be saner to build a link tree _only_ to
source files.

-t