Re: Info-files missing in gcc 3.2.2

[Joe Buck <jbuck at synopsys dot com>]

On Thu, 6 Feb 2003, Joseph S. Myers wrote:
> > You should never create a new tarball with the same name - this will just
> > create problems for people using diffs.

On Thu, Feb 06, 2003 at 07:28:57PM +0100, Christian Cornelssen wrote:
> In this case, existing files would not change, just new ones would get
> added. "cvs rdiff" would not be affected. If you mean diffs between
> actual release tar contents, then this would matter for diff users
> without makeinfo.

There's another problem: we can't re-use the same name because
of the widespread confusion it would cause.  You'll even get people
worrying about Trojans when MD5 sums on tarballs don't match.
Given widespread worry about such issues, we simply cannot quietly
replace one set of tarballs with another set that has the same names,
ever.

> Currently, both complete tar users and diff users would not get info
> files other than by makeinfo, thus updating the tarball would help at
> least the tar users. For the rest, I think that confusion about
> additional release suffixes would be greater than the casual
> inconvenience of having to install makeinfo. Don't you think that
> typical users will download gcc-3.2.2.tar.*z*, regardless of whether
> versions with additional suffixes exist?

If we withdraw the tarball and produce another, we have to rename it.
There's precedent for that; in such cases names like

gcc-3.2.2a.tar.bz2

would be used (this has been done when a packaging error occurred that
did not affect the compiler source).

However, I don't think that this problem is severe enough to justify
such things.  I would suggest adding a new tarball containing only
the info files, for those who care; distros packaging gcc will be
re-arranging things and can run makeinfo themselves.