CVS timestamps [WAS Re: 002-gary-error-msg-quoting-test]

[gary]

address@hidden wrote:
> Gary> I have toyed with the idea of having bootstrap touch sensitive
> Gary> files in the correct order to prevent this kind of thing before.
> 
> I thought about this too, but thought it was also dangerous in the
> sense that if local *.y is more recent, Bison will really not be run.
> This is bad :(

It would be, but that wouldn't happen if the functionality was built in
to cvsutils:

  * clcommit would call, say, `savetimestamps' to commit a time ordered
    list (say `MANIFEST') of cvs controlled files.
  * `restoretimestamps' would set the times of each non-locally-modified
    file, and then touch each M file in pre-existing relative order.

The disadvantage would be that the restore step would probably force a
rebuild of all non-cvs controlled generated files.  But fixing even that
is just a "simple matter of programming" =)O|

> The real good fix is to have configure reject too old Bisons.

That would be bad too.  What if the latest release tarball is not new 
enough?  And even if it is, everyone who doesn't track cvs closely might
end up needing to fetch and build an alpha tarball each time they do an
occasional cvs bootstrap.  There is no need to enforce that when a 2 stage
build will normally suffice (lest configure refuses to continue because
of an old installed version).

Cheers,
        Gary.