Robin Fairbairns wrote:

>> I'm trying to think of
>> how to balance ease of use with availability of older versions.  I think
>> that this was not possible when LaTeX2e was first released, nor was it
>> obvious it would be required.
> 
> actually, except in cases where the original has proved buggy (or the
> like), most of the latex 2.09 stuff is still around.  (and there was a
> "big day" when the link macros/latex was switched from latex209 to
> latex2e...)

Perhaps I wasn't clear here.  I was thinking about when packages move
version (however small), not about major changes (2.09 -> 2e -> 3).  If
I want/need to get hold of "not the current CTAN release" of most
packages, I currently have to mail the author and hope a copy is
available (unless it's one of a small number of larger projects that has
a code repository I can connect to). This discussion started with Will
saying:

> Indeed. I was thinking of something a little more organised, which
> would obviously involve CTAN & TeXLive keeping old versions in their
> repository as well. (Which I think they've discussed in the past.)

I was simply trying to think through how that might work without (1)
putting off package authors or (2) asking too much of the CTAN team.

> so yes, the switchover _was_ thought about (carefully), and reasonable
> action taken.

I have no doubt at all that you thought very carefully about this!  The
system works well, and I'm certainly not keen on change for changes sake.

> first, we can't (as at the 2.09->2e switchover) "just use" unconverted
> packages.
> 
> second, there were myriads of 2.09 packages designed to sort out
> problems with latex as it stood.  an example is the many different ways
> you could configure fonts -- at my work in the 80s, i had built
> different latex formats to use different document fonts.
> 
> so back then we expected there to be far fewer 2e packages, but the
> freedom 2e has given us (that, with the safe assumption that a "big"tex
> can always be used) has led to massive extension of usable function from
> packages (though there may not actually be all that many more -- i've
> not counted).
> 
> (also, cpus are faster.  i remember being excited that we were about to
> get a 1 mips machine, which sounds ridiculous now.  no-one would dream
> of something like siunitx, back then!)

Um yes, I do rely on quite a fast system in siunitx: I've had complaints
about it being too slow.

> i would not be surprised to find that (after a period of hiatus) the set
> of packages (converted and otherwise) started to grow again,
> enourmously.

You are also quite right that in the same way a move to LaTeX3 is an
unknown until it actually happens, and there will be lots of things
people want to do. I suppose I'm looking mainly at the likelihood that a
lot of "utility" packages a lot of people use (obvious examples float,
caption, geometry, booktabs) are likely to be superseded by the new
kernel (I would guess).  That doesn't mean there won't be other things
going on, which as you say will probably mean lots of new packages none
of us have anticipated (or maybe they have, and I've missed it!).

> so what would be a sensible way forward?  personally, for some time, i
> would leave distribution as it is.  

I hope I didn't imply that anything should happen *now*.  I was looking
to some future date where LaTeX3 is released, or at least enough of the
current code is stable and LaTeX3 packages are not placed in
/macros/latex/contrib (which if they don't work under 2e would be bad, I
think).

> so, personally, i would like to see deprecated stuff doing no more than
> move to ctan:obsolete/ (and possibly being superseded there as the newer
> version passes its sell-by date).
> 
> in other words, no change.

Certainly not at the moment.  The current expl3 code looks quite good to
me, and I think Will is finding and sorting inconsistencies in a very
sensible manner.  However, it certainly isn't marked "stable" yet (i.e.
changes to the existing interface only as necessary and not because it
looks cool, but still allowing new features to be added.  A bit like
siunitx!)

> and of course i am thoroughly in agreement with karl's position.

I get the point here.  Please ignore the distribution part of my
suggestion entirely :-)  If people need an old version of something,
they should have some idea what they are up to.
-- 
Joseph Wright