You are viewing a plain text version of this content. The canonical link for it is here.
Posted to c-dev@axis.apache.org by Samisa Abeysinghe <sa...@gmail.com> on 2004/12/13 10:10:08 UTC
Number of decimal places in float (and decimal, double)
Hi Folks,
As per http://nagoya.apache.org/jira/browse/AXISCPP-320, are we
happy with a 4 decimal place acuracy, or do we need more decimal
places be represented correctly?
Thanks,
Samisa...
Re: Number of decimal places in float (and decimal, double)
Posted by Samisa Abeysinghe <sa...@gmail.com>.
Hi John,
Thaks for the fantastic effort to build this matrix. This is very
helpful indeed.
+1 for publishing this.
IMHO, we have to address all the basic types and document the
'error' in cases we are not adhering to the spec 100%.
Also it is good to have a record of standard library functions that
we use in serialization and deserialization so that we can get to knwo
how precise our representations are in relation to those lib
implementations.
PS. Due to some formatting problems in my mail client, the matrix
appear distorted. Would it be possible to post and HTML version of
this?
Thanks,
Samisa...
On Tue, 14 Dec 2004 11:29:13 +0000, John Hawkins <HA...@uk.ibm.com> wrote:
>
>
> Hi Folks,
>
> I've created a top of the head comparison list of types between xsd
> definition, java representation and what we convert them to. There are
> quite a few issues here and quite a few types that I have not dealt with.
> This relates closely to http://nagoya.apache.org/jira/browse/AXISCPP-255.
>
> You can see that we are not really adhering to the spec in some/many cases.
> The question is - what do we wish to do? I'm not sure I have a strong
> opinion here (for once ;-) Certainly some are more important than others
> so, if we decide to purge these issues, we could prioritise those.
>
> At the very least I suggest we complete this list and publish it.
>
> |--------------+----------------------------+-----------------+------------+----------------------------------+---------------------+-------------|
> | XSD Type |XML Constraints | Java | Deserialise| Axis C serialise as. |Comments | Deserialise |
> | | | interpretation | as | | | as |
> |--------------+----------------------------+-----------------+------------+----------------------------------+---------------------+-------------|
> | String | N/A | String |char* | | | char* |
> |--------------+----------------------------+-----------------+------------+----------------------------------+---------------------+-------------|
> | integer | derived from decimal | math.BigInteger |long |"%ld", *((long*)(pValue)) | | long |
> |--------------+----------------------------+-----------------+------------+----------------------------------+---------------------+-------------|
> | int | derived from long | Int |Int |"%d", *((int*)(pValue)) | | int[] ? |
> | | maxInc=2147483647 | | | | | |
> | | minInc=-2147483648 | | | | | |
> |--------------+----------------------------+-----------------+------------+----------------------------------+---------------------+-------------|
> | long | derived from integer | long |Win32: |WIN32: | | Win32: |
> | | maxInc=9223372036854775807 | |__int64 |"%lld", *((__int64*)(pValue)) | | __int64 |
> | | minInc=-9223372036854775808| |#else |#else | | #else |
> | | | |long long |"%lld", *((long long*)(pValue)) | | long long |
> |--------------+----------------------------+-----------------+------------+----------------------------------+---------------------+-------------|
> | short | derived from int | short |short |"%d", *((short*)(pValue)) | | short |
> | | maxInc=32767 | | | | | |
> | | minInc=-32768 | | | | | |
> |--------------+----------------------------+-----------------+------------+----------------------------------+---------------------+-------------|
> | decimal |must support 18 decimal | bigDecimal |double |"%f", *((double*)(pValue)) |'double' has a | double |
> | |digits | | | |precision of 15 | |
> | | | | | |digits, not 18. | |
> |--------------+----------------------------+-----------------+------------+----------------------------------+---------------------+-------------|
> | float | IEEE 32 bit standard-> mx | float |float |"%f", *((float*)(pValue))) |Change to '%E' so | float |
> | | 2^e where | | | |that the number | |
> | |m<2^24 | | | |always appears as | |
> | |-149<e<104 | | | |[+/-]#.#--#E[-]###. | |
> | | | | | |Also type float does | |
> | | | | | |not have sufficient | |
> | | | | | |precision | |
> | | | | | |(3.402823466E38 > n >| |
> | | | | | |1.175494351E-38). | |
> |--------------+----------------------------+-----------------+------------+----------------------------------+---------------------+-------------|
> | double | IEEE 64 bit standard-> | double |double |"%f", *((double*)(pValue)) | | double |
> | | mx2^e where | | | | | |
> | | m<2^53 | | | | | |
> | | -1075<e<970 | | | | | |
> |--------------+----------------------------+-----------------+------------+----------------------------------+---------------------+-------------|
> | boolean | | boolean |Int |{true, false} | | ? |
> |--------------+----------------------------+-----------------+------------+----------------------------------+---------------------+-------------|
> | byte | derived from short | byte |char |"%d", *((char*)(pValue))) | | char |
> | | maxInc=127 | | | | | |
> | | minInc=-128 | | | | | |
> |--------------+----------------------------+-----------------+------------+----------------------------------+---------------------+-------------|
> | dateTime | | Calendar | | | Only 1st Jan 1970 | |
> | | | | | | onwards plus other | |
> | | | | | | issues. | |
> |--------------+----------------------------+-----------------+------------+----------------------------------+---------------------+-------------|
> | duration | ISO 8601 | | |m_AxisTime.serialize( | | |
> | | | | | *((long*)(pValue)), | | |
> | | | | |type).c_str ()) | | |
> |--------------+----------------------------+-----------------+------------+----------------------------------+---------------------+-------------|
> | base64Binary | | byte[] |char* | | | char* |
> |--------------+----------------------------+-----------------+------------+----------------------------------+---------------------+-------------|
> | hexBinary | | byte[] |char* | | | char* |
> |--------------+----------------------------+-----------------+------------+----------------------------------+---------------------+-------------|
> | | | | | | | |
> |--------------+----------------------------+-----------------+------------+----------------------------------+---------------------+-------------|
>
>
> John Hawkins
>
> Samisa Abeysinghe
> <samisa.abeysingh
> e@gmail.com> To
> Apache AXIS C Developers List
> 14/12/2004 04:30 <ax...@ws.apache.org>
> cc
>
> Please respond to Subject
> "Apache AXIS C Re: Number of decimal places in
> Developers List" float (and decimal, double)
>
>
> OK, I found a solution to go beyond the 4 decimal place limit.
>
> However, there still would be some element of error involved (but very
> small, about
> 0.0000000000001) I think this is acceptable.
>
> Samisa...
>
> On Tue, 14 Dec 2004 01:32:25 +0000, Samisa Abeysinghe
> <sa...@gmail.com> wrote:
> > Sanjiva, the standard functions (e.g. sprintf, strtod, etc.) that we
> > are using in serialization/deserialization does not seem to retain
> > precision the way we want. There are some work-arounds but does not
> > work exctly the way we want them to.
> >
> > John, yes I will have a look at XSD spec. I think that is the best
> > thing to do in order to find out what we sould be supporting.
> >
> > Thanks,
> > Samisa...
> >
> >
> >
> >
> > On Mon, 13 Dec 2004 21:57:36 +0600, Sanjiva Weerawarana
> > <sa...@opensource.lk> wrote:
> > > Yes I agree.
> > >
> > > Samisa, why wouldn't we retain all the precision that's available?
> > > Maybe I don't understand the issue (entirely possible!).
> > >
> > > Sanjiva.
> > >
> > >
> > >
> > > ----- Original Message -----
> > > From: "John Hawkins" <HA...@uk.ibm.com>
> > > To: "Apache AXIS C Developers List" <ax...@ws.apache.org>
> > > Sent: Monday, December 13, 2004 7:01 PM
> > > Subject: Re: Number of decimal places in float (and decimal, double)
> > >
> > > >
> > > >
> > > >
> > > >
> > > > This is a huge area that I can't see we've discussed before?
> > > >
> > > > To my logic we should be looking at the attributes on the types for
> their
> > > > description of what they are expecting shouldn't we ?
> > > >
> > > >
> > > > John Hawkins
> > > >
> > > >
> > > >
> > > >
> > > >
> > > > Samisa Abeysinghe
> > > > <samisa.abeysingh
> > > > e@gmail.com>
> To
> > > > Apache AXIS C Developers List
> > > > 13/12/2004 09:10 <ax...@ws.apache.org>
> > > >
> cc
> > > >
> > > > Please respond to
> Subject
> > > > "Apache AXIS C Number of decimal places in
> float
> > > > Developers List" (and decimal, double)
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > > Hi Folks,
> > > > As per http://nagoya.apache.org/jira/browse/AXISCPP-320, are we
> > > > happy with a 4 decimal place acuracy, or do we need more decimal
> > > > places be represented correctly?
> > > >
> > > > Thanks,
> > > > Samisa...
> > > >
> > > >
> > > >
> > >
> > >
> >
>
>
Re: Number of decimal places in float (and decimal, double)
Posted by John Hawkins <HA...@uk.ibm.com>.
Hi Folks,
I've created a top of the head comparison list of types between xsd
definition, java representation and what we convert them to. There are
quite a few issues here and quite a few types that I have not dealt with.
This relates closely to http://nagoya.apache.org/jira/browse/AXISCPP-255.
You can see that we are not really adhering to the spec in some/many cases.
The question is - what do we wish to do? I'm not sure I have a strong
opinion here (for once ;-) Certainly some are more important than others
so, if we decide to purge these issues, we could prioritise those.
At the very least I suggest we complete this list and publish it.
|--------------+----------------------------+-----------------+------------+----------------------------------+---------------------+-------------|
| XSD Type |XML Constraints | Java | Deserialise| Axis C serialise as. |Comments | Deserialise |
| | | interpretation | as | | | as |
|--------------+----------------------------+-----------------+------------+----------------------------------+---------------------+-------------|
| String | N/A | String |char* | | | char* |
|--------------+----------------------------+-----------------+------------+----------------------------------+---------------------+-------------|
| integer | derived from decimal | math.BigInteger |long |"%ld", *((long*)(pValue)) | | long |
|--------------+----------------------------+-----------------+------------+----------------------------------+---------------------+-------------|
| int | derived from long | Int |Int |"%d", *((int*)(pValue)) | | int[] ? |
| | maxInc=2147483647 | | | | | |
| | minInc=-2147483648 | | | | | |
|--------------+----------------------------+-----------------+------------+----------------------------------+---------------------+-------------|
| long | derived from integer | long |Win32: |WIN32: | | Win32: |
| | maxInc=9223372036854775807 | |__int64 |"%lld", *((__int64*)(pValue)) | | __int64 |
| | minInc=-9223372036854775808| |#else |#else | | #else |
| | | |long long |"%lld", *((long long*)(pValue)) | | long long |
|--------------+----------------------------+-----------------+------------+----------------------------------+---------------------+-------------|
| short | derived from int | short |short |"%d", *((short*)(pValue)) | | short |
| | maxInc=32767 | | | | | |
| | minInc=-32768 | | | | | |
|--------------+----------------------------+-----------------+------------+----------------------------------+---------------------+-------------|
| decimal |must support 18 decimal | bigDecimal |double |"%f", *((double*)(pValue)) |'double' has a | double |
| |digits | | | |precision of 15 | |
| | | | | |digits, not 18. | |
|--------------+----------------------------+-----------------+------------+----------------------------------+---------------------+-------------|
| float | IEEE 32 bit standard-> mx | float |float |"%f", *((float*)(pValue))) |Change to '%E' so | float |
| | 2^e where | | | |that the number | |
| |m<2^24 | | | |always appears as | |
| |-149<e<104 | | | |[+/-]#.#--#E[-]###. | |
| | | | | |Also type float does | |
| | | | | |not have sufficient | |
| | | | | |precision | |
| | | | | |(3.402823466E38 > n >| |
| | | | | |1.175494351E-38). | |
|--------------+----------------------------+-----------------+------------+----------------------------------+---------------------+-------------|
| double | IEEE 64 bit standard-> | double |double |"%f", *((double*)(pValue)) | | double |
| | mx2^e where | | | | | |
| | m<2^53 | | | | | |
| | -1075<e<970 | | | | | |
|--------------+----------------------------+-----------------+------------+----------------------------------+---------------------+-------------|
| boolean | | boolean |Int |{true, false} | | ? |
|--------------+----------------------------+-----------------+------------+----------------------------------+---------------------+-------------|
| byte | derived from short | byte |char |"%d", *((char*)(pValue))) | | char |
| | maxInc=127 | | | | | |
| | minInc=-128 | | | | | |
|--------------+----------------------------+-----------------+------------+----------------------------------+---------------------+-------------|
| dateTime | | Calendar | | | Only 1st Jan 1970 | |
| | | | | | onwards plus other | |
| | | | | | issues. | |
|--------------+----------------------------+-----------------+------------+----------------------------------+---------------------+-------------|
| duration | ISO 8601 | | |m_AxisTime.serialize( | | |
| | | | | *((long*)(pValue)), | | |
| | | | |type).c_str ()) | | |
|--------------+----------------------------+-----------------+------------+----------------------------------+---------------------+-------------|
| base64Binary | | byte[] |char* | | | char* |
|--------------+----------------------------+-----------------+------------+----------------------------------+---------------------+-------------|
| hexBinary | | byte[] |char* | | | char* |
|--------------+----------------------------+-----------------+------------+----------------------------------+---------------------+-------------|
| | | | | | | |
|--------------+----------------------------+-----------------+------------+----------------------------------+---------------------+-------------|
John Hawkins
Samisa Abeysinghe
<samisa.abeysingh
e@gmail.com> To
Apache AXIS C Developers List
14/12/2004 04:30 <ax...@ws.apache.org>
cc
Please respond to Subject
"Apache AXIS C Re: Number of decimal places in
Developers List" float (and decimal, double)
OK, I found a solution to go beyond the 4 decimal place limit.
However, there still would be some element of error involved (but very
small, about
0.0000000000001) I think this is acceptable.
Samisa...
On Tue, 14 Dec 2004 01:32:25 +0000, Samisa Abeysinghe
<sa...@gmail.com> wrote:
> Sanjiva, the standard functions (e.g. sprintf, strtod, etc.) that we
> are using in serialization/deserialization does not seem to retain
> precision the way we want. There are some work-arounds but does not
> work exctly the way we want them to.
>
> John, yes I will have a look at XSD spec. I think that is the best
> thing to do in order to find out what we sould be supporting.
>
> Thanks,
> Samisa...
>
>
>
>
> On Mon, 13 Dec 2004 21:57:36 +0600, Sanjiva Weerawarana
> <sa...@opensource.lk> wrote:
> > Yes I agree.
> >
> > Samisa, why wouldn't we retain all the precision that's available?
> > Maybe I don't understand the issue (entirely possible!).
> >
> > Sanjiva.
> >
> >
> >
> > ----- Original Message -----
> > From: "John Hawkins" <HA...@uk.ibm.com>
> > To: "Apache AXIS C Developers List" <ax...@ws.apache.org>
> > Sent: Monday, December 13, 2004 7:01 PM
> > Subject: Re: Number of decimal places in float (and decimal, double)
> >
> > >
> > >
> > >
> > >
> > > This is a huge area that I can't see we've discussed before?
> > >
> > > To my logic we should be looking at the attributes on the types for
their
> > > description of what they are expecting shouldn't we ?
> > >
> > >
> > > John Hawkins
> > >
> > >
> > >
> > >
> > >
> > > Samisa Abeysinghe
> > > <samisa.abeysingh
> > > e@gmail.com>
To
> > > Apache AXIS C Developers List
> > > 13/12/2004 09:10 <ax...@ws.apache.org>
> > >
cc
> > >
> > > Please respond to
Subject
> > > "Apache AXIS C Number of decimal places in
float
> > > Developers List" (and decimal, double)
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > > Hi Folks,
> > > As per http://nagoya.apache.org/jira/browse/AXISCPP-320, are we
> > > happy with a 4 decimal place acuracy, or do we need more decimal
> > > places be represented correctly?
> > >
> > > Thanks,
> > > Samisa...
> > >
> > >
> > >
> >
> >
>
Re: Number of decimal places in float (and decimal, double)
Posted by Samisa Abeysinghe <sa...@gmail.com>.
OK, I found a solution to go beyond the 4 decimal place limit.
However, there still would be some element of error involved (but very
small, about
0.0000000000001) I think this is acceptable.
Samisa...
On Tue, 14 Dec 2004 01:32:25 +0000, Samisa Abeysinghe
<sa...@gmail.com> wrote:
> Sanjiva, the standard functions (e.g. sprintf, strtod, etc.) that we
> are using in serialization/deserialization does not seem to retain
> precision the way we want. There are some work-arounds but does not
> work exctly the way we want them to.
>
> John, yes I will have a look at XSD spec. I think that is the best
> thing to do in order to find out what we sould be supporting.
>
> Thanks,
> Samisa...
>
>
>
>
> On Mon, 13 Dec 2004 21:57:36 +0600, Sanjiva Weerawarana
> <sa...@opensource.lk> wrote:
> > Yes I agree.
> >
> > Samisa, why wouldn't we retain all the precision that's available?
> > Maybe I don't understand the issue (entirely possible!).
> >
> > Sanjiva.
> >
> >
> >
> > ----- Original Message -----
> > From: "John Hawkins" <HA...@uk.ibm.com>
> > To: "Apache AXIS C Developers List" <ax...@ws.apache.org>
> > Sent: Monday, December 13, 2004 7:01 PM
> > Subject: Re: Number of decimal places in float (and decimal, double)
> >
> > >
> > >
> > >
> > >
> > > This is a huge area that I can't see we've discussed before?
> > >
> > > To my logic we should be looking at the attributes on the types for their
> > > description of what they are expecting shouldn't we ?
> > >
> > >
> > > John Hawkins
> > >
> > >
> > >
> > >
> > >
> > > Samisa Abeysinghe
> > > <samisa.abeysingh
> > > e@gmail.com> To
> > > Apache AXIS C Developers List
> > > 13/12/2004 09:10 <ax...@ws.apache.org>
> > > cc
> > >
> > > Please respond to Subject
> > > "Apache AXIS C Number of decimal places in float
> > > Developers List" (and decimal, double)
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > > Hi Folks,
> > > As per http://nagoya.apache.org/jira/browse/AXISCPP-320, are we
> > > happy with a 4 decimal place acuracy, or do we need more decimal
> > > places be represented correctly?
> > >
> > > Thanks,
> > > Samisa...
> > >
> > >
> > >
> >
> >
>
Re: Number of decimal places in float (and decimal, double)
Posted by Samisa Abeysinghe <sa...@gmail.com>.
Sanjiva, the standard functions (e.g. sprintf, strtod, etc.) that we
are using in serialization/deserialization does not seem to retain
precision the way we want. There are some work-arounds but does not
work exctly the way we want them to.
John, yes I will have a look at XSD spec. I think that is the best
thing to do in order to find out what we sould be supporting.
Thanks,
Samisa...
On Mon, 13 Dec 2004 21:57:36 +0600, Sanjiva Weerawarana
<sa...@opensource.lk> wrote:
> Yes I agree.
>
> Samisa, why wouldn't we retain all the precision that's available?
> Maybe I don't understand the issue (entirely possible!).
>
> Sanjiva.
>
>
>
> ----- Original Message -----
> From: "John Hawkins" <HA...@uk.ibm.com>
> To: "Apache AXIS C Developers List" <ax...@ws.apache.org>
> Sent: Monday, December 13, 2004 7:01 PM
> Subject: Re: Number of decimal places in float (and decimal, double)
>
> >
> >
> >
> >
> > This is a huge area that I can't see we've discussed before?
> >
> > To my logic we should be looking at the attributes on the types for their
> > description of what they are expecting shouldn't we ?
> >
> >
> > John Hawkins
> >
> >
> >
> >
> >
> > Samisa Abeysinghe
> > <samisa.abeysingh
> > e@gmail.com> To
> > Apache AXIS C Developers List
> > 13/12/2004 09:10 <ax...@ws.apache.org>
> > cc
> >
> > Please respond to Subject
> > "Apache AXIS C Number of decimal places in float
> > Developers List" (and decimal, double)
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> > Hi Folks,
> > As per http://nagoya.apache.org/jira/browse/AXISCPP-320, are we
> > happy with a 4 decimal place acuracy, or do we need more decimal
> > places be represented correctly?
> >
> > Thanks,
> > Samisa...
> >
> >
> >
>
>
Re: Number of decimal places in float (and decimal, double)
Posted by Sanjiva Weerawarana <sa...@opensource.lk>.
Yes I agree.
Samisa, why wouldn't we retain all the precision that's available?
Maybe I don't understand the issue (entirely possible!).
Sanjiva.
----- Original Message -----
From: "John Hawkins" <HA...@uk.ibm.com>
To: "Apache AXIS C Developers List" <ax...@ws.apache.org>
Sent: Monday, December 13, 2004 7:01 PM
Subject: Re: Number of decimal places in float (and decimal, double)
>
>
>
>
> This is a huge area that I can't see we've discussed before?
>
> To my logic we should be looking at the attributes on the types for their
> description of what they are expecting shouldn't we ?
>
>
> John Hawkins
>
>
>
>
>
> Samisa Abeysinghe
> <samisa.abeysingh
> e@gmail.com> To
> Apache AXIS C Developers List
> 13/12/2004 09:10 <ax...@ws.apache.org>
> cc
>
> Please respond to Subject
> "Apache AXIS C Number of decimal places in float
> Developers List" (and decimal, double)
>
>
>
>
>
>
>
>
>
>
> Hi Folks,
> As per http://nagoya.apache.org/jira/browse/AXISCPP-320, are we
> happy with a 4 decimal place acuracy, or do we need more decimal
> places be represented correctly?
>
> Thanks,
> Samisa...
>
>
>
Re: Number of decimal places in float (and decimal, double)
Posted by John Hawkins <HA...@uk.ibm.com>.
This is a huge area that I can't see we've discussed before?
To my logic we should be looking at the attributes on the types for their
description of what they are expecting shouldn't we ?
John Hawkins
Samisa Abeysinghe
<samisa.abeysingh
e@gmail.com> To
Apache AXIS C Developers List
13/12/2004 09:10 <ax...@ws.apache.org>
cc
Please respond to Subject
"Apache AXIS C Number of decimal places in float
Developers List" (and decimal, double)
Hi Folks,
As per http://nagoya.apache.org/jira/browse/AXISCPP-320, are we
happy with a 4 decimal place acuracy, or do we need more decimal
places be represented correctly?
Thanks,
Samisa...