So your proposal is to always use one set of units, namely 1000ths of a mm (i.e., a micrometer); never use inches for any media sizes.  It certainly is simple. 
1000th of a mm is precise enough so that the English sizes can be represented in 1000ths of a mm without round off error (which would create differences in names, if some rounded and others truncated).
We used the same strategy of using only a single unit system in the IPP Production Printing Attributes - Set1 extension, instead of having both metric and English.  The only minor difference was that we used 100th of a mm, instead of 1000th of a mm for use in the "media-size" member attribute of the "media-col" attribute.  We felt that was precise enough.  See section 3.13.8 in:
ftp://ftp.pwg.org/pub/pwg/standards/pwg5100.3.pdf, .doc, .rtf
The 1000th of a mm is one of the two units used in the Printer MIB (the other being 10000th of an inch).
We did agree at the meeting that the client shouldn't simply display the Media Size name to the user if it doesn't have it in its localization data base (thought there will always be simple minded clients that will).  The client should do some parsing and possible converting of units to the one that the user prefers.  So there is no real advantage to the client to have the units be in inches for English sizes and metric for metric sizes (except for the really simple clients that never convert units).
What do others think of just always using micrometers for the size dimensions for our Media Size Self Describing Names?
-----Original Message-----
From: Norbert Schade [mailto:norbertschade@oaktech.com]
Sent: Thursday, April 26, 2001 13:50
To: IPP Group
Subject: IPP> global media; comment on yesterday's proposal

I have my problems with the new proposal.
I am going to rephrase my previous statements commenting that proposal.
Splitting the media size name into three components (unit, name, dimensions) is a very new idea.
My main problem with this proposal is the first component.
In Ron's current version of the spec we have two units: inch/1000 and mm/10. We have implemented that version to learn about problems.
With the new proposal there is the danger to have an even bigger number of units.
Supporting more than one unit is a serious problem for any driver. Ask any driver developer. It's not about what unit I want to show in the UI. It's about necessary conversions when dealing with applications. Please study Mark's feedback from 4/20. I repeat it in easy words (I hope).
Scenario excerpt
1. Workstation 1 with driver 1
Driver 1 is supporting a media size 'Letter.2159-2794' (the developer of the driver has chosen the metric way). You could do the sample with any other size.
2. User 1 sitting at workstation 1 writes one page of text with application xyz and saves the document.
this means that the size information 'Letter.2159-2794' is saved in the document file as well in many, many applications.
3. User 1 sends his document to user 2.
4. User 2 sitting at workstation 2 with driver 2 (different from driver 1) opens the document. This second the driver 2 is already involved.
5. Imagine driver 2 is not supporting 'Letter.2159-2794', but 'na-Letter.8500-1100' instead, which in fact is the same size with a different name.
Now it's the question what is driver 2 doing.
It could start some investigation to match or emulate the required size. -> bad performance.
The same situation will happen again when printing. It will happen very often, repeatedly.
So we already have a problem with two units. If we now open a gate to be able to define even more units, it will be aweful code and a terrible performance. Every driver developer should be able to prove that.
However everything would be fine, if there'd be one and one only unit.
Make it small enough that any rounding for a UI string or whatever is needed, can be done properly. Our proposal within UPDF was mm/1000, which is certainly small enough (and used in the industry anyhow).
I treated strings like 'jis' or 'iso' just as parts of the name to make it more apparent. 'na' was the only exception so far.
If all names are unique (I think they are in Ron's current concept), I don't have a problem splitting the name and the dimensions into two components. In that case we may even work with the name only and handle the dimensions with an include file.
I thought the idea of combining the name and the dimensions is ok, as we need it for custom size anyhow.
BTW: I am happy to have proper keywords, but my drivers certainly will never, never, never show them in the UI. Be also sure that in UPDF we are providing the chance to assign a proper human readable UI string to it.
So from a driver's point of view the easiest case is to work with 1 unit (mm/1000), remove the prefix 'na-' and convert all the dimensions.
This ensures a good performance, consistent routines and readability.
Whatever the internal unit of a driver is, it most probably has all converting routines available to work with 1 unit, but not all necessary functions to match between different units.
I would be very surprised if Mark does not feel very, very similarly, although I have been told differently today. Unfortunately I couldn't get hold of him on his trip today.
I call this proposal '1unit mm/1000, unchanged naming', where unchanged naming means no separate components, but converted dimensions into that 1 unit. I do not insist on unchanged naming, but I haven't seen the big advantage of it so far.
Norbert Schade
Principle Software Engineer
Host Software Group
Oak Technology, Inc.
10 Presidential Way
Woburn, MA 01801
Phone: 1-781-638-7614
Fax: 1-781-638-7555
email: norbertschade@oaktech.com