Half of PR/12744. Allow tarball builds without flex and bison/byacc

[Joseph S. Myers]

On Wed, 14 Jan 2004, Nathanael Nerode wrote:

> This means the following, I believe:
> We have to make very sure that the timestamps are in the correct order 
> in the tarball.
> If the timestamp for c-parse.in somehow gets newer than the timestamp 
> for c-parse.y, or the timestamp for c-parse.y somehow gets newer than 
> the timestamp for c-parse.c, we'll be back to needing bison, etc. when 
> make runs.
> 
> Ought we to make some note about the timestamp skew issue somewhere?  In 
> the notes about generating a release tarball, or something?

I don't see how such a problem could occur unless the system building the
release has a wrong clock (time in the past, before some files in CVS).  
First CVS is checked out, then gcc_update corrects the timestamps on that
checkout, then the tree is built which puts the files in the source
directory for the tarball.  Timestamp skew could occur for users using
diffs, but that's inevitable without support in diff for a topological
sort on the diffs of individual files.

In any case, this would not be a new issue; only the mechanism by which
the files get in the source directory has changed, not the principle that
they are there in release tarballs and need the correct timestamp
sequence.

-- 
Joseph S. Myers
jsm@polyomino.org.uk