Thanks for these clarifications.
Do we have a way to represent “unpacked”
decimal numbers. This is like zoned, except the “zones” are zero
instead of “F” (in ebcdic encodings).
Also, can a BCD number have a sign?
…mikeb
From: Steve Hanson
[mailto:smh@uk.ibm.com]
Sent: Wednesday, April 09, 2008
10:00 AM
To: mbeckerle.dfdl@gmail.com
Cc: 'Mike Beckerle'; Alan Powell;
Ian W Parkinson
Subject: RE: Fw: DFDL Decimal -
proposal - correcting wrong attachment
Hi Mike - answers in-line below.
Regards, Steve
Steve Hanson
Programming Model Architect
WebSphere Message Brokers
Hursley, UK
Internet: smh@uk.ibm.com
Phone (+44)/(0) 1962-815848
|
"Mike Beckerle"
<mbeckerle.dfdl@gmail.com> 09/04/2008 01:43
|
|
I prefer one property dfdl:numberFormat, the
valid values of which depend on dfdl:representation
<smh>The
advantage of two properties is that you can set scoping for text and binary
numbers separately.
I like the analysis that text formats are ones which depend
on encoding, and not byteOrder, and binary depend on byte order, and NOT
encoding.
<smh>Me
too.
There’s also format specifiers for floating point.
Should those also go on here, be allowed only for
representation=”binary”?
<smh>I
did think about this, but I think we are better off keeping floats separate.
Otherwise people might think you can declare a logical float to be rep'd by
physical integer. MRM allows this, and I wish it didn't. It also exacerbates
the problem noted above - I couldn't set a default float format, which is
something that would almost certainly never vary within a data stream.
The rest of the proposal looks fine. I found
decimalVirtualPoint an odd name, but it is clear and obeys the conventions.
<smh>I
agree it's a bit odd. An alternative is 'decimalimpliedPlaces' which uses TX
terminology - but that doesn't match the 'V' pattern character we are proposing
in the ICU pattern (which matches COBOL)
I was a bit unclear on how do you represent an unsigned
packed decimal. This is common. There is no sign nibble at all. It lets you do
an even number of digits. MMDDYY is commonly this, 3 unsigned packed numbers.
<smh>What
you have described is dfdl:numberFormat="BCD". An unsigned packed
decimal is dfdl:numberFormat="packed" with the sign nibble always
unsigned, so dfdl:packedDecimalSignCodes="F F F".
…mikeb
From: Steve Hanson
[mailto:smh@uk.ibm.com]
Sent: Wednesday, April 02, 2008 11:54 AM
To: Alan Powell
Cc: Ian W Parkinson; Mike Beckerle
Subject: Re: Fw: DFDL Decimal - proposal - correcting wrong
attachment
Alan, Ian and myself reviewed this today.
The main issue was that the loss of dfdl:representation="decimal"
means that it is no longer clear when to use dfdl:integerFormat and
dfdl:decimalFormat, because an xs:decimal can have a binary integer rep and an
xs:int can have a binary decimal rep. It was noted that both IBM models (MRM
and TX type tree) handle this by having a single property. I don't want to
re-introduce rep=decimal, I think we shoiuld stick with text (implying encoding
sensitive) and binary (potentially byte order sensitive). Options:
a) One property dfdl:numberFormat with values "text",
"zoned", "packed", "BCD",
"twosComplement", "onesComplement", "signMagnitude".
- "text" and "zoned" when
dfdl:representation="text"
- "packed", "BCD", "twosComplement",
"onesComplement", "signMagnitude" when
dfdl:representation="binary"
Number xs:int, xs:decimal
text =>
numberFormat
xs:float, xs:double
text =>
xs:int, xs:decimal
binary =>
numberFormat
xs:float
binary => floatFormat
b) Two properties dfdl:textNumberFormat and dfdl:binaryNumberFormat, allowable
enums split as above.
- this means the existing dfdl:textNumberFormat property gets renamed to
dfdl:textNumberPattern or dfdl:textNumberScheme
Number xs:int, xs:decimal
text => textNumberFormat
xs:float, xs:double
text =>
xs:int, xs:decimal
binary =>
binaryNumberFormat
xs:float
binary => floatFormat
Other suggestions?
Regards, Steve
Steve Hanson
WebSphere Message Brokers
Hursley, UK
Internet: smh@uk.ibm.com
Phone (+44)/(0) 1962-815848
|
Alan Powell/UK/IBM
28/03/2008 16:45 |
|
Steve
Technically seems OK.
Need quite a bit of editorial work before it can be included in the spec which
I have started.
Alan Powell
MP 211, IBM UK Labs, Hursley, Winchester, SO21 2JN, England
Notes Id: Alan Powell/UK/IBM email: alan_powell@uk.ibm.com
Tel: +44 (0)1962 815073
Fax: +44 (0)1962 816898
|
From:
|
Steve Hanson/UK/IBM |
|
To:
|
mbeckerle@oco-inc.com |
|
Cc:
|
Alan Powell/UK/IBM, Ian W Parkinson/UK/IBM
|
|
Date:
|
28/03/2008 13:59 |
|
Subject:
|
Fw: DFDL Decimal - proposal - correcting wrong
attachment |
Here's an attempt at a revised decimal supplement, that takes into account the
stuff in my mail below.
[attachment "ggf-dfdl-supplement-advanced-decimal-properties-v1.0-003.doc"
deleted by Alan Powell/UK/IBM]
Some discussion points:
1) I've removed the representation 'Decimal' - a decimal is either 'Text' or
'Binary'. Property decimalFormat says whether it is text or zoned (for
text) or packed or BCD (for binary).
2) There's no need for a decimalSigned property, as zoned uses numberPattern
for this, BCD is always unsigned, and packed indicates this via sign code
3) I've added VDP property for BCD and packed - zoned uses numberPattern for
this. However, VDP property is also needed for binary integers - this is
missing from spec. COBOL PIC 99V99 COMP will create an xs:decimal with binary
integer rep, so we need to support this. I suggest we have a single VDP property
that applies to all binary reps that can be used to represent xs:decimal. So my
VDP property gets removed to main spec.
4) The resultant properties are less than before. I'm not sure that a separate
supplement is justified.
5) I would like to remove numberCheckPolicy from dfdl:DefineNumberFormat, and
make it a separate property. Two reasons:
- I think the decision to use strict/lax checking is not an attribute of the
number format but more an attribute of the schema as a whole.
- It means we can control packed decimal sign nibble oddities with the same
property as other strict/lax number checking,
Let's review on next OGF WG call.
Regards, Steve
Steve Hanson
Programming Model Architect
WebSphere Message Brokers
Hursley, UK
Internet: smh@uk.ibm.com
Phone (+44)/(0) 1962-815848
----- Forwarded by Steve Hanson/UK/IBM on 28/03/2008 12:33 -----
|
Steve Hanson/UK/IBM 27/03/2008 15:29 |
|
Hi Mike
I've finally got round to looking at the decimal supplement, and I'd like to
get your opinion on something. The WTX team have been reviewing draft 031 and
had the following observation (actually they had quite a few good ones, and
when they've finished we need to discuss them all on a OGF WG call).
"13.3. Is a zoned decimal textual or non-textual? If all overpunched
variants result in well-known characters then the data is scannable and
therefore more like a textual field."
It turns out that the type hierarchy in TX for decimal looks like below. They
consider Zoned as text as it always consists of reasonable characters and is
subject to encoding conversion, padding, justification, etc. There's a lot of
appeal in that. It's always bothered me a bit that MRM viewed it as a binary
type.
Number -> Character -> Decimal (meaning text decimal)
Integer (meaning text integer)
Zoned
-> Binary -> Integer (meaning binary
integer)
Float
Packed
BCD
Also, their Zoned does not have separate sign option. They point out that a
separate signed Zoned is just a Text decimal. And they are correct. We got the
separate sign thing from MRM, which after some digging turns out to have got it
from the CAM Type Descriptor model, which had no other way of representing a
text decimal number with a separate sign.
As part of my rework of the decimal supplement, I'd like to take both these
into account. The implications are:
- Zoned => overpunched only
- Zoned decimal can pick up on the textNumberxxx properties, including
textNumberFormat
=> use the numberPattern (ie, ICU pattern) property to
say which end the (overpunched) sign goes
=> can get away without a separate pattern language for
binary decimals, which as you point out has endian-ness issues
- Binary decimals are packed and BCD
- There are a lot fewer properties for decimals
- dfdl:representation = "text" can have subdivisions - that's not
occurred until now (we could think about making dfdl:representation =
"xml" a subdivision of "text"?)
If you think there is merit in this approach then let me know by return and
I'll see if I can write something up tomorrow.
I'm WAH on +44-1794-340899 if you want to discuss.
Your "crazy idea" below is interesting - but I think is a tooling
thought rather than a core spec thing.
(Sorry about call yesterday - I thought I mailed something out a couple of
calls ago about DST mismatch, but perhaps I didn't).
Regards, Steve
Steve Hanson
Programming Model Architect
WebSphere Message Brokers
Hursley, UK
Internet: smh@uk.ibm.com
Phone (+44)/(0) 1962-815848
----- Forwarded by Steve Hanson/UK/IBM on 27/03/2008 15:04 -----
|
Mike Beckerle/Worcester/IBM@IBMUS
21/11/2007 15:26 |
|
I think decimal has signed and unsigned variants based on dfdl:decimalSigned boolean.
If this is false then it's unsigned and packedUnsignedRep specifies the sign
nibble used for unsigned. The doc doesn't specify that one can say ""
for this indicating no sign nibble at all.
I've been rereading the decimal properties supplement and starting v002 of it
based on changes to dfdl:representation in the core spec. This needs a general
clean up. There's errors here in that there is a decimalType="zoned",
or "packed" or "BCD" and also a bcdIsPacked, and bcdUnpackedRep="ebcdic",
which is the same as zoned I think.
We need there to be one way to express these things. Right now the bias is a
set of orthogonal flags: signed or unsigned, what's the sign nibble for
unsigned, what sign nibbles for signed, packed or unpacked, what's in the zones
- the unused nibbles - (ebcdic, i.e., "F", ascii, i.e.,
"3", or zero - but that's not enough as I've seen data with
"2" in the zones - some non IBM cobol compiler does this.).
A better choice may be to specify decimalType as a larger enum which includes
most of these properties, so that we don't end up with too much ability to
express variants that have simply never existed.
A list of the use cases needs to be added to the doc also.
Here's a few:
-1234 as expressed as bytes in hex in increasing position order, i.e., LSB
first.
packed ibm, signed, D01234
zoned ibm, overpunched leading sign D1F2F3F4 (are signs usually leading or
trailing.... I think trailing actually.)
big endian zoned ascii, ascii-translated overpunched leading sign 4A323334
(yuck - so much for treating decimal as "binary" data).
Here's a crazy idea: I believe there is a set of magic numbers which if you
give me their translations in bytes, I can determine exactly what the encoding
properties are.
E.g., if you give me the bytes for +0000, -1234, +789 I believe I can
determine all of the properties.
This might be a better way to specify decimal formats. I.e., give me those byte
patterns expressed as hex, and I reverse engineer all the property settings.
e.g., decimalFormat="+0000=C00000-1234=D01234 +789=C789" (signed,
packed, leading sign, padded to even number of nibbles, big endian, zero
carries a sign, "C" is plus, "D" is minus)
or decimalFormat="+0000=00000000 -1234=D1F2F3F4 +789=C7F8F9" (ebcdic
zoned, leading overpunched sign, big endian, zero is allowed to have zero as
sign and all zero bytes, "C" is plus, "D" is minus)
This may make more sense for the tooling than the DFDL language though. I.e.,
point it at some data and it tries to guess these properties.
Mike Beckerle
STSM, Architect, Scalable Computing
IBM Software Group
Information Platform and Solutions
Westborough, MA 01581
direct: voice and FAX 508-599-7148
assistant: Pam Riordan
priordan@us.ibm.com
508-599-7046
Unless stated otherwise above:
IBM United Kingdom Limited - Registered in England and Wales with number
741598.
Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU
Unless stated otherwise above:
IBM United Kingdom Limited - Registered in England and Wales with number
741598.
Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU