IGNITE-3196 - ready for review

classic Classic list List threaded Threaded
26 messages Options
12
Reply | Threaded
Open this post in threaded view
|

IGNITE-3196 - ready for review

daradurvs
Hello. I fixed it. Please, review.

https://issues.apache.org/jira/browse/IGNITE-3196 - Marshaling works wrong
for the BigDecimals that have negative scale
Reply | Threaded
Open this post in threaded view
|

Re: IGNITE-3196 - ready for review

Igor Sapego
Vyacheslav,

I had a look at your PR and left some comments in Jira.

Best Regards,
Igor

On Mon, Jan 30, 2017 at 12:52 PM, Vyacheslav Daradur <[hidden email]>
wrote:

> Hello. I fixed it. Please, review.
>
> https://issues.apache.org/jira/browse/IGNITE-3196 - Marshaling works wrong
> for the BigDecimals that have negative scale
>
Reply | Threaded
Open this post in threaded view
|

Re: IGNITE-3196 - ready for review

daradurvs
Pavel, I see that you are the main contributor of Ignite.NET.

We should repair BinaryUtils#ReadDecimal and BinaryUtils#WriteDecimal.

*ReadDecimal:
byte[] mag = ReadByteArray(stream); // including at least one sign bit,
which is (ceil((this.bitLength() + 1)/8))
bool neg = (mag[0] < 0);
if (scale < 0)
// TODO: a scale of -3 means the unscaled value is multiplied by 1000

*WriteDecimal:
int sign = vals[3] < 0 ? -1 : 0;
stream.WriteInt(sign);

Can you help with this task?


2017-01-31 12:46 GMT+03:00 Igor Sapego <[hidden email]>:

> Vyacheslav,
>
> I had a look at your PR and left some comments in Jira.
>
> Best Regards,
> Igor
>
> On Mon, Jan 30, 2017 at 12:52 PM, Vyacheslav Daradur <[hidden email]>
> wrote:
>
> > Hello. I fixed it. Please, review.
> >
> > https://issues.apache.org/jira/browse/IGNITE-3196 - Marshaling works
> wrong
> > for the BigDecimals that have negative scale
> >
>
Reply | Threaded
Open this post in threaded view
|

Re: IGNITE-3196 - ready for review

Pavel Tupitsyn
Vyacheslav,

I'm not sure I understand the code you attached.

If you know how to fix the .NET part, can you just do it in your PR and run
"Platform .NET" on TeamCity to verify?
http://ci.ignite.apache.org/viewType.html?buildTypeId=IgniteTests_IgnitePlatformNet

Thanks,

Pavel

On Tue, Jan 31, 2017 at 1:35 PM, Vyacheslav Daradur <[hidden email]>
wrote:

> Pavel, I see that you are the main contributor of Ignite.NET.
>
> We should repair BinaryUtils#ReadDecimal and BinaryUtils#WriteDecimal.
>
> *ReadDecimal:
> byte[] mag = ReadByteArray(stream); // including at least one sign bit,
> which is (ceil((this.bitLength() + 1)/8))
> bool neg = (mag[0] < 0);
> if (scale < 0)
> // TODO: a scale of -3 means the unscaled value is multiplied by 1000
>
> *WriteDecimal:
> int sign = vals[3] < 0 ? -1 : 0;
> stream.WriteInt(sign);
>
> Can you help with this task?
>
>
> 2017-01-31 12:46 GMT+03:00 Igor Sapego <[hidden email]>:
>
>> Vyacheslav,
>>
>> I had a look at your PR and left some comments in Jira.
>>
>> Best Regards,
>> Igor
>>
>> On Mon, Jan 30, 2017 at 12:52 PM, Vyacheslav Daradur <[hidden email]
>> >
>> wrote:
>>
>> > Hello. I fixed it. Please, review.
>> >
>> > https://issues.apache.org/jira/browse/IGNITE-3196 - Marshaling works
>> wrong
>> > for the BigDecimals that have negative scale
>> >
>>
>
>
Reply | Threaded
Open this post in threaded view
|

Re: IGNITE-3196 - ready for review

daradurvs
Hello.

I looked and understood the code of methods ReadDecimal and WriteDecimal on
.NET platform.

1. At the moment remaking of this methods for my Java-decimal-fix is very
difficult, it needs to write new methods for serialization/deserialization
of negative decimals.

I can make it, but there is simpler decision: to add additional byte for
sign.

I need advice: difficult solution (new methods .net) Versus : simple
solutions (additional byte)?

*I don't know yet, what changes are necessary on С++ platform.

2. I see a problem with the negative scale on .NET platform.

Now negative scale is forbidden.

We can make:
if (scale < 0) return Decimal.Multiply(new decimal(lo, mid, hi, neg, 0),
new decimal(Math.Pow(10, -scale)));

But there is the problem:
* 1 Serialize in Java; number=123456789, scale=-4
* 2 Deserialize in .NET; number=1234567890000, scale=0
* 3 Serialize in .NET; number=1234567890000, scale=0
* 4 Deserialize in Java; number=1234567890000, scale=0

Logically: (1) 123456789 * 10^4 == (2)  1234567890000

In Java (1) not equal (2), because scales are different.

Any thougths?

2017-01-31 14:08 GMT+03:00 Pavel Tupitsyn <[hidden email]>:

> Vyacheslav,
>
> I'm not sure I understand the code you attached.
>
> If you know how to fix the .NET part, can you just do it in your PR and
> run "Platform .NET" on TeamCity to verify?
> http://ci.ignite.apache.org/viewType.html?buildTypeId=
> IgniteTests_IgnitePlatformNet
>
> Thanks,
>
> Pavel
>
> On Tue, Jan 31, 2017 at 1:35 PM, Vyacheslav Daradur <[hidden email]>
> wrote:
>
>> Pavel, I see that you are the main contributor of Ignite.NET.
>>
>> We should repair BinaryUtils#ReadDecimal and BinaryUtils#WriteDecimal.
>>
>> *ReadDecimal:
>> byte[] mag = ReadByteArray(stream); // including at least one sign bit,
>> which is (ceil((this.bitLength() + 1)/8))
>> bool neg = (mag[0] < 0);
>> if (scale < 0)
>> // TODO: a scale of -3 means the unscaled value is multiplied by 1000
>>
>> *WriteDecimal:
>> int sign = vals[3] < 0 ? -1 : 0;
>> stream.WriteInt(sign);
>>
>> Can you help with this task?
>>
>>
>> 2017-01-31 12:46 GMT+03:00 Igor Sapego <[hidden email]>:
>>
>>> Vyacheslav,
>>>
>>> I had a look at your PR and left some comments in Jira.
>>>
>>> Best Regards,
>>> Igor
>>>
>>> On Mon, Jan 30, 2017 at 12:52 PM, Vyacheslav Daradur <
>>> [hidden email]>
>>> wrote:
>>>
>>> > Hello. I fixed it. Please, review.
>>> >
>>> > https://issues.apache.org/jira/browse/IGNITE-3196 - Marshaling works
>>> wrong
>>> > for the BigDecimals that have negative scale
>>> >
>>>
>>
>>
>
Reply | Threaded
Open this post in threaded view
|

Re: IGNITE-3196 - ready for review

Pavel Tupitsyn
Vyacheslav,

I see the problem now. Yes, negative scale is not supported in .NET.

I don't think we should do the multiplication. As you described, this will
break equality on Java side. SQL queries might be broken, etc.
I think we should throw an exception in .NET when encountering negative
decimal scale.

Vladimir O, any thoughts?

Pavel

On Fri, Feb 3, 2017 at 12:01 PM, Vyacheslav Daradur <[hidden email]>
wrote:

> Hello.
>
> I looked and understood the code of methods ReadDecimal and WriteDecimal
> on .NET platform.
>
> 1. At the moment remaking of this methods for my Java-decimal-fix is very
> difficult, it needs to write new methods for serialization/deserialization
> of negative decimals.
>
> I can make it, but there is simpler decision: to add additional byte for
> sign.
>
> I need advice: difficult solution (new methods .net) Versus : simple
> solutions (additional byte)?
>
> *I don't know yet, what changes are necessary on С++ platform.
>
> 2. I see a problem with the negative scale on .NET platform.
>
> Now negative scale is forbidden.
>
> We can make:
> if (scale < 0) return Decimal.Multiply(new decimal(lo, mid, hi, neg, 0),
> new decimal(Math.Pow(10, -scale)));
>
> But there is the problem:
> * 1 Serialize in Java; number=123456789, scale=-4
> * 2 Deserialize in .NET; number=1234567890000, scale=0
> * 3 Serialize in .NET; number=1234567890000, scale=0
> * 4 Deserialize in Java; number=1234567890000, scale=0
>
> Logically: (1) 123456789 * 10^4 == (2)  1234567890000
>
> In Java (1) not equal (2), because scales are different.
>
> Any thougths?
>
> 2017-01-31 14:08 GMT+03:00 Pavel Tupitsyn <[hidden email]>:
>
>> Vyacheslav,
>>
>> I'm not sure I understand the code you attached.
>>
>> If you know how to fix the .NET part, can you just do it in your PR and
>> run "Platform .NET" on TeamCity to verify?
>> http://ci.ignite.apache.org/viewType.html?buildTypeId=Ignite
>> Tests_IgnitePlatformNet
>>
>> Thanks,
>>
>> Pavel
>>
>> On Tue, Jan 31, 2017 at 1:35 PM, Vyacheslav Daradur <[hidden email]>
>> wrote:
>>
>>> Pavel, I see that you are the main contributor of Ignite.NET.
>>>
>>> We should repair BinaryUtils#ReadDecimal and BinaryUtils#WriteDecimal.
>>>
>>> *ReadDecimal:
>>> byte[] mag = ReadByteArray(stream); // including at least one sign bit,
>>> which is (ceil((this.bitLength() + 1)/8))
>>> bool neg = (mag[0] < 0);
>>> if (scale < 0)
>>> // TODO: a scale of -3 means the unscaled value is multiplied by 1000
>>>
>>> *WriteDecimal:
>>> int sign = vals[3] < 0 ? -1 : 0;
>>> stream.WriteInt(sign);
>>>
>>> Can you help with this task?
>>>
>>>
>>> 2017-01-31 12:46 GMT+03:00 Igor Sapego <[hidden email]>:
>>>
>>>> Vyacheslav,
>>>>
>>>> I had a look at your PR and left some comments in Jira.
>>>>
>>>> Best Regards,
>>>> Igor
>>>>
>>>> On Mon, Jan 30, 2017 at 12:52 PM, Vyacheslav Daradur <
>>>> [hidden email]>
>>>> wrote:
>>>>
>>>> > Hello. I fixed it. Please, review.
>>>> >
>>>> > https://issues.apache.org/jira/browse/IGNITE-3196 - Marshaling works
>>>> wrong
>>>> > for the BigDecimals that have negative scale
>>>> >
>>>>
>>>
>>>
>>
>
Reply | Threaded
Open this post in threaded view
|

Re: IGNITE-3196 - ready for review

daradurvs
1. On my first question
I think up, if we serialize only positive numbers, we can write sign in
first byte, because it is positive always.
I will try to make this decision

2017-02-03 12:48 GMT+03:00 Pavel Tupitsyn <[hidden email]>:

> Vyacheslav,
>
> I see the problem now. Yes, negative scale is not supported in .NET.
>
> I don't think we should do the multiplication. As you described, this will
> break equality on Java side. SQL queries might be broken, etc.
> I think we should throw an exception in .NET when encountering negative
> decimal scale.
>
> Vladimir O, any thoughts?
>
> Pavel
>
> On Fri, Feb 3, 2017 at 12:01 PM, Vyacheslav Daradur <[hidden email]>
> wrote:
>
> > Hello.
> >
> > I looked and understood the code of methods ReadDecimal and WriteDecimal
> > on .NET platform.
> >
> > 1. At the moment remaking of this methods for my Java-decimal-fix is very
> > difficult, it needs to write new methods for
> serialization/deserialization
> > of negative decimals.
> >
> > I can make it, but there is simpler decision: to add additional byte for
> > sign.
> >
> > I need advice: difficult solution (new methods .net) Versus : simple
> > solutions (additional byte)?
> >
> > *I don't know yet, what changes are necessary on С++ platform.
> >
> > 2. I see a problem with the negative scale on .NET platform.
> >
> > Now negative scale is forbidden.
> >
> > We can make:
> > if (scale < 0) return Decimal.Multiply(new decimal(lo, mid, hi, neg, 0),
> > new decimal(Math.Pow(10, -scale)));
> >
> > But there is the problem:
> > * 1 Serialize in Java; number=123456789, scale=-4
> > * 2 Deserialize in .NET; number=1234567890000, scale=0
> > * 3 Serialize in .NET; number=1234567890000, scale=0
> > * 4 Deserialize in Java; number=1234567890000, scale=0
> >
> > Logically: (1) 123456789 * 10^4 == (2)  1234567890000
> >
> > In Java (1) not equal (2), because scales are different.
> >
> > Any thougths?
> >
> > 2017-01-31 14:08 GMT+03:00 Pavel Tupitsyn <[hidden email]>:
> >
> >> Vyacheslav,
> >>
> >> I'm not sure I understand the code you attached.
> >>
> >> If you know how to fix the .NET part, can you just do it in your PR and
> >> run "Platform .NET" on TeamCity to verify?
> >> http://ci.ignite.apache.org/viewType.html?buildTypeId=Ignite
> >> Tests_IgnitePlatformNet
> >>
> >> Thanks,
> >>
> >> Pavel
> >>
> >> On Tue, Jan 31, 2017 at 1:35 PM, Vyacheslav Daradur <
> [hidden email]>
> >> wrote:
> >>
> >>> Pavel, I see that you are the main contributor of Ignite.NET.
> >>>
> >>> We should repair BinaryUtils#ReadDecimal and BinaryUtils#WriteDecimal.
> >>>
> >>> *ReadDecimal:
> >>> byte[] mag = ReadByteArray(stream); // including at least one sign bit,
> >>> which is (ceil((this.bitLength() + 1)/8))
> >>> bool neg = (mag[0] < 0);
> >>> if (scale < 0)
> >>> // TODO: a scale of -3 means the unscaled value is multiplied by 1000
> >>>
> >>> *WriteDecimal:
> >>> int sign = vals[3] < 0 ? -1 : 0;
> >>> stream.WriteInt(sign);
> >>>
> >>> Can you help with this task?
> >>>
> >>>
> >>> 2017-01-31 12:46 GMT+03:00 Igor Sapego <[hidden email]>:
> >>>
> >>>> Vyacheslav,
> >>>>
> >>>> I had a look at your PR and left some comments in Jira.
> >>>>
> >>>> Best Regards,
> >>>> Igor
> >>>>
> >>>> On Mon, Jan 30, 2017 at 12:52 PM, Vyacheslav Daradur <
> >>>> [hidden email]>
> >>>> wrote:
> >>>>
> >>>> > Hello. I fixed it. Please, review.
> >>>> >
> >>>> > https://issues.apache.org/jira/browse/IGNITE-3196 - Marshaling
> works
> >>>> wrong
> >>>> > for the BigDecimals that have negative scale
> >>>> >
> >>>>
> >>>
> >>>
> >>
> >
>
Reply | Threaded
Open this post in threaded view
|

Re: IGNITE-3196 - ready for review

daradurvs
Pavel, Igor

Please, review it again.

https://github.com/apache/ignite/pull/1473/files

All tests
<http://ci.ignite.apache.org/viewLog.html?buildId=444231&tab=buildResultsDiv&buildTypeId=IgniteTests_RunAll>
.NET tests
<http://ci.ignite.apache.org/viewLog.html?buildId=443439&tab=buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformNet>

How about this solution?

2017-02-03 13:59 GMT+03:00 Vyacheslav Daradur <[hidden email]>:

> 1. On my first question
> I think up, if we serialize only positive numbers, we can write sign in
> first byte, because it is positive always.
> I will try to make this decision
>
> 2017-02-03 12:48 GMT+03:00 Pavel Tupitsyn <[hidden email]>:
>
>> Vyacheslav,
>>
>> I see the problem now. Yes, negative scale is not supported in .NET.
>>
>> I don't think we should do the multiplication. As you described, this will
>> break equality on Java side. SQL queries might be broken, etc.
>> I think we should throw an exception in .NET when encountering negative
>> decimal scale.
>>
>> Vladimir O, any thoughts?
>>
>> Pavel
>>
>> On Fri, Feb 3, 2017 at 12:01 PM, Vyacheslav Daradur <[hidden email]>
>> wrote:
>>
>> > Hello.
>> >
>> > I looked and understood the code of methods ReadDecimal and WriteDecimal
>> > on .NET platform.
>> >
>> > 1. At the moment remaking of this methods for my Java-decimal-fix is
>> very
>> > difficult, it needs to write new methods for
>> serialization/deserialization
>> > of negative decimals.
>> >
>> > I can make it, but there is simpler decision: to add additional byte for
>> > sign.
>> >
>> > I need advice: difficult solution (new methods .net) Versus : simple
>> > solutions (additional byte)?
>> >
>> > *I don't know yet, what changes are necessary on С++ platform.
>> >
>> > 2. I see a problem with the negative scale on .NET platform.
>> >
>> > Now negative scale is forbidden.
>> >
>> > We can make:
>> > if (scale < 0) return Decimal.Multiply(new decimal(lo, mid, hi, neg, 0),
>> > new decimal(Math.Pow(10, -scale)));
>> >
>> > But there is the problem:
>> > * 1 Serialize in Java; number=123456789, scale=-4
>> > * 2 Deserialize in .NET; number=1234567890000, scale=0
>> > * 3 Serialize in .NET; number=1234567890000, scale=0
>> > * 4 Deserialize in Java; number=1234567890000, scale=0
>> >
>> > Logically: (1) 123456789 * 10^4 == (2)  1234567890000
>> >
>> > In Java (1) not equal (2), because scales are different.
>> >
>> > Any thougths?
>> >
>> > 2017-01-31 14:08 GMT+03:00 Pavel Tupitsyn <[hidden email]>:
>> >
>> >> Vyacheslav,
>> >>
>> >> I'm not sure I understand the code you attached.
>> >>
>> >> If you know how to fix the .NET part, can you just do it in your PR and
>> >> run "Platform .NET" on TeamCity to verify?
>> >> http://ci.ignite.apache.org/viewType.html?buildTypeId=Ignite
>> >> Tests_IgnitePlatformNet
>> >>
>> >> Thanks,
>> >>
>> >> Pavel
>> >>
>> >> On Tue, Jan 31, 2017 at 1:35 PM, Vyacheslav Daradur <
>> [hidden email]>
>> >> wrote:
>> >>
>> >>> Pavel, I see that you are the main contributor of Ignite.NET.
>> >>>
>> >>> We should repair BinaryUtils#ReadDecimal and BinaryUtils#WriteDecimal.
>> >>>
>> >>> *ReadDecimal:
>> >>> byte[] mag = ReadByteArray(stream); // including at least one sign
>> bit,
>> >>> which is (ceil((this.bitLength() + 1)/8))
>> >>> bool neg = (mag[0] < 0);
>> >>> if (scale < 0)
>> >>> // TODO: a scale of -3 means the unscaled value is multiplied by 1000
>> >>>
>> >>> *WriteDecimal:
>> >>> int sign = vals[3] < 0 ? -1 : 0;
>> >>> stream.WriteInt(sign);
>> >>>
>> >>> Can you help with this task?
>> >>>
>> >>>
>> >>> 2017-01-31 12:46 GMT+03:00 Igor Sapego <[hidden email]>:
>> >>>
>> >>>> Vyacheslav,
>> >>>>
>> >>>> I had a look at your PR and left some comments in Jira.
>> >>>>
>> >>>> Best Regards,
>> >>>> Igor
>> >>>>
>> >>>> On Mon, Jan 30, 2017 at 12:52 PM, Vyacheslav Daradur <
>> >>>> [hidden email]>
>> >>>> wrote:
>> >>>>
>> >>>> > Hello. I fixed it. Please, review.
>> >>>> >
>> >>>> > https://issues.apache.org/jira/browse/IGNITE-3196 - Marshaling
>> works
>> >>>> wrong
>> >>>> > for the BigDecimals that have negative scale
>> >>>> >
>> >>>>
>> >>>
>> >>>
>> >>
>> >
>>
>
>
Reply | Threaded
Open this post in threaded view
|

Re: IGNITE-3196 - ready for review

Igor Sapego
Vyacheslav, I can see two ODBC tests fail in C++ test suits that should not:
- TestEscConvertFunctionFloat
<http://ci.ignite.apache.org/viewLog.html?buildId=444207&tab=buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformCppLinux#testNameId-9178617718508801660>
- TestEscConvertFunctionDouble
<http://ci.ignite.apache.org/viewLog.html?buildId=444207&tab=buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformCppLinux#testNameId5432107083822590090>
.

I believe, this is because I can't see any changes in C++ Decimal
marshaling code.
Please, pay attention to file
ignite\modules\platforms\cpp\odbc\src\utility.cpp,
functions ReadDecimal and WriteDecimal.

Best Regards,
Igor

On Mon, Feb 6, 2017 at 11:21 AM, Vyacheslav Daradur <[hidden email]>
wrote:

> Pavel, Igor
>
> Please, review it again.
>
> https://github.com/apache/ignite/pull/1473/files
>
> All tests
> <http://ci.ignite.apache.org/viewLog.html?buildId=444231&tab=buildResultsDiv&buildTypeId=IgniteTests_RunAll>
> .NET tests
> <http://ci.ignite.apache.org/viewLog.html?buildId=443439&tab=buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformNet>
>
> How about this solution?
>
> 2017-02-03 13:59 GMT+03:00 Vyacheslav Daradur <[hidden email]>:
>
>> 1. On my first question
>> I think up, if we serialize only positive numbers, we can write sign in
>> first byte, because it is positive always.
>> I will try to make this decision
>>
>> 2017-02-03 12:48 GMT+03:00 Pavel Tupitsyn <[hidden email]>:
>>
>>> Vyacheslav,
>>>
>>> I see the problem now. Yes, negative scale is not supported in .NET.
>>>
>>> I don't think we should do the multiplication. As you described, this
>>> will
>>> break equality on Java side. SQL queries might be broken, etc.
>>> I think we should throw an exception in .NET when encountering negative
>>> decimal scale.
>>>
>>> Vladimir O, any thoughts?
>>>
>>> Pavel
>>>
>>> On Fri, Feb 3, 2017 at 12:01 PM, Vyacheslav Daradur <[hidden email]
>>> >
>>> wrote:
>>>
>>> > Hello.
>>> >
>>> > I looked and understood the code of methods ReadDecimal and
>>> WriteDecimal
>>> > on .NET platform.
>>> >
>>> > 1. At the moment remaking of this methods for my Java-decimal-fix is
>>> very
>>> > difficult, it needs to write new methods for
>>> serialization/deserialization
>>> > of negative decimals.
>>> >
>>> > I can make it, but there is simpler decision: to add additional byte
>>> for
>>> > sign.
>>> >
>>> > I need advice: difficult solution (new methods .net) Versus : simple
>>> > solutions (additional byte)?
>>> >
>>> > *I don't know yet, what changes are necessary on С++ platform.
>>> >
>>> > 2. I see a problem with the negative scale on .NET platform.
>>> >
>>> > Now negative scale is forbidden.
>>> >
>>> > We can make:
>>> > if (scale < 0) return Decimal.Multiply(new decimal(lo, mid, hi, neg,
>>> 0),
>>> > new decimal(Math.Pow(10, -scale)));
>>> >
>>> > But there is the problem:
>>> > * 1 Serialize in Java; number=123456789, scale=-4
>>> > * 2 Deserialize in .NET; number=1234567890000, scale=0
>>> > * 3 Serialize in .NET; number=1234567890000, scale=0
>>> > * 4 Deserialize in Java; number=1234567890000, scale=0
>>> >
>>> > Logically: (1) 123456789 * 10^4 == (2)  1234567890000
>>> >
>>> > In Java (1) not equal (2), because scales are different.
>>> >
>>> > Any thougths?
>>> >
>>> > 2017-01-31 14:08 GMT+03:00 Pavel Tupitsyn <[hidden email]>:
>>> >
>>> >> Vyacheslav,
>>> >>
>>> >> I'm not sure I understand the code you attached.
>>> >>
>>> >> If you know how to fix the .NET part, can you just do it in your PR
>>> and
>>> >> run "Platform .NET" on TeamCity to verify?
>>> >> http://ci.ignite.apache.org/viewType.html?buildTypeId=Ignite
>>> >> Tests_IgnitePlatformNet
>>> >>
>>> >> Thanks,
>>> >>
>>> >> Pavel
>>> >>
>>> >> On Tue, Jan 31, 2017 at 1:35 PM, Vyacheslav Daradur <
>>> [hidden email]>
>>> >> wrote:
>>> >>
>>> >>> Pavel, I see that you are the main contributor of Ignite.NET.
>>> >>>
>>> >>> We should repair BinaryUtils#ReadDecimal and
>>> BinaryUtils#WriteDecimal.
>>> >>>
>>> >>> *ReadDecimal:
>>> >>> byte[] mag = ReadByteArray(stream); // including at least one sign
>>> bit,
>>> >>> which is (ceil((this.bitLength() + 1)/8))
>>> >>> bool neg = (mag[0] < 0);
>>> >>> if (scale < 0)
>>> >>> // TODO: a scale of -3 means the unscaled value is multiplied by 1000
>>> >>>
>>> >>> *WriteDecimal:
>>> >>> int sign = vals[3] < 0 ? -1 : 0;
>>> >>> stream.WriteInt(sign);
>>> >>>
>>> >>> Can you help with this task?
>>> >>>
>>> >>>
>>> >>> 2017-01-31 12:46 GMT+03:00 Igor Sapego <[hidden email]>:
>>> >>>
>>> >>>> Vyacheslav,
>>> >>>>
>>> >>>> I had a look at your PR and left some comments in Jira.
>>> >>>>
>>> >>>> Best Regards,
>>> >>>> Igor
>>> >>>>
>>> >>>> On Mon, Jan 30, 2017 at 12:52 PM, Vyacheslav Daradur <
>>> >>>> [hidden email]>
>>> >>>> wrote:
>>> >>>>
>>> >>>> > Hello. I fixed it. Please, review.
>>> >>>> >
>>> >>>> > https://issues.apache.org/jira/browse/IGNITE-3196 - Marshaling
>>> works
>>> >>>> wrong
>>> >>>> > for the BigDecimals that have negative scale
>>> >>>> >
>>> >>>>
>>> >>>
>>> >>>
>>> >>
>>> >
>>>
>>
>>
>
Reply | Threaded
Open this post in threaded view
|

Re: IGNITE-3196 - ready for review

Pavel Tupitsyn
.NET changes look good to me.

Pavel

On Mon, Feb 6, 2017 at 1:10 PM, Igor Sapego <[hidden email]> wrote:

> Vyacheslav, I can see two ODBC tests fail in C++ test suits that should
> not:
> - TestEscConvertFunctionFloat
> <http://ci.ignite.apache.org/viewLog.html?buildId=444207&tab=buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformCppLinux#testNameId-9178617718508801660>
> - TestEscConvertFunctionDouble
> <http://ci.ignite.apache.org/viewLog.html?buildId=444207&tab=buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformCppLinux#testNameId5432107083822590090>
> .
>
> I believe, this is because I can't see any changes in C++ Decimal
> marshaling code.
> Please, pay attention to file ignite\modules\platforms\cpp\
> odbc\src\utility.cpp,
> functions ReadDecimal and WriteDecimal.
>
> Best Regards,
> Igor
>
> On Mon, Feb 6, 2017 at 11:21 AM, Vyacheslav Daradur <[hidden email]>
> wrote:
>
>> Pavel, Igor
>>
>> Please, review it again.
>>
>> https://github.com/apache/ignite/pull/1473/files
>>
>> All tests
>> <http://ci.ignite.apache.org/viewLog.html?buildId=444231&tab=buildResultsDiv&buildTypeId=IgniteTests_RunAll>
>> .NET tests
>> <http://ci.ignite.apache.org/viewLog.html?buildId=443439&tab=buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformNet>
>>
>> How about this solution?
>>
>> 2017-02-03 13:59 GMT+03:00 Vyacheslav Daradur <[hidden email]>:
>>
>>> 1. On my first question
>>> I think up, if we serialize only positive numbers, we can write sign in
>>> first byte, because it is positive always.
>>> I will try to make this decision
>>>
>>> 2017-02-03 12:48 GMT+03:00 Pavel Tupitsyn <[hidden email]>:
>>>
>>>> Vyacheslav,
>>>>
>>>> I see the problem now. Yes, negative scale is not supported in .NET.
>>>>
>>>> I don't think we should do the multiplication. As you described, this
>>>> will
>>>> break equality on Java side. SQL queries might be broken, etc.
>>>> I think we should throw an exception in .NET when encountering negative
>>>> decimal scale.
>>>>
>>>> Vladimir O, any thoughts?
>>>>
>>>> Pavel
>>>>
>>>> On Fri, Feb 3, 2017 at 12:01 PM, Vyacheslav Daradur <
>>>> [hidden email]>
>>>> wrote:
>>>>
>>>> > Hello.
>>>> >
>>>> > I looked and understood the code of methods ReadDecimal and
>>>> WriteDecimal
>>>> > on .NET platform.
>>>> >
>>>> > 1. At the moment remaking of this methods for my Java-decimal-fix is
>>>> very
>>>> > difficult, it needs to write new methods for
>>>> serialization/deserialization
>>>> > of negative decimals.
>>>> >
>>>> > I can make it, but there is simpler decision: to add additional byte
>>>> for
>>>> > sign.
>>>> >
>>>> > I need advice: difficult solution (new methods .net) Versus : simple
>>>> > solutions (additional byte)?
>>>> >
>>>> > *I don't know yet, what changes are necessary on С++ platform.
>>>> >
>>>> > 2. I see a problem with the negative scale on .NET platform.
>>>> >
>>>> > Now negative scale is forbidden.
>>>> >
>>>> > We can make:
>>>> > if (scale < 0) return Decimal.Multiply(new decimal(lo, mid, hi, neg,
>>>> 0),
>>>> > new decimal(Math.Pow(10, -scale)));
>>>> >
>>>> > But there is the problem:
>>>> > * 1 Serialize in Java; number=123456789, scale=-4
>>>> > * 2 Deserialize in .NET; number=1234567890000, scale=0
>>>> > * 3 Serialize in .NET; number=1234567890000, scale=0
>>>> > * 4 Deserialize in Java; number=1234567890000, scale=0
>>>> >
>>>> > Logically: (1) 123456789 * 10^4 == (2)  1234567890000
>>>> >
>>>> > In Java (1) not equal (2), because scales are different.
>>>> >
>>>> > Any thougths?
>>>> >
>>>> > 2017-01-31 14:08 GMT+03:00 Pavel Tupitsyn <[hidden email]>:
>>>> >
>>>> >> Vyacheslav,
>>>> >>
>>>> >> I'm not sure I understand the code you attached.
>>>> >>
>>>> >> If you know how to fix the .NET part, can you just do it in your PR
>>>> and
>>>> >> run "Platform .NET" on TeamCity to verify?
>>>> >> http://ci.ignite.apache.org/viewType.html?buildTypeId=Ignite
>>>> >> Tests_IgnitePlatformNet
>>>> >>
>>>> >> Thanks,
>>>> >>
>>>> >> Pavel
>>>> >>
>>>> >> On Tue, Jan 31, 2017 at 1:35 PM, Vyacheslav Daradur <
>>>> [hidden email]>
>>>> >> wrote:
>>>> >>
>>>> >>> Pavel, I see that you are the main contributor of Ignite.NET.
>>>> >>>
>>>> >>> We should repair BinaryUtils#ReadDecimal and
>>>> BinaryUtils#WriteDecimal.
>>>> >>>
>>>> >>> *ReadDecimal:
>>>> >>> byte[] mag = ReadByteArray(stream); // including at least one sign
>>>> bit,
>>>> >>> which is (ceil((this.bitLength() + 1)/8))
>>>> >>> bool neg = (mag[0] < 0);
>>>> >>> if (scale < 0)
>>>> >>> // TODO: a scale of -3 means the unscaled value is multiplied by
>>>> 1000
>>>> >>>
>>>> >>> *WriteDecimal:
>>>> >>> int sign = vals[3] < 0 ? -1 : 0;
>>>> >>> stream.WriteInt(sign);
>>>> >>>
>>>> >>> Can you help with this task?
>>>> >>>
>>>> >>>
>>>> >>> 2017-01-31 12:46 GMT+03:00 Igor Sapego <[hidden email]>:
>>>> >>>
>>>> >>>> Vyacheslav,
>>>> >>>>
>>>> >>>> I had a look at your PR and left some comments in Jira.
>>>> >>>>
>>>> >>>> Best Regards,
>>>> >>>> Igor
>>>> >>>>
>>>> >>>> On Mon, Jan 30, 2017 at 12:52 PM, Vyacheslav Daradur <
>>>> >>>> [hidden email]>
>>>> >>>> wrote:
>>>> >>>>
>>>> >>>> > Hello. I fixed it. Please, review.
>>>> >>>> >
>>>> >>>> > https://issues.apache.org/jira/browse/IGNITE-3196 - Marshaling
>>>> works
>>>> >>>> wrong
>>>> >>>> > for the BigDecimals that have negative scale
>>>> >>>> >
>>>> >>>>
>>>> >>>
>>>> >>>
>>>> >>
>>>> >
>>>>
>>>
>>>
>>
>
Reply | Threaded
Open this post in threaded view
|

Re: IGNITE-3196 - ready for review

daradurvs
Igor,

I didn't change the CPP code before approval approach.
I shall write directly, sorry.

But I made CPP changes already.

> TestEscConvertFunctionFloat
> TestEscConvertFunctionDouble.
These tests were passed
<http://ci.ignite.apache.org/viewQueued.html?itemId=445824>



2017-02-06 13:20 GMT+03:00 Pavel Tupitsyn <[hidden email]>:

> .NET changes look good to me.
>
> Pavel
>
> On Mon, Feb 6, 2017 at 1:10 PM, Igor Sapego <[hidden email]> wrote:
>
> > Vyacheslav, I can see two ODBC tests fail in C++ test suits that should
> > not:
> > - TestEscConvertFunctionFloat
> > <http://ci.ignite.apache.org/viewLog.html?buildId=444207&tab
> =buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformCppLi
> nux#testNameId-9178617718508801660>
> > - TestEscConvertFunctionDouble
> > <http://ci.ignite.apache.org/viewLog.html?buildId=444207&tab
> =buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformCppLi
> nux#testNameId5432107083822590090>
> > .
> >
> > I believe, this is because I can't see any changes in C++ Decimal
> > marshaling code.
> > Please, pay attention to file ignite\modules\platforms\cpp\
> > odbc\src\utility.cpp,
> > functions ReadDecimal and WriteDecimal.
> >
> > Best Regards,
> > Igor
> >
> > On Mon, Feb 6, 2017 at 11:21 AM, Vyacheslav Daradur <[hidden email]
> >
> > wrote:
> >
> >> Pavel, Igor
> >>
> >> Please, review it again.
> >>
> >> https://github.com/apache/ignite/pull/1473/files
> >>
> >> All tests
> >> <http://ci.ignite.apache.org/viewLog.html?buildId=444231&tab
> =buildResultsDiv&buildTypeId=IgniteTests_RunAll>
> >> .NET tests
> >> <http://ci.ignite.apache.org/viewLog.html?buildId=443439&tab
> =buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformNet>
> >>
> >> How about this solution?
> >>
> >> 2017-02-03 13:59 GMT+03:00 Vyacheslav Daradur <[hidden email]>:
> >>
> >>> 1. On my first question
> >>> I think up, if we serialize only positive numbers, we can write sign in
> >>> first byte, because it is positive always.
> >>> I will try to make this decision
> >>>
> >>> 2017-02-03 12:48 GMT+03:00 Pavel Tupitsyn <[hidden email]>:
> >>>
> >>>> Vyacheslav,
> >>>>
> >>>> I see the problem now. Yes, negative scale is not supported in .NET.
> >>>>
> >>>> I don't think we should do the multiplication. As you described, this
> >>>> will
> >>>> break equality on Java side. SQL queries might be broken, etc.
> >>>> I think we should throw an exception in .NET when encountering
> negative
> >>>> decimal scale.
> >>>>
> >>>> Vladimir O, any thoughts?
> >>>>
> >>>> Pavel
> >>>>
> >>>> On Fri, Feb 3, 2017 at 12:01 PM, Vyacheslav Daradur <
> >>>> [hidden email]>
> >>>> wrote:
> >>>>
> >>>> > Hello.
> >>>> >
> >>>> > I looked and understood the code of methods ReadDecimal and
> >>>> WriteDecimal
> >>>> > on .NET platform.
> >>>> >
> >>>> > 1. At the moment remaking of this methods for my Java-decimal-fix is
> >>>> very
> >>>> > difficult, it needs to write new methods for
> >>>> serialization/deserialization
> >>>> > of negative decimals.
> >>>> >
> >>>> > I can make it, but there is simpler decision: to add additional byte
> >>>> for
> >>>> > sign.
> >>>> >
> >>>> > I need advice: difficult solution (new methods .net) Versus : simple
> >>>> > solutions (additional byte)?
> >>>> >
> >>>> > *I don't know yet, what changes are necessary on С++ platform.
> >>>> >
> >>>> > 2. I see a problem with the negative scale on .NET platform.
> >>>> >
> >>>> > Now negative scale is forbidden.
> >>>> >
> >>>> > We can make:
> >>>> > if (scale < 0) return Decimal.Multiply(new decimal(lo, mid, hi, neg,
> >>>> 0),
> >>>> > new decimal(Math.Pow(10, -scale)));
> >>>> >
> >>>> > But there is the problem:
> >>>> > * 1 Serialize in Java; number=123456789, scale=-4
> >>>> > * 2 Deserialize in .NET; number=1234567890000, scale=0
> >>>> > * 3 Serialize in .NET; number=1234567890000, scale=0
> >>>> > * 4 Deserialize in Java; number=1234567890000, scale=0
> >>>> >
> >>>> > Logically: (1) 123456789 * 10^4 == (2)  1234567890000
> >>>> >
> >>>> > In Java (1) not equal (2), because scales are different.
> >>>> >
> >>>> > Any thougths?
> >>>> >
> >>>> > 2017-01-31 14:08 GMT+03:00 Pavel Tupitsyn <[hidden email]>:
> >>>> >
> >>>> >> Vyacheslav,
> >>>> >>
> >>>> >> I'm not sure I understand the code you attached.
> >>>> >>
> >>>> >> If you know how to fix the .NET part, can you just do it in your PR
> >>>> and
> >>>> >> run "Platform .NET" on TeamCity to verify?
> >>>> >> http://ci.ignite.apache.org/viewType.html?buildTypeId=Ignite
> >>>> >> Tests_IgnitePlatformNet
> >>>> >>
> >>>> >> Thanks,
> >>>> >>
> >>>> >> Pavel
> >>>> >>
> >>>> >> On Tue, Jan 31, 2017 at 1:35 PM, Vyacheslav Daradur <
> >>>> [hidden email]>
> >>>> >> wrote:
> >>>> >>
> >>>> >>> Pavel, I see that you are the main contributor of Ignite.NET.
> >>>> >>>
> >>>> >>> We should repair BinaryUtils#ReadDecimal and
> >>>> BinaryUtils#WriteDecimal.
> >>>> >>>
> >>>> >>> *ReadDecimal:
> >>>> >>> byte[] mag = ReadByteArray(stream); // including at least one sign
> >>>> bit,
> >>>> >>> which is (ceil((this.bitLength() + 1)/8))
> >>>> >>> bool neg = (mag[0] < 0);
> >>>> >>> if (scale < 0)
> >>>> >>> // TODO: a scale of -3 means the unscaled value is multiplied by
> >>>> 1000
> >>>> >>>
> >>>> >>> *WriteDecimal:
> >>>> >>> int sign = vals[3] < 0 ? -1 : 0;
> >>>> >>> stream.WriteInt(sign);
> >>>> >>>
> >>>> >>> Can you help with this task?
> >>>> >>>
> >>>> >>>
> >>>> >>> 2017-01-31 12:46 GMT+03:00 Igor Sapego <[hidden email]>:
> >>>> >>>
> >>>> >>>> Vyacheslav,
> >>>> >>>>
> >>>> >>>> I had a look at your PR and left some comments in Jira.
> >>>> >>>>
> >>>> >>>> Best Regards,
> >>>> >>>> Igor
> >>>> >>>>
> >>>> >>>> On Mon, Jan 30, 2017 at 12:52 PM, Vyacheslav Daradur <
> >>>> >>>> [hidden email]>
> >>>> >>>> wrote:
> >>>> >>>>
> >>>> >>>> > Hello. I fixed it. Please, review.
> >>>> >>>> >
> >>>> >>>> > https://issues.apache.org/jira/browse/IGNITE-3196 - Marshaling
> >>>> works
> >>>> >>>> wrong
> >>>> >>>> > for the BigDecimals that have negative scale
> >>>> >>>> >
> >>>> >>>>
> >>>> >>>
> >>>> >>>
> >>>> >>
> >>>> >
> >>>>
> >>>
> >>>
> >>
> >
>
Reply | Threaded
Open this post in threaded view
|

Re: IGNITE-3196 - ready for review

Igor Sapego
Vyacheslav,

Overall looks good. But why do you use -0x80 instead of 0x80?

Best Regards,
Igor

On Mon, Feb 6, 2017 at 5:36 PM, Vyacheslav Daradur <[hidden email]>
wrote:

> Igor,
>
> I didn't change the CPP code before approval approach.
> I shall write directly, sorry.
>
> But I made CPP changes already.
>
> > TestEscConvertFunctionFloat
> > TestEscConvertFunctionDouble.
> These tests were passed
> <http://ci.ignite.apache.org/viewQueued.html?itemId=445824>
>
>
>
> 2017-02-06 13:20 GMT+03:00 Pavel Tupitsyn <[hidden email]>:
>
>> .NET changes look good to me.
>>
>> Pavel
>>
>> On Mon, Feb 6, 2017 at 1:10 PM, Igor Sapego <[hidden email]> wrote:
>>
>> > Vyacheslav, I can see two ODBC tests fail in C++ test suits that should
>> > not:
>> > - TestEscConvertFunctionFloat
>> > <http://ci.ignite.apache.org/viewLog.html?buildId=444207&tab
>> =buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformCppLi
>> nux#testNameId-9178617718508801660>
>> > - TestEscConvertFunctionDouble
>> > <http://ci.ignite.apache.org/viewLog.html?buildId=444207&tab
>> =buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformCppLi
>> nux#testNameId5432107083822590090>
>> > .
>> >
>> > I believe, this is because I can't see any changes in C++ Decimal
>> > marshaling code.
>> > Please, pay attention to file ignite\modules\platforms\cpp\
>> > odbc\src\utility.cpp,
>> > functions ReadDecimal and WriteDecimal.
>> >
>> > Best Regards,
>> > Igor
>> >
>> > On Mon, Feb 6, 2017 at 11:21 AM, Vyacheslav Daradur <
>> [hidden email]>
>> > wrote:
>> >
>> >> Pavel, Igor
>> >>
>> >> Please, review it again.
>> >>
>> >> https://github.com/apache/ignite/pull/1473/files
>> >>
>> >> All tests
>> >> <http://ci.ignite.apache.org/viewLog.html?buildId=444231&tab
>> =buildResultsDiv&buildTypeId=IgniteTests_RunAll>
>> >> .NET tests
>> >> <http://ci.ignite.apache.org/viewLog.html?buildId=443439&tab
>> =buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformNet>
>> >>
>> >> How about this solution?
>> >>
>> >> 2017-02-03 13:59 GMT+03:00 Vyacheslav Daradur <[hidden email]>:
>> >>
>> >>> 1. On my first question
>> >>> I think up, if we serialize only positive numbers, we can write sign
>> in
>> >>> first byte, because it is positive always.
>> >>> I will try to make this decision
>> >>>
>> >>> 2017-02-03 12:48 GMT+03:00 Pavel Tupitsyn <[hidden email]>:
>> >>>
>> >>>> Vyacheslav,
>> >>>>
>> >>>> I see the problem now. Yes, negative scale is not supported in .NET.
>> >>>>
>> >>>> I don't think we should do the multiplication. As you described, this
>> >>>> will
>> >>>> break equality on Java side. SQL queries might be broken, etc.
>> >>>> I think we should throw an exception in .NET when encountering
>> negative
>> >>>> decimal scale.
>> >>>>
>> >>>> Vladimir O, any thoughts?
>> >>>>
>> >>>> Pavel
>> >>>>
>> >>>> On Fri, Feb 3, 2017 at 12:01 PM, Vyacheslav Daradur <
>> >>>> [hidden email]>
>> >>>> wrote:
>> >>>>
>> >>>> > Hello.
>> >>>> >
>> >>>> > I looked and understood the code of methods ReadDecimal and
>> >>>> WriteDecimal
>> >>>> > on .NET platform.
>> >>>> >
>> >>>> > 1. At the moment remaking of this methods for my Java-decimal-fix
>> is
>> >>>> very
>> >>>> > difficult, it needs to write new methods for
>> >>>> serialization/deserialization
>> >>>> > of negative decimals.
>> >>>> >
>> >>>> > I can make it, but there is simpler decision: to add additional
>> byte
>> >>>> for
>> >>>> > sign.
>> >>>> >
>> >>>> > I need advice: difficult solution (new methods .net) Versus :
>> simple
>> >>>> > solutions (additional byte)?
>> >>>> >
>> >>>> > *I don't know yet, what changes are necessary on С++ platform.
>> >>>> >
>> >>>> > 2. I see a problem with the negative scale on .NET platform.
>> >>>> >
>> >>>> > Now negative scale is forbidden.
>> >>>> >
>> >>>> > We can make:
>> >>>> > if (scale < 0) return Decimal.Multiply(new decimal(lo, mid, hi,
>> neg,
>> >>>> 0),
>> >>>> > new decimal(Math.Pow(10, -scale)));
>> >>>> >
>> >>>> > But there is the problem:
>> >>>> > * 1 Serialize in Java; number=123456789, scale=-4
>> >>>> > * 2 Deserialize in .NET; number=1234567890000, scale=0
>> >>>> > * 3 Serialize in .NET; number=1234567890000, scale=0
>> >>>> > * 4 Deserialize in Java; number=1234567890000, scale=0
>> >>>> >
>> >>>> > Logically: (1) 123456789 * 10^4 == (2)  1234567890000
>> >>>> >
>> >>>> > In Java (1) not equal (2), because scales are different.
>> >>>> >
>> >>>> > Any thougths?
>> >>>> >
>> >>>> > 2017-01-31 14:08 GMT+03:00 Pavel Tupitsyn <[hidden email]>:
>> >>>> >
>> >>>> >> Vyacheslav,
>> >>>> >>
>> >>>> >> I'm not sure I understand the code you attached.
>> >>>> >>
>> >>>> >> If you know how to fix the .NET part, can you just do it in your
>> PR
>> >>>> and
>> >>>> >> run "Platform .NET" on TeamCity to verify?
>> >>>> >> http://ci.ignite.apache.org/viewType.html?buildTypeId=Ignite
>> >>>> >> Tests_IgnitePlatformNet
>> >>>> >>
>> >>>> >> Thanks,
>> >>>> >>
>> >>>> >> Pavel
>> >>>> >>
>> >>>> >> On Tue, Jan 31, 2017 at 1:35 PM, Vyacheslav Daradur <
>> >>>> [hidden email]>
>> >>>> >> wrote:
>> >>>> >>
>> >>>> >>> Pavel, I see that you are the main contributor of Ignite.NET.
>> >>>> >>>
>> >>>> >>> We should repair BinaryUtils#ReadDecimal and
>> >>>> BinaryUtils#WriteDecimal.
>> >>>> >>>
>> >>>> >>> *ReadDecimal:
>> >>>> >>> byte[] mag = ReadByteArray(stream); // including at least one
>> sign
>> >>>> bit,
>> >>>> >>> which is (ceil((this.bitLength() + 1)/8))
>> >>>> >>> bool neg = (mag[0] < 0);
>> >>>> >>> if (scale < 0)
>> >>>> >>> // TODO: a scale of -3 means the unscaled value is multiplied by
>> >>>> 1000
>> >>>> >>>
>> >>>> >>> *WriteDecimal:
>> >>>> >>> int sign = vals[3] < 0 ? -1 : 0;
>> >>>> >>> stream.WriteInt(sign);
>> >>>> >>>
>> >>>> >>> Can you help with this task?
>> >>>> >>>
>> >>>> >>>
>> >>>> >>> 2017-01-31 12:46 GMT+03:00 Igor Sapego <[hidden email]>:
>> >>>> >>>
>> >>>> >>>> Vyacheslav,
>> >>>> >>>>
>> >>>> >>>> I had a look at your PR and left some comments in Jira.
>> >>>> >>>>
>> >>>> >>>> Best Regards,
>> >>>> >>>> Igor
>> >>>> >>>>
>> >>>> >>>> On Mon, Jan 30, 2017 at 12:52 PM, Vyacheslav Daradur <
>> >>>> >>>> [hidden email]>
>> >>>> >>>> wrote:
>> >>>> >>>>
>> >>>> >>>> > Hello. I fixed it. Please, review.
>> >>>> >>>> >
>> >>>> >>>> > https://issues.apache.org/jira/browse/IGNITE-3196 -
>> Marshaling
>> >>>> works
>> >>>> >>>> wrong
>> >>>> >>>> > for the BigDecimals that have negative scale
>> >>>> >>>> >
>> >>>> >>>>
>> >>>> >>>
>> >>>> >>>
>> >>>> >>
>> >>>> >
>> >>>>
>> >>>
>> >>>
>> >>
>> >
>>
>
>
Reply | Threaded
Open this post in threaded view
|

Re: IGNITE-3196 - ready for review

daradurvs
Byte.MIN_VALUE = -128 = -0x80
Byte.MAX_VALUE = 127 = 0x7F

It is just more evident for me.

Maybe, I just have the Java programming style.

In Java:
byte a = 100 | -0x80;  // compiled
byte b = 100 | 0x80;  // doesn't compile, explicit type casting is
neccessary (byte)(100 | 0x80)
System.out.println(a | -0x80); // -28
System.out.println(a | 0x80); // 228 - cast to int

Is it bad style?

2017-02-06 20:04 GMT+03:00 Igor Sapego <[hidden email]>:

> Vyacheslav,
>
> Overall looks good. But why do you use -0x80 instead of 0x80?
>
> Best Regards,
> Igor
>
> On Mon, Feb 6, 2017 at 5:36 PM, Vyacheslav Daradur <[hidden email]>
> wrote:
>
>> Igor,
>>
>> I didn't change the CPP code before approval approach.
>> I shall write directly, sorry.
>>
>> But I made CPP changes already.
>>
>> > TestEscConvertFunctionFloat
>> > TestEscConvertFunctionDouble.
>> These tests were passed
>> <http://ci.ignite.apache.org/viewQueued.html?itemId=445824>
>>
>>
>>
>> 2017-02-06 13:20 GMT+03:00 Pavel Tupitsyn <[hidden email]>:
>>
>>> .NET changes look good to me.
>>>
>>> Pavel
>>>
>>> On Mon, Feb 6, 2017 at 1:10 PM, Igor Sapego <[hidden email]>
>>> wrote:
>>>
>>> > Vyacheslav, I can see two ODBC tests fail in C++ test suits that should
>>> > not:
>>> > - TestEscConvertFunctionFloat
>>> > <http://ci.ignite.apache.org/viewLog.html?buildId=444207&tab
>>> =buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformCppLi
>>> nux#testNameId-9178617718508801660>
>>> > - TestEscConvertFunctionDouble
>>> > <http://ci.ignite.apache.org/viewLog.html?buildId=444207&tab
>>> =buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformCppLi
>>> nux#testNameId5432107083822590090>
>>> > .
>>> >
>>> > I believe, this is because I can't see any changes in C++ Decimal
>>> > marshaling code.
>>> > Please, pay attention to file ignite\modules\platforms\cpp\
>>> > odbc\src\utility.cpp,
>>> > functions ReadDecimal and WriteDecimal.
>>> >
>>> > Best Regards,
>>> > Igor
>>> >
>>> > On Mon, Feb 6, 2017 at 11:21 AM, Vyacheslav Daradur <
>>> [hidden email]>
>>> > wrote:
>>> >
>>> >> Pavel, Igor
>>> >>
>>> >> Please, review it again.
>>> >>
>>> >> https://github.com/apache/ignite/pull/1473/files
>>> >>
>>> >> All tests
>>> >> <http://ci.ignite.apache.org/viewLog.html?buildId=444231&tab
>>> =buildResultsDiv&buildTypeId=IgniteTests_RunAll>
>>> >> .NET tests
>>> >> <http://ci.ignite.apache.org/viewLog.html?buildId=443439&tab
>>> =buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformNet>
>>> >>
>>> >> How about this solution?
>>> >>
>>> >> 2017-02-03 13:59 GMT+03:00 Vyacheslav Daradur <[hidden email]>:
>>> >>
>>> >>> 1. On my first question
>>> >>> I think up, if we serialize only positive numbers, we can write sign
>>> in
>>> >>> first byte, because it is positive always.
>>> >>> I will try to make this decision
>>> >>>
>>> >>> 2017-02-03 12:48 GMT+03:00 Pavel Tupitsyn <[hidden email]>:
>>> >>>
>>> >>>> Vyacheslav,
>>> >>>>
>>> >>>> I see the problem now. Yes, negative scale is not supported in .NET.
>>> >>>>
>>> >>>> I don't think we should do the multiplication. As you described,
>>> this
>>> >>>> will
>>> >>>> break equality on Java side. SQL queries might be broken, etc.
>>> >>>> I think we should throw an exception in .NET when encountering
>>> negative
>>> >>>> decimal scale.
>>> >>>>
>>> >>>> Vladimir O, any thoughts?
>>> >>>>
>>> >>>> Pavel
>>> >>>>
>>> >>>> On Fri, Feb 3, 2017 at 12:01 PM, Vyacheslav Daradur <
>>> >>>> [hidden email]>
>>> >>>> wrote:
>>> >>>>
>>> >>>> > Hello.
>>> >>>> >
>>> >>>> > I looked and understood the code of methods ReadDecimal and
>>> >>>> WriteDecimal
>>> >>>> > on .NET platform.
>>> >>>> >
>>> >>>> > 1. At the moment remaking of this methods for my Java-decimal-fix
>>> is
>>> >>>> very
>>> >>>> > difficult, it needs to write new methods for
>>> >>>> serialization/deserialization
>>> >>>> > of negative decimals.
>>> >>>> >
>>> >>>> > I can make it, but there is simpler decision: to add additional
>>> byte
>>> >>>> for
>>> >>>> > sign.
>>> >>>> >
>>> >>>> > I need advice: difficult solution (new methods .net) Versus :
>>> simple
>>> >>>> > solutions (additional byte)?
>>> >>>> >
>>> >>>> > *I don't know yet, what changes are necessary on С++ platform.
>>> >>>> >
>>> >>>> > 2. I see a problem with the negative scale on .NET platform.
>>> >>>> >
>>> >>>> > Now negative scale is forbidden.
>>> >>>> >
>>> >>>> > We can make:
>>> >>>> > if (scale < 0) return Decimal.Multiply(new decimal(lo, mid, hi,
>>> neg,
>>> >>>> 0),
>>> >>>> > new decimal(Math.Pow(10, -scale)));
>>> >>>> >
>>> >>>> > But there is the problem:
>>> >>>> > * 1 Serialize in Java; number=123456789, scale=-4
>>> >>>> > * 2 Deserialize in .NET; number=1234567890000, scale=0
>>> >>>> > * 3 Serialize in .NET; number=1234567890000, scale=0
>>> >>>> > * 4 Deserialize in Java; number=1234567890000, scale=0
>>> >>>> >
>>> >>>> > Logically: (1) 123456789 * 10^4 == (2)  1234567890000
>>> >>>> >
>>> >>>> > In Java (1) not equal (2), because scales are different.
>>> >>>> >
>>> >>>> > Any thougths?
>>> >>>> >
>>> >>>> > 2017-01-31 14:08 GMT+03:00 Pavel Tupitsyn <[hidden email]>:
>>> >>>> >
>>> >>>> >> Vyacheslav,
>>> >>>> >>
>>> >>>> >> I'm not sure I understand the code you attached.
>>> >>>> >>
>>> >>>> >> If you know how to fix the .NET part, can you just do it in your
>>> PR
>>> >>>> and
>>> >>>> >> run "Platform .NET" on TeamCity to verify?
>>> >>>> >> http://ci.ignite.apache.org/viewType.html?buildTypeId=Ignite
>>> >>>> >> Tests_IgnitePlatformNet
>>> >>>> >>
>>> >>>> >> Thanks,
>>> >>>> >>
>>> >>>> >> Pavel
>>> >>>> >>
>>> >>>> >> On Tue, Jan 31, 2017 at 1:35 PM, Vyacheslav Daradur <
>>> >>>> [hidden email]>
>>> >>>> >> wrote:
>>> >>>> >>
>>> >>>> >>> Pavel, I see that you are the main contributor of Ignite.NET.
>>> >>>> >>>
>>> >>>> >>> We should repair BinaryUtils#ReadDecimal and
>>> >>>> BinaryUtils#WriteDecimal.
>>> >>>> >>>
>>> >>>> >>> *ReadDecimal:
>>> >>>> >>> byte[] mag = ReadByteArray(stream); // including at least one
>>> sign
>>> >>>> bit,
>>> >>>> >>> which is (ceil((this.bitLength() + 1)/8))
>>> >>>> >>> bool neg = (mag[0] < 0);
>>> >>>> >>> if (scale < 0)
>>> >>>> >>> // TODO: a scale of -3 means the unscaled value is multiplied by
>>> >>>> 1000
>>> >>>> >>>
>>> >>>> >>> *WriteDecimal:
>>> >>>> >>> int sign = vals[3] < 0 ? -1 : 0;
>>> >>>> >>> stream.WriteInt(sign);
>>> >>>> >>>
>>> >>>> >>> Can you help with this task?
>>> >>>> >>>
>>> >>>> >>>
>>> >>>> >>> 2017-01-31 12:46 GMT+03:00 Igor Sapego <[hidden email]>:
>>> >>>> >>>
>>> >>>> >>>> Vyacheslav,
>>> >>>> >>>>
>>> >>>> >>>> I had a look at your PR and left some comments in Jira.
>>> >>>> >>>>
>>> >>>> >>>> Best Regards,
>>> >>>> >>>> Igor
>>> >>>> >>>>
>>> >>>> >>>> On Mon, Jan 30, 2017 at 12:52 PM, Vyacheslav Daradur <
>>> >>>> >>>> [hidden email]>
>>> >>>> >>>> wrote:
>>> >>>> >>>>
>>> >>>> >>>> > Hello. I fixed it. Please, review.
>>> >>>> >>>> >
>>> >>>> >>>> > https://issues.apache.org/jira/browse/IGNITE-3196 -
>>> Marshaling
>>> >>>> works
>>> >>>> >>>> wrong
>>> >>>> >>>> > for the BigDecimals that have negative scale
>>> >>>> >>>> >
>>> >>>> >>>>
>>> >>>> >>>
>>> >>>> >>>
>>> >>>> >>
>>> >>>> >
>>> >>>>
>>> >>>
>>> >>>
>>> >>
>>> >
>>>
>>
>>
>
Reply | Threaded
Open this post in threaded view
|

Re: IGNITE-3196 - ready for review

Igor Sapego
I was just curious.

In C++ both constants 0x80 and -0x80 are of type 'int' and have the same
lower byte, so they give the same result. Though first number is actually
0x00000080 when the second one is 0xFFFFFF80.

So it's just made a minus sign look a little redundant and pointless to me
in C++ code.

Best Regards,
Igor

On Mon, Feb 6, 2017 at 10:15 PM, Vyacheslav Daradur <[hidden email]>
wrote:

> Byte.MIN_VALUE = -128 = -0x80
> Byte.MAX_VALUE = 127 = 0x7F
>
> It is just more evident for me.
>
> Maybe, I just have the Java programming style.
>
> In Java:
> byte a = 100 | -0x80;  // compiled
> byte b = 100 | 0x80;  // doesn't compile, explicit type casting is
> neccessary (byte)(100 | 0x80)
> System.out.println(a | -0x80); // -28
> System.out.println(a | 0x80); // 228 - cast to int
>
> Is it bad style?
>
> 2017-02-06 20:04 GMT+03:00 Igor Sapego <[hidden email]>:
>
>> Vyacheslav,
>>
>> Overall looks good. But why do you use -0x80 instead of 0x80?
>>
>> Best Regards,
>> Igor
>>
>> On Mon, Feb 6, 2017 at 5:36 PM, Vyacheslav Daradur <[hidden email]>
>> wrote:
>>
>>> Igor,
>>>
>>> I didn't change the CPP code before approval approach.
>>> I shall write directly, sorry.
>>>
>>> But I made CPP changes already.
>>>
>>> > TestEscConvertFunctionFloat
>>> > TestEscConvertFunctionDouble.
>>> These tests were passed
>>> <http://ci.ignite.apache.org/viewQueued.html?itemId=445824>
>>>
>>>
>>>
>>> 2017-02-06 13:20 GMT+03:00 Pavel Tupitsyn <[hidden email]>:
>>>
>>>> .NET changes look good to me.
>>>>
>>>> Pavel
>>>>
>>>> On Mon, Feb 6, 2017 at 1:10 PM, Igor Sapego <[hidden email]>
>>>> wrote:
>>>>
>>>> > Vyacheslav, I can see two ODBC tests fail in C++ test suits that
>>>> should
>>>> > not:
>>>> > - TestEscConvertFunctionFloat
>>>> > <http://ci.ignite.apache.org/viewLog.html?buildId=444207&tab
>>>> =buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformCppLi
>>>> nux#testNameId-9178617718508801660>
>>>> > - TestEscConvertFunctionDouble
>>>> > <http://ci.ignite.apache.org/viewLog.html?buildId=444207&tab
>>>> =buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformCppLi
>>>> nux#testNameId5432107083822590090>
>>>> > .
>>>> >
>>>> > I believe, this is because I can't see any changes in C++ Decimal
>>>> > marshaling code.
>>>> > Please, pay attention to file ignite\modules\platforms\cpp\
>>>> > odbc\src\utility.cpp,
>>>> > functions ReadDecimal and WriteDecimal.
>>>> >
>>>> > Best Regards,
>>>> > Igor
>>>> >
>>>> > On Mon, Feb 6, 2017 at 11:21 AM, Vyacheslav Daradur <
>>>> [hidden email]>
>>>> > wrote:
>>>> >
>>>> >> Pavel, Igor
>>>> >>
>>>> >> Please, review it again.
>>>> >>
>>>> >> https://github.com/apache/ignite/pull/1473/files
>>>> >>
>>>> >> All tests
>>>> >> <http://ci.ignite.apache.org/viewLog.html?buildId=444231&tab
>>>> =buildResultsDiv&buildTypeId=IgniteTests_RunAll>
>>>> >> .NET tests
>>>> >> <http://ci.ignite.apache.org/viewLog.html?buildId=443439&tab
>>>> =buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformNet>
>>>> >>
>>>> >> How about this solution?
>>>> >>
>>>> >> 2017-02-03 13:59 GMT+03:00 Vyacheslav Daradur <[hidden email]>:
>>>> >>
>>>> >>> 1. On my first question
>>>> >>> I think up, if we serialize only positive numbers, we can write
>>>> sign in
>>>> >>> first byte, because it is positive always.
>>>> >>> I will try to make this decision
>>>> >>>
>>>> >>> 2017-02-03 12:48 GMT+03:00 Pavel Tupitsyn <[hidden email]>:
>>>> >>>
>>>> >>>> Vyacheslav,
>>>> >>>>
>>>> >>>> I see the problem now. Yes, negative scale is not supported in
>>>> .NET.
>>>> >>>>
>>>> >>>> I don't think we should do the multiplication. As you described,
>>>> this
>>>> >>>> will
>>>> >>>> break equality on Java side. SQL queries might be broken, etc.
>>>> >>>> I think we should throw an exception in .NET when encountering
>>>> negative
>>>> >>>> decimal scale.
>>>> >>>>
>>>> >>>> Vladimir O, any thoughts?
>>>> >>>>
>>>> >>>> Pavel
>>>> >>>>
>>>> >>>> On Fri, Feb 3, 2017 at 12:01 PM, Vyacheslav Daradur <
>>>> >>>> [hidden email]>
>>>> >>>> wrote:
>>>> >>>>
>>>> >>>> > Hello.
>>>> >>>> >
>>>> >>>> > I looked and understood the code of methods ReadDecimal and
>>>> >>>> WriteDecimal
>>>> >>>> > on .NET platform.
>>>> >>>> >
>>>> >>>> > 1. At the moment remaking of this methods for my
>>>> Java-decimal-fix is
>>>> >>>> very
>>>> >>>> > difficult, it needs to write new methods for
>>>> >>>> serialization/deserialization
>>>> >>>> > of negative decimals.
>>>> >>>> >
>>>> >>>> > I can make it, but there is simpler decision: to add additional
>>>> byte
>>>> >>>> for
>>>> >>>> > sign.
>>>> >>>> >
>>>> >>>> > I need advice: difficult solution (new methods .net) Versus :
>>>> simple
>>>> >>>> > solutions (additional byte)?
>>>> >>>> >
>>>> >>>> > *I don't know yet, what changes are necessary on С++ platform.
>>>> >>>> >
>>>> >>>> > 2. I see a problem with the negative scale on .NET platform.
>>>> >>>> >
>>>> >>>> > Now negative scale is forbidden.
>>>> >>>> >
>>>> >>>> > We can make:
>>>> >>>> > if (scale < 0) return Decimal.Multiply(new decimal(lo, mid, hi,
>>>> neg,
>>>> >>>> 0),
>>>> >>>> > new decimal(Math.Pow(10, -scale)));
>>>> >>>> >
>>>> >>>> > But there is the problem:
>>>> >>>> > * 1 Serialize in Java; number=123456789, scale=-4
>>>> >>>> > * 2 Deserialize in .NET; number=1234567890000, scale=0
>>>> >>>> > * 3 Serialize in .NET; number=1234567890000, scale=0
>>>> >>>> > * 4 Deserialize in Java; number=1234567890000, scale=0
>>>> >>>> >
>>>> >>>> > Logically: (1) 123456789 * 10^4 == (2)  1234567890000
>>>> >>>> >
>>>> >>>> > In Java (1) not equal (2), because scales are different.
>>>> >>>> >
>>>> >>>> > Any thougths?
>>>> >>>> >
>>>> >>>> > 2017-01-31 14:08 GMT+03:00 Pavel Tupitsyn <[hidden email]
>>>> >:
>>>> >>>> >
>>>> >>>> >> Vyacheslav,
>>>> >>>> >>
>>>> >>>> >> I'm not sure I understand the code you attached.
>>>> >>>> >>
>>>> >>>> >> If you know how to fix the .NET part, can you just do it in
>>>> your PR
>>>> >>>> and
>>>> >>>> >> run "Platform .NET" on TeamCity to verify?
>>>> >>>> >> http://ci.ignite.apache.org/viewType.html?buildTypeId=Ignite
>>>> >>>> >> Tests_IgnitePlatformNet
>>>> >>>> >>
>>>> >>>> >> Thanks,
>>>> >>>> >>
>>>> >>>> >> Pavel
>>>> >>>> >>
>>>> >>>> >> On Tue, Jan 31, 2017 at 1:35 PM, Vyacheslav Daradur <
>>>> >>>> [hidden email]>
>>>> >>>> >> wrote:
>>>> >>>> >>
>>>> >>>> >>> Pavel, I see that you are the main contributor of Ignite.NET.
>>>> >>>> >>>
>>>> >>>> >>> We should repair BinaryUtils#ReadDecimal and
>>>> >>>> BinaryUtils#WriteDecimal.
>>>> >>>> >>>
>>>> >>>> >>> *ReadDecimal:
>>>> >>>> >>> byte[] mag = ReadByteArray(stream); // including at least one
>>>> sign
>>>> >>>> bit,
>>>> >>>> >>> which is (ceil((this.bitLength() + 1)/8))
>>>> >>>> >>> bool neg = (mag[0] < 0);
>>>> >>>> >>> if (scale < 0)
>>>> >>>> >>> // TODO: a scale of -3 means the unscaled value is multiplied
>>>> by
>>>> >>>> 1000
>>>> >>>> >>>
>>>> >>>> >>> *WriteDecimal:
>>>> >>>> >>> int sign = vals[3] < 0 ? -1 : 0;
>>>> >>>> >>> stream.WriteInt(sign);
>>>> >>>> >>>
>>>> >>>> >>> Can you help with this task?
>>>> >>>> >>>
>>>> >>>> >>>
>>>> >>>> >>> 2017-01-31 12:46 GMT+03:00 Igor Sapego <[hidden email]>:
>>>> >>>> >>>
>>>> >>>> >>>> Vyacheslav,
>>>> >>>> >>>>
>>>> >>>> >>>> I had a look at your PR and left some comments in Jira.
>>>> >>>> >>>>
>>>> >>>> >>>> Best Regards,
>>>> >>>> >>>> Igor
>>>> >>>> >>>>
>>>> >>>> >>>> On Mon, Jan 30, 2017 at 12:52 PM, Vyacheslav Daradur <
>>>> >>>> >>>> [hidden email]>
>>>> >>>> >>>> wrote:
>>>> >>>> >>>>
>>>> >>>> >>>> > Hello. I fixed it. Please, review.
>>>> >>>> >>>> >
>>>> >>>> >>>> > https://issues.apache.org/jira/browse/IGNITE-3196 -
>>>> Marshaling
>>>> >>>> works
>>>> >>>> >>>> wrong
>>>> >>>> >>>> > for the BigDecimals that have negative scale
>>>> >>>> >>>> >
>>>> >>>> >>>>
>>>> >>>> >>>
>>>> >>>> >>>
>>>> >>>> >>
>>>> >>>> >
>>>> >>>>
>>>> >>>
>>>> >>>
>>>> >>
>>>> >
>>>>
>>>
>>>
>>
>
Reply | Threaded
Open this post in threaded view
|

Re: IGNITE-3196 - ready for review

Igor Sapego
But that's Ok. Since we use int8_t for bytes in C++ as well I guess
your -0x80 may have more sense than 0x80.

Best Regards,
Igor

On Tue, Feb 7, 2017 at 1:54 PM, Igor Sapego <[hidden email]> wrote:

> I was just curious.
>
> In C++ both constants 0x80 and -0x80 are of type 'int' and have the same
> lower byte, so they give the same result. Though first number is actually
> 0x00000080 when the second one is 0xFFFFFF80.
>
> So it's just made a minus sign look a little redundant and pointless to me
> in C++ code.
>
> Best Regards,
> Igor
>
> On Mon, Feb 6, 2017 at 10:15 PM, Vyacheslav Daradur <[hidden email]>
> wrote:
>
>> Byte.MIN_VALUE = -128 = -0x80
>> Byte.MAX_VALUE = 127 = 0x7F
>>
>> It is just more evident for me.
>>
>> Maybe, I just have the Java programming style.
>>
>> In Java:
>> byte a = 100 | -0x80;  // compiled
>> byte b = 100 | 0x80;  // doesn't compile, explicit type casting is
>> neccessary (byte)(100 | 0x80)
>> System.out.println(a | -0x80); // -28
>> System.out.println(a | 0x80); // 228 - cast to int
>>
>> Is it bad style?
>>
>> 2017-02-06 20:04 GMT+03:00 Igor Sapego <[hidden email]>:
>>
>>> Vyacheslav,
>>>
>>> Overall looks good. But why do you use -0x80 instead of 0x80?
>>>
>>> Best Regards,
>>> Igor
>>>
>>> On Mon, Feb 6, 2017 at 5:36 PM, Vyacheslav Daradur <[hidden email]>
>>> wrote:
>>>
>>>> Igor,
>>>>
>>>> I didn't change the CPP code before approval approach.
>>>> I shall write directly, sorry.
>>>>
>>>> But I made CPP changes already.
>>>>
>>>> > TestEscConvertFunctionFloat
>>>> > TestEscConvertFunctionDouble.
>>>> These tests were passed
>>>> <http://ci.ignite.apache.org/viewQueued.html?itemId=445824>
>>>>
>>>>
>>>>
>>>> 2017-02-06 13:20 GMT+03:00 Pavel Tupitsyn <[hidden email]>:
>>>>
>>>>> .NET changes look good to me.
>>>>>
>>>>> Pavel
>>>>>
>>>>> On Mon, Feb 6, 2017 at 1:10 PM, Igor Sapego <[hidden email]>
>>>>> wrote:
>>>>>
>>>>> > Vyacheslav, I can see two ODBC tests fail in C++ test suits that
>>>>> should
>>>>> > not:
>>>>> > - TestEscConvertFunctionFloat
>>>>> > <http://ci.ignite.apache.org/viewLog.html?buildId=444207&tab
>>>>> =buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformCppLi
>>>>> nux#testNameId-9178617718508801660>
>>>>> > - TestEscConvertFunctionDouble
>>>>> > <http://ci.ignite.apache.org/viewLog.html?buildId=444207&tab
>>>>> =buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformCppLi
>>>>> nux#testNameId5432107083822590090>
>>>>> > .
>>>>> >
>>>>> > I believe, this is because I can't see any changes in C++ Decimal
>>>>> > marshaling code.
>>>>> > Please, pay attention to file ignite\modules\platforms\cpp\
>>>>> > odbc\src\utility.cpp,
>>>>> > functions ReadDecimal and WriteDecimal.
>>>>> >
>>>>> > Best Regards,
>>>>> > Igor
>>>>> >
>>>>> > On Mon, Feb 6, 2017 at 11:21 AM, Vyacheslav Daradur <
>>>>> [hidden email]>
>>>>> > wrote:
>>>>> >
>>>>> >> Pavel, Igor
>>>>> >>
>>>>> >> Please, review it again.
>>>>> >>
>>>>> >> https://github.com/apache/ignite/pull/1473/files
>>>>> >>
>>>>> >> All tests
>>>>> >> <http://ci.ignite.apache.org/viewLog.html?buildId=444231&tab
>>>>> =buildResultsDiv&buildTypeId=IgniteTests_RunAll>
>>>>> >> .NET tests
>>>>> >> <http://ci.ignite.apache.org/viewLog.html?buildId=443439&tab
>>>>> =buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformNet>
>>>>> >>
>>>>> >> How about this solution?
>>>>> >>
>>>>> >> 2017-02-03 13:59 GMT+03:00 Vyacheslav Daradur <[hidden email]
>>>>> >:
>>>>> >>
>>>>> >>> 1. On my first question
>>>>> >>> I think up, if we serialize only positive numbers, we can write
>>>>> sign in
>>>>> >>> first byte, because it is positive always.
>>>>> >>> I will try to make this decision
>>>>> >>>
>>>>> >>> 2017-02-03 12:48 GMT+03:00 Pavel Tupitsyn <[hidden email]>:
>>>>> >>>
>>>>> >>>> Vyacheslav,
>>>>> >>>>
>>>>> >>>> I see the problem now. Yes, negative scale is not supported in
>>>>> .NET.
>>>>> >>>>
>>>>> >>>> I don't think we should do the multiplication. As you described,
>>>>> this
>>>>> >>>> will
>>>>> >>>> break equality on Java side. SQL queries might be broken, etc.
>>>>> >>>> I think we should throw an exception in .NET when encountering
>>>>> negative
>>>>> >>>> decimal scale.
>>>>> >>>>
>>>>> >>>> Vladimir O, any thoughts?
>>>>> >>>>
>>>>> >>>> Pavel
>>>>> >>>>
>>>>> >>>> On Fri, Feb 3, 2017 at 12:01 PM, Vyacheslav Daradur <
>>>>> >>>> [hidden email]>
>>>>> >>>> wrote:
>>>>> >>>>
>>>>> >>>> > Hello.
>>>>> >>>> >
>>>>> >>>> > I looked and understood the code of methods ReadDecimal and
>>>>> >>>> WriteDecimal
>>>>> >>>> > on .NET platform.
>>>>> >>>> >
>>>>> >>>> > 1. At the moment remaking of this methods for my
>>>>> Java-decimal-fix is
>>>>> >>>> very
>>>>> >>>> > difficult, it needs to write new methods for
>>>>> >>>> serialization/deserialization
>>>>> >>>> > of negative decimals.
>>>>> >>>> >
>>>>> >>>> > I can make it, but there is simpler decision: to add additional
>>>>> byte
>>>>> >>>> for
>>>>> >>>> > sign.
>>>>> >>>> >
>>>>> >>>> > I need advice: difficult solution (new methods .net) Versus :
>>>>> simple
>>>>> >>>> > solutions (additional byte)?
>>>>> >>>> >
>>>>> >>>> > *I don't know yet, what changes are necessary on С++ platform.
>>>>> >>>> >
>>>>> >>>> > 2. I see a problem with the negative scale on .NET platform.
>>>>> >>>> >
>>>>> >>>> > Now negative scale is forbidden.
>>>>> >>>> >
>>>>> >>>> > We can make:
>>>>> >>>> > if (scale < 0) return Decimal.Multiply(new decimal(lo, mid, hi,
>>>>> neg,
>>>>> >>>> 0),
>>>>> >>>> > new decimal(Math.Pow(10, -scale)));
>>>>> >>>> >
>>>>> >>>> > But there is the problem:
>>>>> >>>> > * 1 Serialize in Java; number=123456789, scale=-4
>>>>> >>>> > * 2 Deserialize in .NET; number=1234567890000, scale=0
>>>>> >>>> > * 3 Serialize in .NET; number=1234567890000, scale=0
>>>>> >>>> > * 4 Deserialize in Java; number=1234567890000, scale=0
>>>>> >>>> >
>>>>> >>>> > Logically: (1) 123456789 * 10^4 == (2)  1234567890000
>>>>> >>>> >
>>>>> >>>> > In Java (1) not equal (2), because scales are different.
>>>>> >>>> >
>>>>> >>>> > Any thougths?
>>>>> >>>> >
>>>>> >>>> > 2017-01-31 14:08 GMT+03:00 Pavel Tupitsyn <[hidden email]
>>>>> >:
>>>>> >>>> >
>>>>> >>>> >> Vyacheslav,
>>>>> >>>> >>
>>>>> >>>> >> I'm not sure I understand the code you attached.
>>>>> >>>> >>
>>>>> >>>> >> If you know how to fix the .NET part, can you just do it in
>>>>> your PR
>>>>> >>>> and
>>>>> >>>> >> run "Platform .NET" on TeamCity to verify?
>>>>> >>>> >> http://ci.ignite.apache.org/viewType.html?buildTypeId=Ignite
>>>>> >>>> >> Tests_IgnitePlatformNet
>>>>> >>>> >>
>>>>> >>>> >> Thanks,
>>>>> >>>> >>
>>>>> >>>> >> Pavel
>>>>> >>>> >>
>>>>> >>>> >> On Tue, Jan 31, 2017 at 1:35 PM, Vyacheslav Daradur <
>>>>> >>>> [hidden email]>
>>>>> >>>> >> wrote:
>>>>> >>>> >>
>>>>> >>>> >>> Pavel, I see that you are the main contributor of Ignite.NET.
>>>>> >>>> >>>
>>>>> >>>> >>> We should repair BinaryUtils#ReadDecimal and
>>>>> >>>> BinaryUtils#WriteDecimal.
>>>>> >>>> >>>
>>>>> >>>> >>> *ReadDecimal:
>>>>> >>>> >>> byte[] mag = ReadByteArray(stream); // including at least one
>>>>> sign
>>>>> >>>> bit,
>>>>> >>>> >>> which is (ceil((this.bitLength() + 1)/8))
>>>>> >>>> >>> bool neg = (mag[0] < 0);
>>>>> >>>> >>> if (scale < 0)
>>>>> >>>> >>> // TODO: a scale of -3 means the unscaled value is multiplied
>>>>> by
>>>>> >>>> 1000
>>>>> >>>> >>>
>>>>> >>>> >>> *WriteDecimal:
>>>>> >>>> >>> int sign = vals[3] < 0 ? -1 : 0;
>>>>> >>>> >>> stream.WriteInt(sign);
>>>>> >>>> >>>
>>>>> >>>> >>> Can you help with this task?
>>>>> >>>> >>>
>>>>> >>>> >>>
>>>>> >>>> >>> 2017-01-31 12:46 GMT+03:00 Igor Sapego <[hidden email]
>>>>> >:
>>>>> >>>> >>>
>>>>> >>>> >>>> Vyacheslav,
>>>>> >>>> >>>>
>>>>> >>>> >>>> I had a look at your PR and left some comments in Jira.
>>>>> >>>> >>>>
>>>>> >>>> >>>> Best Regards,
>>>>> >>>> >>>> Igor
>>>>> >>>> >>>>
>>>>> >>>> >>>> On Mon, Jan 30, 2017 at 12:52 PM, Vyacheslav Daradur <
>>>>> >>>> >>>> [hidden email]>
>>>>> >>>> >>>> wrote:
>>>>> >>>> >>>>
>>>>> >>>> >>>> > Hello. I fixed it. Please, review.
>>>>> >>>> >>>> >
>>>>> >>>> >>>> > https://issues.apache.org/jira/browse/IGNITE-3196 -
>>>>> Marshaling
>>>>> >>>> works
>>>>> >>>> >>>> wrong
>>>>> >>>> >>>> > for the BigDecimals that have negative scale
>>>>> >>>> >>>> >
>>>>> >>>> >>>>
>>>>> >>>> >>>
>>>>> >>>> >>>
>>>>> >>>> >>
>>>>> >>>> >
>>>>> >>>>
>>>>> >>>
>>>>> >>>
>>>>> >>
>>>>> >
>>>>>
>>>>
>>>>
>>>
>>
>
Reply | Threaded
Open this post in threaded view
|

Re: IGNITE-3196 - ready for review

daradurvs
Ok, thanks for explanations.

What about this task?

2017-02-07 13:57 GMT+03:00 Igor Sapego <[hidden email]>:

> But that's Ok. Since we use int8_t for bytes in C++ as well I guess
> your -0x80 may have more sense than 0x80.
>
> Best Regards,
> Igor
>
> On Tue, Feb 7, 2017 at 1:54 PM, Igor Sapego <[hidden email]> wrote:
>
>> I was just curious.
>>
>> In C++ both constants 0x80 and -0x80 are of type 'int' and have the same
>> lower byte, so they give the same result. Though first number is actually
>> 0x00000080 when the second one is 0xFFFFFF80.
>>
>> So it's just made a minus sign look a little redundant and pointless to me
>> in C++ code.
>>
>> Best Regards,
>> Igor
>>
>> On Mon, Feb 6, 2017 at 10:15 PM, Vyacheslav Daradur <[hidden email]>
>> wrote:
>>
>>> Byte.MIN_VALUE = -128 = -0x80
>>> Byte.MAX_VALUE = 127 = 0x7F
>>>
>>> It is just more evident for me.
>>>
>>> Maybe, I just have the Java programming style.
>>>
>>> In Java:
>>> byte a = 100 | -0x80;  // compiled
>>> byte b = 100 | 0x80;  // doesn't compile, explicit type casting is
>>> neccessary (byte)(100 | 0x80)
>>> System.out.println(a | -0x80); // -28
>>> System.out.println(a | 0x80); // 228 - cast to int
>>>
>>> Is it bad style?
>>>
>>> 2017-02-06 20:04 GMT+03:00 Igor Sapego <[hidden email]>:
>>>
>>>> Vyacheslav,
>>>>
>>>> Overall looks good. But why do you use -0x80 instead of 0x80?
>>>>
>>>> Best Regards,
>>>> Igor
>>>>
>>>> On Mon, Feb 6, 2017 at 5:36 PM, Vyacheslav Daradur <[hidden email]
>>>> > wrote:
>>>>
>>>>> Igor,
>>>>>
>>>>> I didn't change the CPP code before approval approach.
>>>>> I shall write directly, sorry.
>>>>>
>>>>> But I made CPP changes already.
>>>>>
>>>>> > TestEscConvertFunctionFloat
>>>>> > TestEscConvertFunctionDouble.
>>>>> These tests were passed
>>>>> <http://ci.ignite.apache.org/viewQueued.html?itemId=445824>
>>>>>
>>>>>
>>>>>
>>>>> 2017-02-06 13:20 GMT+03:00 Pavel Tupitsyn <[hidden email]>:
>>>>>
>>>>>> .NET changes look good to me.
>>>>>>
>>>>>> Pavel
>>>>>>
>>>>>> On Mon, Feb 6, 2017 at 1:10 PM, Igor Sapego <[hidden email]>
>>>>>> wrote:
>>>>>>
>>>>>> > Vyacheslav, I can see two ODBC tests fail in C++ test suits that
>>>>>> should
>>>>>> > not:
>>>>>> > - TestEscConvertFunctionFloat
>>>>>> > <http://ci.ignite.apache.org/viewLog.html?buildId=444207&tab
>>>>>> =buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformCppLi
>>>>>> nux#testNameId-9178617718508801660>
>>>>>> > - TestEscConvertFunctionDouble
>>>>>> > <http://ci.ignite.apache.org/viewLog.html?buildId=444207&tab
>>>>>> =buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformCppLi
>>>>>> nux#testNameId5432107083822590090>
>>>>>> > .
>>>>>> >
>>>>>> > I believe, this is because I can't see any changes in C++ Decimal
>>>>>> > marshaling code.
>>>>>> > Please, pay attention to file ignite\modules\platforms\cpp\
>>>>>> > odbc\src\utility.cpp,
>>>>>> > functions ReadDecimal and WriteDecimal.
>>>>>> >
>>>>>> > Best Regards,
>>>>>> > Igor
>>>>>> >
>>>>>> > On Mon, Feb 6, 2017 at 11:21 AM, Vyacheslav Daradur <
>>>>>> [hidden email]>
>>>>>> > wrote:
>>>>>> >
>>>>>> >> Pavel, Igor
>>>>>> >>
>>>>>> >> Please, review it again.
>>>>>> >>
>>>>>> >> https://github.com/apache/ignite/pull/1473/files
>>>>>> >>
>>>>>> >> All tests
>>>>>> >> <http://ci.ignite.apache.org/viewLog.html?buildId=444231&tab
>>>>>> =buildResultsDiv&buildTypeId=IgniteTests_RunAll>
>>>>>> >> .NET tests
>>>>>> >> <http://ci.ignite.apache.org/viewLog.html?buildId=443439&tab
>>>>>> =buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformNet>
>>>>>> >>
>>>>>> >> How about this solution?
>>>>>> >>
>>>>>> >> 2017-02-03 13:59 GMT+03:00 Vyacheslav Daradur <[hidden email]
>>>>>> >:
>>>>>> >>
>>>>>> >>> 1. On my first question
>>>>>> >>> I think up, if we serialize only positive numbers, we can write
>>>>>> sign in
>>>>>> >>> first byte, because it is positive always.
>>>>>> >>> I will try to make this decision
>>>>>> >>>
>>>>>> >>> 2017-02-03 12:48 GMT+03:00 Pavel Tupitsyn <[hidden email]>:
>>>>>> >>>
>>>>>> >>>> Vyacheslav,
>>>>>> >>>>
>>>>>> >>>> I see the problem now. Yes, negative scale is not supported in
>>>>>> .NET.
>>>>>> >>>>
>>>>>> >>>> I don't think we should do the multiplication. As you described,
>>>>>> this
>>>>>> >>>> will
>>>>>> >>>> break equality on Java side. SQL queries might be broken, etc.
>>>>>> >>>> I think we should throw an exception in .NET when encountering
>>>>>> negative
>>>>>> >>>> decimal scale.
>>>>>> >>>>
>>>>>> >>>> Vladimir O, any thoughts?
>>>>>> >>>>
>>>>>> >>>> Pavel
>>>>>> >>>>
>>>>>> >>>> On Fri, Feb 3, 2017 at 12:01 PM, Vyacheslav Daradur <
>>>>>> >>>> [hidden email]>
>>>>>> >>>> wrote:
>>>>>> >>>>
>>>>>> >>>> > Hello.
>>>>>> >>>> >
>>>>>> >>>> > I looked and understood the code of methods ReadDecimal and
>>>>>> >>>> WriteDecimal
>>>>>> >>>> > on .NET platform.
>>>>>> >>>> >
>>>>>> >>>> > 1. At the moment remaking of this methods for my
>>>>>> Java-decimal-fix is
>>>>>> >>>> very
>>>>>> >>>> > difficult, it needs to write new methods for
>>>>>> >>>> serialization/deserialization
>>>>>> >>>> > of negative decimals.
>>>>>> >>>> >
>>>>>> >>>> > I can make it, but there is simpler decision: to add
>>>>>> additional byte
>>>>>> >>>> for
>>>>>> >>>> > sign.
>>>>>> >>>> >
>>>>>> >>>> > I need advice: difficult solution (new methods .net) Versus :
>>>>>> simple
>>>>>> >>>> > solutions (additional byte)?
>>>>>> >>>> >
>>>>>> >>>> > *I don't know yet, what changes are necessary on С++ platform.
>>>>>> >>>> >
>>>>>> >>>> > 2. I see a problem with the negative scale on .NET platform.
>>>>>> >>>> >
>>>>>> >>>> > Now negative scale is forbidden.
>>>>>> >>>> >
>>>>>> >>>> > We can make:
>>>>>> >>>> > if (scale < 0) return Decimal.Multiply(new decimal(lo, mid,
>>>>>> hi, neg,
>>>>>> >>>> 0),
>>>>>> >>>> > new decimal(Math.Pow(10, -scale)));
>>>>>> >>>> >
>>>>>> >>>> > But there is the problem:
>>>>>> >>>> > * 1 Serialize in Java; number=123456789, scale=-4
>>>>>> >>>> > * 2 Deserialize in .NET; number=1234567890000, scale=0
>>>>>> >>>> > * 3 Serialize in .NET; number=1234567890000, scale=0
>>>>>> >>>> > * 4 Deserialize in Java; number=1234567890000, scale=0
>>>>>> >>>> >
>>>>>> >>>> > Logically: (1) 123456789 * 10^4 == (2)  1234567890000
>>>>>> >>>> >
>>>>>> >>>> > In Java (1) not equal (2), because scales are different.
>>>>>> >>>> >
>>>>>> >>>> > Any thougths?
>>>>>> >>>> >
>>>>>> >>>> > 2017-01-31 14:08 GMT+03:00 Pavel Tupitsyn <
>>>>>> [hidden email]>:
>>>>>> >>>> >
>>>>>> >>>> >> Vyacheslav,
>>>>>> >>>> >>
>>>>>> >>>> >> I'm not sure I understand the code you attached.
>>>>>> >>>> >>
>>>>>> >>>> >> If you know how to fix the .NET part, can you just do it in
>>>>>> your PR
>>>>>> >>>> and
>>>>>> >>>> >> run "Platform .NET" on TeamCity to verify?
>>>>>> >>>> >> http://ci.ignite.apache.org/viewType.html?buildTypeId=Ignite
>>>>>> >>>> >> Tests_IgnitePlatformNet
>>>>>> >>>> >>
>>>>>> >>>> >> Thanks,
>>>>>> >>>> >>
>>>>>> >>>> >> Pavel
>>>>>> >>>> >>
>>>>>> >>>> >> On Tue, Jan 31, 2017 at 1:35 PM, Vyacheslav Daradur <
>>>>>> >>>> [hidden email]>
>>>>>> >>>> >> wrote:
>>>>>> >>>> >>
>>>>>> >>>> >>> Pavel, I see that you are the main contributor of Ignite.NET.
>>>>>> >>>> >>>
>>>>>> >>>> >>> We should repair BinaryUtils#ReadDecimal and
>>>>>> >>>> BinaryUtils#WriteDecimal.
>>>>>> >>>> >>>
>>>>>> >>>> >>> *ReadDecimal:
>>>>>> >>>> >>> byte[] mag = ReadByteArray(stream); // including at least
>>>>>> one sign
>>>>>> >>>> bit,
>>>>>> >>>> >>> which is (ceil((this.bitLength() + 1)/8))
>>>>>> >>>> >>> bool neg = (mag[0] < 0);
>>>>>> >>>> >>> if (scale < 0)
>>>>>> >>>> >>> // TODO: a scale of -3 means the unscaled value is
>>>>>> multiplied by
>>>>>> >>>> 1000
>>>>>> >>>> >>>
>>>>>> >>>> >>> *WriteDecimal:
>>>>>> >>>> >>> int sign = vals[3] < 0 ? -1 : 0;
>>>>>> >>>> >>> stream.WriteInt(sign);
>>>>>> >>>> >>>
>>>>>> >>>> >>> Can you help with this task?
>>>>>> >>>> >>>
>>>>>> >>>> >>>
>>>>>> >>>> >>> 2017-01-31 12:46 GMT+03:00 Igor Sapego <[hidden email]
>>>>>> >:
>>>>>> >>>> >>>
>>>>>> >>>> >>>> Vyacheslav,
>>>>>> >>>> >>>>
>>>>>> >>>> >>>> I had a look at your PR and left some comments in Jira.
>>>>>> >>>> >>>>
>>>>>> >>>> >>>> Best Regards,
>>>>>> >>>> >>>> Igor
>>>>>> >>>> >>>>
>>>>>> >>>> >>>> On Mon, Jan 30, 2017 at 12:52 PM, Vyacheslav Daradur <
>>>>>> >>>> >>>> [hidden email]>
>>>>>> >>>> >>>> wrote:
>>>>>> >>>> >>>>
>>>>>> >>>> >>>> > Hello. I fixed it. Please, review.
>>>>>> >>>> >>>> >
>>>>>> >>>> >>>> > https://issues.apache.org/jira/browse/IGNITE-3196 -
>>>>>> Marshaling
>>>>>> >>>> works
>>>>>> >>>> >>>> wrong
>>>>>> >>>> >>>> > for the BigDecimals that have negative scale
>>>>>> >>>> >>>> >
>>>>>> >>>> >>>>
>>>>>> >>>> >>>
>>>>>> >>>> >>>
>>>>>> >>>> >>
>>>>>> >>>> >
>>>>>> >>>>
>>>>>> >>>
>>>>>> >>>
>>>>>> >>
>>>>>> >
>>>>>>
>>>>>
>>>>>
>>>>
>>>
>>
>
Reply | Threaded
Open this post in threaded view
|

Re: IGNITE-3196 - ready for review

Igor Sapego
Looks good to me.

Best Regards,
Igor

On Tue, Feb 7, 2017 at 2:55 PM, Vyacheslav Daradur <[hidden email]>
wrote:

> Ok, thanks for explanations.
>
> What about this task?
>
> 2017-02-07 13:57 GMT+03:00 Igor Sapego <[hidden email]>:
>
>> But that's Ok. Since we use int8_t for bytes in C++ as well I guess
>> your -0x80 may have more sense than 0x80.
>>
>> Best Regards,
>> Igor
>>
>> On Tue, Feb 7, 2017 at 1:54 PM, Igor Sapego <[hidden email]> wrote:
>>
>>> I was just curious.
>>>
>>> In C++ both constants 0x80 and -0x80 are of type 'int' and have the same
>>> lower byte, so they give the same result. Though first number is actually
>>> 0x00000080 when the second one is 0xFFFFFF80.
>>>
>>> So it's just made a minus sign look a little redundant and pointless to
>>> me
>>> in C++ code.
>>>
>>> Best Regards,
>>> Igor
>>>
>>> On Mon, Feb 6, 2017 at 10:15 PM, Vyacheslav Daradur <[hidden email]
>>> > wrote:
>>>
>>>> Byte.MIN_VALUE = -128 = -0x80
>>>> Byte.MAX_VALUE = 127 = 0x7F
>>>>
>>>> It is just more evident for me.
>>>>
>>>> Maybe, I just have the Java programming style.
>>>>
>>>> In Java:
>>>> byte a = 100 | -0x80;  // compiled
>>>> byte b = 100 | 0x80;  // doesn't compile, explicit type casting is
>>>> neccessary (byte)(100 | 0x80)
>>>> System.out.println(a | -0x80); // -28
>>>> System.out.println(a | 0x80); // 228 - cast to int
>>>>
>>>> Is it bad style?
>>>>
>>>> 2017-02-06 20:04 GMT+03:00 Igor Sapego <[hidden email]>:
>>>>
>>>>> Vyacheslav,
>>>>>
>>>>> Overall looks good. But why do you use -0x80 instead of 0x80?
>>>>>
>>>>> Best Regards,
>>>>> Igor
>>>>>
>>>>> On Mon, Feb 6, 2017 at 5:36 PM, Vyacheslav Daradur <
>>>>> [hidden email]> wrote:
>>>>>
>>>>>> Igor,
>>>>>>
>>>>>> I didn't change the CPP code before approval approach.
>>>>>> I shall write directly, sorry.
>>>>>>
>>>>>> But I made CPP changes already.
>>>>>>
>>>>>> > TestEscConvertFunctionFloat
>>>>>> > TestEscConvertFunctionDouble.
>>>>>> These tests were passed
>>>>>> <http://ci.ignite.apache.org/viewQueued.html?itemId=445824>
>>>>>>
>>>>>>
>>>>>>
>>>>>> 2017-02-06 13:20 GMT+03:00 Pavel Tupitsyn <[hidden email]>:
>>>>>>
>>>>>>> .NET changes look good to me.
>>>>>>>
>>>>>>> Pavel
>>>>>>>
>>>>>>> On Mon, Feb 6, 2017 at 1:10 PM, Igor Sapego <[hidden email]>
>>>>>>> wrote:
>>>>>>>
>>>>>>> > Vyacheslav, I can see two ODBC tests fail in C++ test suits that
>>>>>>> should
>>>>>>> > not:
>>>>>>> > - TestEscConvertFunctionFloat
>>>>>>> > <http://ci.ignite.apache.org/viewLog.html?buildId=444207&tab
>>>>>>> =buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformCppLi
>>>>>>> nux#testNameId-9178617718508801660>
>>>>>>> > - TestEscConvertFunctionDouble
>>>>>>> > <http://ci.ignite.apache.org/viewLog.html?buildId=444207&tab
>>>>>>> =buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformCppLi
>>>>>>> nux#testNameId5432107083822590090>
>>>>>>> > .
>>>>>>> >
>>>>>>> > I believe, this is because I can't see any changes in C++ Decimal
>>>>>>> > marshaling code.
>>>>>>> > Please, pay attention to file ignite\modules\platforms\cpp\
>>>>>>> > odbc\src\utility.cpp,
>>>>>>> > functions ReadDecimal and WriteDecimal.
>>>>>>> >
>>>>>>> > Best Regards,
>>>>>>> > Igor
>>>>>>> >
>>>>>>> > On Mon, Feb 6, 2017 at 11:21 AM, Vyacheslav Daradur <
>>>>>>> [hidden email]>
>>>>>>> > wrote:
>>>>>>> >
>>>>>>> >> Pavel, Igor
>>>>>>> >>
>>>>>>> >> Please, review it again.
>>>>>>> >>
>>>>>>> >> https://github.com/apache/ignite/pull/1473/files
>>>>>>> >>
>>>>>>> >> All tests
>>>>>>> >> <http://ci.ignite.apache.org/viewLog.html?buildId=444231&tab
>>>>>>> =buildResultsDiv&buildTypeId=IgniteTests_RunAll>
>>>>>>> >> .NET tests
>>>>>>> >> <http://ci.ignite.apache.org/viewLog.html?buildId=443439&tab
>>>>>>> =buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformNet>
>>>>>>> >>
>>>>>>> >> How about this solution?
>>>>>>> >>
>>>>>>> >> 2017-02-03 13:59 GMT+03:00 Vyacheslav Daradur <
>>>>>>> [hidden email]>:
>>>>>>> >>
>>>>>>> >>> 1. On my first question
>>>>>>> >>> I think up, if we serialize only positive numbers, we can write
>>>>>>> sign in
>>>>>>> >>> first byte, because it is positive always.
>>>>>>> >>> I will try to make this decision
>>>>>>> >>>
>>>>>>> >>> 2017-02-03 12:48 GMT+03:00 Pavel Tupitsyn <[hidden email]
>>>>>>> >:
>>>>>>> >>>
>>>>>>> >>>> Vyacheslav,
>>>>>>> >>>>
>>>>>>> >>>> I see the problem now. Yes, negative scale is not supported in
>>>>>>> .NET.
>>>>>>> >>>>
>>>>>>> >>>> I don't think we should do the multiplication. As you
>>>>>>> described, this
>>>>>>> >>>> will
>>>>>>> >>>> break equality on Java side. SQL queries might be broken, etc.
>>>>>>> >>>> I think we should throw an exception in .NET when encountering
>>>>>>> negative
>>>>>>> >>>> decimal scale.
>>>>>>> >>>>
>>>>>>> >>>> Vladimir O, any thoughts?
>>>>>>> >>>>
>>>>>>> >>>> Pavel
>>>>>>> >>>>
>>>>>>> >>>> On Fri, Feb 3, 2017 at 12:01 PM, Vyacheslav Daradur <
>>>>>>> >>>> [hidden email]>
>>>>>>> >>>> wrote:
>>>>>>> >>>>
>>>>>>> >>>> > Hello.
>>>>>>> >>>> >
>>>>>>> >>>> > I looked and understood the code of methods ReadDecimal and
>>>>>>> >>>> WriteDecimal
>>>>>>> >>>> > on .NET platform.
>>>>>>> >>>> >
>>>>>>> >>>> > 1. At the moment remaking of this methods for my
>>>>>>> Java-decimal-fix is
>>>>>>> >>>> very
>>>>>>> >>>> > difficult, it needs to write new methods for
>>>>>>> >>>> serialization/deserialization
>>>>>>> >>>> > of negative decimals.
>>>>>>> >>>> >
>>>>>>> >>>> > I can make it, but there is simpler decision: to add
>>>>>>> additional byte
>>>>>>> >>>> for
>>>>>>> >>>> > sign.
>>>>>>> >>>> >
>>>>>>> >>>> > I need advice: difficult solution (new methods .net) Versus :
>>>>>>> simple
>>>>>>> >>>> > solutions (additional byte)?
>>>>>>> >>>> >
>>>>>>> >>>> > *I don't know yet, what changes are necessary on С++ platform.
>>>>>>> >>>> >
>>>>>>> >>>> > 2. I see a problem with the negative scale on .NET platform.
>>>>>>> >>>> >
>>>>>>> >>>> > Now negative scale is forbidden.
>>>>>>> >>>> >
>>>>>>> >>>> > We can make:
>>>>>>> >>>> > if (scale < 0) return Decimal.Multiply(new decimal(lo, mid,
>>>>>>> hi, neg,
>>>>>>> >>>> 0),
>>>>>>> >>>> > new decimal(Math.Pow(10, -scale)));
>>>>>>> >>>> >
>>>>>>> >>>> > But there is the problem:
>>>>>>> >>>> > * 1 Serialize in Java; number=123456789, scale=-4
>>>>>>> >>>> > * 2 Deserialize in .NET; number=1234567890000, scale=0
>>>>>>> >>>> > * 3 Serialize in .NET; number=1234567890000, scale=0
>>>>>>> >>>> > * 4 Deserialize in Java; number=1234567890000, scale=0
>>>>>>> >>>> >
>>>>>>> >>>> > Logically: (1) 123456789 * 10^4 == (2)  1234567890000
>>>>>>> >>>> >
>>>>>>> >>>> > In Java (1) not equal (2), because scales are different.
>>>>>>> >>>> >
>>>>>>> >>>> > Any thougths?
>>>>>>> >>>> >
>>>>>>> >>>> > 2017-01-31 14:08 GMT+03:00 Pavel Tupitsyn <
>>>>>>> [hidden email]>:
>>>>>>> >>>> >
>>>>>>> >>>> >> Vyacheslav,
>>>>>>> >>>> >>
>>>>>>> >>>> >> I'm not sure I understand the code you attached.
>>>>>>> >>>> >>
>>>>>>> >>>> >> If you know how to fix the .NET part, can you just do it in
>>>>>>> your PR
>>>>>>> >>>> and
>>>>>>> >>>> >> run "Platform .NET" on TeamCity to verify?
>>>>>>> >>>> >> http://ci.ignite.apache.org/viewType.html?buildTypeId=Ignite
>>>>>>> >>>> >> Tests_IgnitePlatformNet
>>>>>>> >>>> >>
>>>>>>> >>>> >> Thanks,
>>>>>>> >>>> >>
>>>>>>> >>>> >> Pavel
>>>>>>> >>>> >>
>>>>>>> >>>> >> On Tue, Jan 31, 2017 at 1:35 PM, Vyacheslav Daradur <
>>>>>>> >>>> [hidden email]>
>>>>>>> >>>> >> wrote:
>>>>>>> >>>> >>
>>>>>>> >>>> >>> Pavel, I see that you are the main contributor of
>>>>>>> Ignite.NET.
>>>>>>> >>>> >>>
>>>>>>> >>>> >>> We should repair BinaryUtils#ReadDecimal and
>>>>>>> >>>> BinaryUtils#WriteDecimal.
>>>>>>> >>>> >>>
>>>>>>> >>>> >>> *ReadDecimal:
>>>>>>> >>>> >>> byte[] mag = ReadByteArray(stream); // including at least
>>>>>>> one sign
>>>>>>> >>>> bit,
>>>>>>> >>>> >>> which is (ceil((this.bitLength() + 1)/8))
>>>>>>> >>>> >>> bool neg = (mag[0] < 0);
>>>>>>> >>>> >>> if (scale < 0)
>>>>>>> >>>> >>> // TODO: a scale of -3 means the unscaled value is
>>>>>>> multiplied by
>>>>>>> >>>> 1000
>>>>>>> >>>> >>>
>>>>>>> >>>> >>> *WriteDecimal:
>>>>>>> >>>> >>> int sign = vals[3] < 0 ? -1 : 0;
>>>>>>> >>>> >>> stream.WriteInt(sign);
>>>>>>> >>>> >>>
>>>>>>> >>>> >>> Can you help with this task?
>>>>>>> >>>> >>>
>>>>>>> >>>> >>>
>>>>>>> >>>> >>> 2017-01-31 12:46 GMT+03:00 Igor Sapego <
>>>>>>> [hidden email]>:
>>>>>>> >>>> >>>
>>>>>>> >>>> >>>> Vyacheslav,
>>>>>>> >>>> >>>>
>>>>>>> >>>> >>>> I had a look at your PR and left some comments in Jira.
>>>>>>> >>>> >>>>
>>>>>>> >>>> >>>> Best Regards,
>>>>>>> >>>> >>>> Igor
>>>>>>> >>>> >>>>
>>>>>>> >>>> >>>> On Mon, Jan 30, 2017 at 12:52 PM, Vyacheslav Daradur <
>>>>>>> >>>> >>>> [hidden email]>
>>>>>>> >>>> >>>> wrote:
>>>>>>> >>>> >>>>
>>>>>>> >>>> >>>> > Hello. I fixed it. Please, review.
>>>>>>> >>>> >>>> >
>>>>>>> >>>> >>>> > https://issues.apache.org/jira/browse/IGNITE-3196 -
>>>>>>> Marshaling
>>>>>>> >>>> works
>>>>>>> >>>> >>>> wrong
>>>>>>> >>>> >>>> > for the BigDecimals that have negative scale
>>>>>>> >>>> >>>> >
>>>>>>> >>>> >>>>
>>>>>>> >>>> >>>
>>>>>>> >>>> >>>
>>>>>>> >>>> >>
>>>>>>> >>>> >
>>>>>>> >>>>
>>>>>>> >>>
>>>>>>> >>>
>>>>>>> >>
>>>>>>> >
>>>>>>>
>>>>>>
>>>>>>
>>>>>
>>>>
>>>
>>
>
Reply | Threaded
Open this post in threaded view
|

Re: IGNITE-3196 - ready for review

Pavel Tupitsyn
Ok, so we agree on .NET and C++ parts, only Java part is to be reviewed.

Denis, can you have a look?

Pavel

On Tue, Feb 7, 2017 at 3:23 PM, Igor Sapego <[hidden email]> wrote:

> Looks good to me.
>
> Best Regards,
> Igor
>
> On Tue, Feb 7, 2017 at 2:55 PM, Vyacheslav Daradur <[hidden email]>
> wrote:
>
> > Ok, thanks for explanations.
> >
> > What about this task?
> >
> > 2017-02-07 13:57 GMT+03:00 Igor Sapego <[hidden email]>:
> >
> >> But that's Ok. Since we use int8_t for bytes in C++ as well I guess
> >> your -0x80 may have more sense than 0x80.
> >>
> >> Best Regards,
> >> Igor
> >>
> >> On Tue, Feb 7, 2017 at 1:54 PM, Igor Sapego <[hidden email]>
> wrote:
> >>
> >>> I was just curious.
> >>>
> >>> In C++ both constants 0x80 and -0x80 are of type 'int' and have the
> same
> >>> lower byte, so they give the same result. Though first number is
> actually
> >>> 0x00000080 when the second one is 0xFFFFFF80.
> >>>
> >>> So it's just made a minus sign look a little redundant and pointless to
> >>> me
> >>> in C++ code.
> >>>
> >>> Best Regards,
> >>> Igor
> >>>
> >>> On Mon, Feb 6, 2017 at 10:15 PM, Vyacheslav Daradur <
> [hidden email]
> >>> > wrote:
> >>>
> >>>> Byte.MIN_VALUE = -128 = -0x80
> >>>> Byte.MAX_VALUE = 127 = 0x7F
> >>>>
> >>>> It is just more evident for me.
> >>>>
> >>>> Maybe, I just have the Java programming style.
> >>>>
> >>>> In Java:
> >>>> byte a = 100 | -0x80;  // compiled
> >>>> byte b = 100 | 0x80;  // doesn't compile, explicit type casting is
> >>>> neccessary (byte)(100 | 0x80)
> >>>> System.out.println(a | -0x80); // -28
> >>>> System.out.println(a | 0x80); // 228 - cast to int
> >>>>
> >>>> Is it bad style?
> >>>>
> >>>> 2017-02-06 20:04 GMT+03:00 Igor Sapego <[hidden email]>:
> >>>>
> >>>>> Vyacheslav,
> >>>>>
> >>>>> Overall looks good. But why do you use -0x80 instead of 0x80?
> >>>>>
> >>>>> Best Regards,
> >>>>> Igor
> >>>>>
> >>>>> On Mon, Feb 6, 2017 at 5:36 PM, Vyacheslav Daradur <
> >>>>> [hidden email]> wrote:
> >>>>>
> >>>>>> Igor,
> >>>>>>
> >>>>>> I didn't change the CPP code before approval approach.
> >>>>>> I shall write directly, sorry.
> >>>>>>
> >>>>>> But I made CPP changes already.
> >>>>>>
> >>>>>> > TestEscConvertFunctionFloat
> >>>>>> > TestEscConvertFunctionDouble.
> >>>>>> These tests were passed
> >>>>>> <http://ci.ignite.apache.org/viewQueued.html?itemId=445824>
> >>>>>>
> >>>>>>
> >>>>>>
> >>>>>> 2017-02-06 13:20 GMT+03:00 Pavel Tupitsyn <[hidden email]>:
> >>>>>>
> >>>>>>> .NET changes look good to me.
> >>>>>>>
> >>>>>>> Pavel
> >>>>>>>
> >>>>>>> On Mon, Feb 6, 2017 at 1:10 PM, Igor Sapego <[hidden email]>
> >>>>>>> wrote:
> >>>>>>>
> >>>>>>> > Vyacheslav, I can see two ODBC tests fail in C++ test suits that
> >>>>>>> should
> >>>>>>> > not:
> >>>>>>> > - TestEscConvertFunctionFloat
> >>>>>>> > <http://ci.ignite.apache.org/viewLog.html?buildId=444207&tab
> >>>>>>> =buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformCppLi
> >>>>>>> nux#testNameId-9178617718508801660>
> >>>>>>> > - TestEscConvertFunctionDouble
> >>>>>>> > <http://ci.ignite.apache.org/viewLog.html?buildId=444207&tab
> >>>>>>> =buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformCppLi
> >>>>>>> nux#testNameId5432107083822590090>
> >>>>>>> > .
> >>>>>>> >
> >>>>>>> > I believe, this is because I can't see any changes in C++ Decimal
> >>>>>>> > marshaling code.
> >>>>>>> > Please, pay attention to file ignite\modules\platforms\cpp\
> >>>>>>> > odbc\src\utility.cpp,
> >>>>>>> > functions ReadDecimal and WriteDecimal.
> >>>>>>> >
> >>>>>>> > Best Regards,
> >>>>>>> > Igor
> >>>>>>> >
> >>>>>>> > On Mon, Feb 6, 2017 at 11:21 AM, Vyacheslav Daradur <
> >>>>>>> [hidden email]>
> >>>>>>> > wrote:
> >>>>>>> >
> >>>>>>> >> Pavel, Igor
> >>>>>>> >>
> >>>>>>> >> Please, review it again.
> >>>>>>> >>
> >>>>>>> >> https://github.com/apache/ignite/pull/1473/files
> >>>>>>> >>
> >>>>>>> >> All tests
> >>>>>>> >> <http://ci.ignite.apache.org/viewLog.html?buildId=444231&tab
> >>>>>>> =buildResultsDiv&buildTypeId=IgniteTests_RunAll>
> >>>>>>> >> .NET tests
> >>>>>>> >> <http://ci.ignite.apache.org/viewLog.html?buildId=443439&tab
> >>>>>>> =buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformNet>
> >>>>>>> >>
> >>>>>>> >> How about this solution?
> >>>>>>> >>
> >>>>>>> >> 2017-02-03 13:59 GMT+03:00 Vyacheslav Daradur <
> >>>>>>> [hidden email]>:
> >>>>>>> >>
> >>>>>>> >>> 1. On my first question
> >>>>>>> >>> I think up, if we serialize only positive numbers, we can write
> >>>>>>> sign in
> >>>>>>> >>> first byte, because it is positive always.
> >>>>>>> >>> I will try to make this decision
> >>>>>>> >>>
> >>>>>>> >>> 2017-02-03 12:48 GMT+03:00 Pavel Tupitsyn <
> [hidden email]
> >>>>>>> >:
> >>>>>>> >>>
> >>>>>>> >>>> Vyacheslav,
> >>>>>>> >>>>
> >>>>>>> >>>> I see the problem now. Yes, negative scale is not supported in
> >>>>>>> .NET.
> >>>>>>> >>>>
> >>>>>>> >>>> I don't think we should do the multiplication. As you
> >>>>>>> described, this
> >>>>>>> >>>> will
> >>>>>>> >>>> break equality on Java side. SQL queries might be broken, etc.
> >>>>>>> >>>> I think we should throw an exception in .NET when encountering
> >>>>>>> negative
> >>>>>>> >>>> decimal scale.
> >>>>>>> >>>>
> >>>>>>> >>>> Vladimir O, any thoughts?
> >>>>>>> >>>>
> >>>>>>> >>>> Pavel
> >>>>>>> >>>>
> >>>>>>> >>>> On Fri, Feb 3, 2017 at 12:01 PM, Vyacheslav Daradur <
> >>>>>>> >>>> [hidden email]>
> >>>>>>> >>>> wrote:
> >>>>>>> >>>>
> >>>>>>> >>>> > Hello.
> >>>>>>> >>>> >
> >>>>>>> >>>> > I looked and understood the code of methods ReadDecimal and
> >>>>>>> >>>> WriteDecimal
> >>>>>>> >>>> > on .NET platform.
> >>>>>>> >>>> >
> >>>>>>> >>>> > 1. At the moment remaking of this methods for my
> >>>>>>> Java-decimal-fix is
> >>>>>>> >>>> very
> >>>>>>> >>>> > difficult, it needs to write new methods for
> >>>>>>> >>>> serialization/deserialization
> >>>>>>> >>>> > of negative decimals.
> >>>>>>> >>>> >
> >>>>>>> >>>> > I can make it, but there is simpler decision: to add
> >>>>>>> additional byte
> >>>>>>> >>>> for
> >>>>>>> >>>> > sign.
> >>>>>>> >>>> >
> >>>>>>> >>>> > I need advice: difficult solution (new methods .net) Versus
> :
> >>>>>>> simple
> >>>>>>> >>>> > solutions (additional byte)?
> >>>>>>> >>>> >
> >>>>>>> >>>> > *I don't know yet, what changes are necessary on С++
> platform.
> >>>>>>> >>>> >
> >>>>>>> >>>> > 2. I see a problem with the negative scale on .NET platform.
> >>>>>>> >>>> >
> >>>>>>> >>>> > Now negative scale is forbidden.
> >>>>>>> >>>> >
> >>>>>>> >>>> > We can make:
> >>>>>>> >>>> > if (scale < 0) return Decimal.Multiply(new decimal(lo, mid,
> >>>>>>> hi, neg,
> >>>>>>> >>>> 0),
> >>>>>>> >>>> > new decimal(Math.Pow(10, -scale)));
> >>>>>>> >>>> >
> >>>>>>> >>>> > But there is the problem:
> >>>>>>> >>>> > * 1 Serialize in Java; number=123456789, scale=-4
> >>>>>>> >>>> > * 2 Deserialize in .NET; number=1234567890000, scale=0
> >>>>>>> >>>> > * 3 Serialize in .NET; number=1234567890000, scale=0
> >>>>>>> >>>> > * 4 Deserialize in Java; number=1234567890000, scale=0
> >>>>>>> >>>> >
> >>>>>>> >>>> > Logically: (1) 123456789 * 10^4 == (2)  1234567890000
> >>>>>>> >>>> >
> >>>>>>> >>>> > In Java (1) not equal (2), because scales are different.
> >>>>>>> >>>> >
> >>>>>>> >>>> > Any thougths?
> >>>>>>> >>>> >
> >>>>>>> >>>> > 2017-01-31 14:08 GMT+03:00 Pavel Tupitsyn <
> >>>>>>> [hidden email]>:
> >>>>>>> >>>> >
> >>>>>>> >>>> >> Vyacheslav,
> >>>>>>> >>>> >>
> >>>>>>> >>>> >> I'm not sure I understand the code you attached.
> >>>>>>> >>>> >>
> >>>>>>> >>>> >> If you know how to fix the .NET part, can you just do it in
> >>>>>>> your PR
> >>>>>>> >>>> and
> >>>>>>> >>>> >> run "Platform .NET" on TeamCity to verify?
> >>>>>>> >>>> >> http://ci.ignite.apache.org/viewType.html?buildTypeId=
> Ignite
> >>>>>>> >>>> >> Tests_IgnitePlatformNet
> >>>>>>> >>>> >>
> >>>>>>> >>>> >> Thanks,
> >>>>>>> >>>> >>
> >>>>>>> >>>> >> Pavel
> >>>>>>> >>>> >>
> >>>>>>> >>>> >> On Tue, Jan 31, 2017 at 1:35 PM, Vyacheslav Daradur <
> >>>>>>> >>>> [hidden email]>
> >>>>>>> >>>> >> wrote:
> >>>>>>> >>>> >>
> >>>>>>> >>>> >>> Pavel, I see that you are the main contributor of
> >>>>>>> Ignite.NET.
> >>>>>>> >>>> >>>
> >>>>>>> >>>> >>> We should repair BinaryUtils#ReadDecimal and
> >>>>>>> >>>> BinaryUtils#WriteDecimal.
> >>>>>>> >>>> >>>
> >>>>>>> >>>> >>> *ReadDecimal:
> >>>>>>> >>>> >>> byte[] mag = ReadByteArray(stream); // including at least
> >>>>>>> one sign
> >>>>>>> >>>> bit,
> >>>>>>> >>>> >>> which is (ceil((this.bitLength() + 1)/8))
> >>>>>>> >>>> >>> bool neg = (mag[0] < 0);
> >>>>>>> >>>> >>> if (scale < 0)
> >>>>>>> >>>> >>> // TODO: a scale of -3 means the unscaled value is
> >>>>>>> multiplied by
> >>>>>>> >>>> 1000
> >>>>>>> >>>> >>>
> >>>>>>> >>>> >>> *WriteDecimal:
> >>>>>>> >>>> >>> int sign = vals[3] < 0 ? -1 : 0;
> >>>>>>> >>>> >>> stream.WriteInt(sign);
> >>>>>>> >>>> >>>
> >>>>>>> >>>> >>> Can you help with this task?
> >>>>>>> >>>> >>>
> >>>>>>> >>>> >>>
> >>>>>>> >>>> >>> 2017-01-31 12:46 GMT+03:00 Igor Sapego <
> >>>>>>> [hidden email]>:
> >>>>>>> >>>> >>>
> >>>>>>> >>>> >>>> Vyacheslav,
> >>>>>>> >>>> >>>>
> >>>>>>> >>>> >>>> I had a look at your PR and left some comments in Jira.
> >>>>>>> >>>> >>>>
> >>>>>>> >>>> >>>> Best Regards,
> >>>>>>> >>>> >>>> Igor
> >>>>>>> >>>> >>>>
> >>>>>>> >>>> >>>> On Mon, Jan 30, 2017 at 12:52 PM, Vyacheslav Daradur <
> >>>>>>> >>>> >>>> [hidden email]>
> >>>>>>> >>>> >>>> wrote:
> >>>>>>> >>>> >>>>
> >>>>>>> >>>> >>>> > Hello. I fixed it. Please, review.
> >>>>>>> >>>> >>>> >
> >>>>>>> >>>> >>>> > https://issues.apache.org/jira/browse/IGNITE-3196 -
> >>>>>>> Marshaling
> >>>>>>> >>>> works
> >>>>>>> >>>> >>>> wrong
> >>>>>>> >>>> >>>> > for the BigDecimals that have negative scale
> >>>>>>> >>>> >>>> >
> >>>>>>> >>>> >>>>
> >>>>>>> >>>> >>>
> >>>>>>> >>>> >>>
> >>>>>>> >>>> >>
> >>>>>>> >>>> >
> >>>>>>> >>>>
> >>>>>>> >>>
> >>>>>>> >>>
> >>>>>>> >>
> >>>>>>> >
> >>>>>>>
> >>>>>>
> >>>>>>
> >>>>>
> >>>>
> >>>
> >>
> >
>
Reply | Threaded
Open this post in threaded view
|

Re: IGNITE-3196 - ready for review

daradurvs
+Denis

>>Ok, so we agree on .NET and C++ parts, only Java part is to be reviewed.
>>Denis, can you have a look?

2017-02-07 15:27 GMT+03:00 Pavel Tupitsyn <[hidden email]>:

> Ok, so we agree on .NET and C++ parts, only Java part is to be reviewed.
>
> Denis, can you have a look?
>
> Pavel
>
> On Tue, Feb 7, 2017 at 3:23 PM, Igor Sapego <[hidden email]> wrote:
>
> > Looks good to me.
> >
> > Best Regards,
> > Igor
> >
> > On Tue, Feb 7, 2017 at 2:55 PM, Vyacheslav Daradur <[hidden email]>
> > wrote:
> >
> > > Ok, thanks for explanations.
> > >
> > > What about this task?
> > >
> > > 2017-02-07 13:57 GMT+03:00 Igor Sapego <[hidden email]>:
> > >
> > >> But that's Ok. Since we use int8_t for bytes in C++ as well I guess
> > >> your -0x80 may have more sense than 0x80.
> > >>
> > >> Best Regards,
> > >> Igor
> > >>
> > >> On Tue, Feb 7, 2017 at 1:54 PM, Igor Sapego <[hidden email]>
> > wrote:
> > >>
> > >>> I was just curious.
> > >>>
> > >>> In C++ both constants 0x80 and -0x80 are of type 'int' and have the
> > same
> > >>> lower byte, so they give the same result. Though first number is
> > actually
> > >>> 0x00000080 when the second one is 0xFFFFFF80.
> > >>>
> > >>> So it's just made a minus sign look a little redundant and pointless
> to
> > >>> me
> > >>> in C++ code.
> > >>>
> > >>> Best Regards,
> > >>> Igor
> > >>>
> > >>> On Mon, Feb 6, 2017 at 10:15 PM, Vyacheslav Daradur <
> > [hidden email]
> > >>> > wrote:
> > >>>
> > >>>> Byte.MIN_VALUE = -128 = -0x80
> > >>>> Byte.MAX_VALUE = 127 = 0x7F
> > >>>>
> > >>>> It is just more evident for me.
> > >>>>
> > >>>> Maybe, I just have the Java programming style.
> > >>>>
> > >>>> In Java:
> > >>>> byte a = 100 | -0x80;  // compiled
> > >>>> byte b = 100 | 0x80;  // doesn't compile, explicit type casting is
> > >>>> neccessary (byte)(100 | 0x80)
> > >>>> System.out.println(a | -0x80); // -28
> > >>>> System.out.println(a | 0x80); // 228 - cast to int
> > >>>>
> > >>>> Is it bad style?
> > >>>>
> > >>>> 2017-02-06 20:04 GMT+03:00 Igor Sapego <[hidden email]>:
> > >>>>
> > >>>>> Vyacheslav,
> > >>>>>
> > >>>>> Overall looks good. But why do you use -0x80 instead of 0x80?
> > >>>>>
> > >>>>> Best Regards,
> > >>>>> Igor
> > >>>>>
> > >>>>> On Mon, Feb 6, 2017 at 5:36 PM, Vyacheslav Daradur <
> > >>>>> [hidden email]> wrote:
> > >>>>>
> > >>>>>> Igor,
> > >>>>>>
> > >>>>>> I didn't change the CPP code before approval approach.
> > >>>>>> I shall write directly, sorry.
> > >>>>>>
> > >>>>>> But I made CPP changes already.
> > >>>>>>
> > >>>>>> > TestEscConvertFunctionFloat
> > >>>>>> > TestEscConvertFunctionDouble.
> > >>>>>> These tests were passed
> > >>>>>> <http://ci.ignite.apache.org/viewQueued.html?itemId=445824>
> > >>>>>>
> > >>>>>>
> > >>>>>>
> > >>>>>> 2017-02-06 13:20 GMT+03:00 Pavel Tupitsyn <[hidden email]>:
> > >>>>>>
> > >>>>>>> .NET changes look good to me.
> > >>>>>>>
> > >>>>>>> Pavel
> > >>>>>>>
> > >>>>>>> On Mon, Feb 6, 2017 at 1:10 PM, Igor Sapego <
> [hidden email]>
> > >>>>>>> wrote:
> > >>>>>>>
> > >>>>>>> > Vyacheslav, I can see two ODBC tests fail in C++ test suits
> that
> > >>>>>>> should
> > >>>>>>> > not:
> > >>>>>>> > - TestEscConvertFunctionFloat
> > >>>>>>> > <http://ci.ignite.apache.org/viewLog.html?buildId=444207&tab
> > >>>>>>> =buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformCppLi
> > >>>>>>> nux#testNameId-9178617718508801660>
> > >>>>>>> > - TestEscConvertFunctionDouble
> > >>>>>>> > <http://ci.ignite.apache.org/viewLog.html?buildId=444207&tab
> > >>>>>>> =buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformCppLi
> > >>>>>>> nux#testNameId5432107083822590090>
> > >>>>>>> > .
> > >>>>>>> >
> > >>>>>>> > I believe, this is because I can't see any changes in C++
> Decimal
> > >>>>>>> > marshaling code.
> > >>>>>>> > Please, pay attention to file ignite\modules\platforms\cpp\
> > >>>>>>> > odbc\src\utility.cpp,
> > >>>>>>> > functions ReadDecimal and WriteDecimal.
> > >>>>>>> >
> > >>>>>>> > Best Regards,
> > >>>>>>> > Igor
> > >>>>>>> >
> > >>>>>>> > On Mon, Feb 6, 2017 at 11:21 AM, Vyacheslav Daradur <
> > >>>>>>> [hidden email]>
> > >>>>>>> > wrote:
> > >>>>>>> >
> > >>>>>>> >> Pavel, Igor
> > >>>>>>> >>
> > >>>>>>> >> Please, review it again.
> > >>>>>>> >>
> > >>>>>>> >> https://github.com/apache/ignite/pull/1473/files
> > >>>>>>> >>
> > >>>>>>> >> All tests
> > >>>>>>> >> <http://ci.ignite.apache.org/viewLog.html?buildId=444231&tab
> > >>>>>>> =buildResultsDiv&buildTypeId=IgniteTests_RunAll>
> > >>>>>>> >> .NET tests
> > >>>>>>> >> <http://ci.ignite.apache.org/viewLog.html?buildId=443439&tab
> > >>>>>>> =buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformNet>
> > >>>>>>> >>
> > >>>>>>> >> How about this solution?
> > >>>>>>> >>
> > >>>>>>> >> 2017-02-03 13:59 GMT+03:00 Vyacheslav Daradur <
> > >>>>>>> [hidden email]>:
> > >>>>>>> >>
> > >>>>>>> >>> 1. On my first question
> > >>>>>>> >>> I think up, if we serialize only positive numbers, we can
> write
> > >>>>>>> sign in
> > >>>>>>> >>> first byte, because it is positive always.
> > >>>>>>> >>> I will try to make this decision
> > >>>>>>> >>>
> > >>>>>>> >>> 2017-02-03 12:48 GMT+03:00 Pavel Tupitsyn <
> > [hidden email]
> > >>>>>>> >:
> > >>>>>>> >>>
> > >>>>>>> >>>> Vyacheslav,
> > >>>>>>> >>>>
> > >>>>>>> >>>> I see the problem now. Yes, negative scale is not supported
> in
> > >>>>>>> .NET.
> > >>>>>>> >>>>
> > >>>>>>> >>>> I don't think we should do the multiplication. As you
> > >>>>>>> described, this
> > >>>>>>> >>>> will
> > >>>>>>> >>>> break equality on Java side. SQL queries might be broken,
> etc.
> > >>>>>>> >>>> I think we should throw an exception in .NET when
> encountering
> > >>>>>>> negative
> > >>>>>>> >>>> decimal scale.
> > >>>>>>> >>>>
> > >>>>>>> >>>> Vladimir O, any thoughts?
> > >>>>>>> >>>>
> > >>>>>>> >>>> Pavel
> > >>>>>>> >>>>
> > >>>>>>> >>>> On Fri, Feb 3, 2017 at 12:01 PM, Vyacheslav Daradur <
> > >>>>>>> >>>> [hidden email]>
> > >>>>>>> >>>> wrote:
> > >>>>>>> >>>>
> > >>>>>>> >>>> > Hello.
> > >>>>>>> >>>> >
> > >>>>>>> >>>> > I looked and understood the code of methods ReadDecimal
> and
> > >>>>>>> >>>> WriteDecimal
> > >>>>>>> >>>> > on .NET platform.
> > >>>>>>> >>>> >
> > >>>>>>> >>>> > 1. At the moment remaking of this methods for my
> > >>>>>>> Java-decimal-fix is
> > >>>>>>> >>>> very
> > >>>>>>> >>>> > difficult, it needs to write new methods for
> > >>>>>>> >>>> serialization/deserialization
> > >>>>>>> >>>> > of negative decimals.
> > >>>>>>> >>>> >
> > >>>>>>> >>>> > I can make it, but there is simpler decision: to add
> > >>>>>>> additional byte
> > >>>>>>> >>>> for
> > >>>>>>> >>>> > sign.
> > >>>>>>> >>>> >
> > >>>>>>> >>>> > I need advice: difficult solution (new methods .net)
> Versus
> > :
> > >>>>>>> simple
> > >>>>>>> >>>> > solutions (additional byte)?
> > >>>>>>> >>>> >
> > >>>>>>> >>>> > *I don't know yet, what changes are necessary on С++
> > platform.
> > >>>>>>> >>>> >
> > >>>>>>> >>>> > 2. I see a problem with the negative scale on .NET
> platform.
> > >>>>>>> >>>> >
> > >>>>>>> >>>> > Now negative scale is forbidden.
> > >>>>>>> >>>> >
> > >>>>>>> >>>> > We can make:
> > >>>>>>> >>>> > if (scale < 0) return Decimal.Multiply(new decimal(lo,
> mid,
> > >>>>>>> hi, neg,
> > >>>>>>> >>>> 0),
> > >>>>>>> >>>> > new decimal(Math.Pow(10, -scale)));
> > >>>>>>> >>>> >
> > >>>>>>> >>>> > But there is the problem:
> > >>>>>>> >>>> > * 1 Serialize in Java; number=123456789, scale=-4
> > >>>>>>> >>>> > * 2 Deserialize in .NET; number=1234567890000, scale=0
> > >>>>>>> >>>> > * 3 Serialize in .NET; number=1234567890000, scale=0
> > >>>>>>> >>>> > * 4 Deserialize in Java; number=1234567890000, scale=0
> > >>>>>>> >>>> >
> > >>>>>>> >>>> > Logically: (1) 123456789 * 10^4 == (2)  1234567890000
> > >>>>>>> >>>> >
> > >>>>>>> >>>> > In Java (1) not equal (2), because scales are different.
> > >>>>>>> >>>> >
> > >>>>>>> >>>> > Any thougths?
> > >>>>>>> >>>> >
> > >>>>>>> >>>> > 2017-01-31 14:08 GMT+03:00 Pavel Tupitsyn <
> > >>>>>>> [hidden email]>:
> > >>>>>>> >>>> >
> > >>>>>>> >>>> >> Vyacheslav,
> > >>>>>>> >>>> >>
> > >>>>>>> >>>> >> I'm not sure I understand the code you attached.
> > >>>>>>> >>>> >>
> > >>>>>>> >>>> >> If you know how to fix the .NET part, can you just do it
> in
> > >>>>>>> your PR
> > >>>>>>> >>>> and
> > >>>>>>> >>>> >> run "Platform .NET" on TeamCity to verify?
> > >>>>>>> >>>> >> http://ci.ignite.apache.org/viewType.html?buildTypeId=
> > Ignite
> > >>>>>>> >>>> >> Tests_IgnitePlatformNet
> > >>>>>>> >>>> >>
> > >>>>>>> >>>> >> Thanks,
> > >>>>>>> >>>> >>
> > >>>>>>> >>>> >> Pavel
> > >>>>>>> >>>> >>
> > >>>>>>> >>>> >> On Tue, Jan 31, 2017 at 1:35 PM, Vyacheslav Daradur <
> > >>>>>>> >>>> [hidden email]>
> > >>>>>>> >>>> >> wrote:
> > >>>>>>> >>>> >>
> > >>>>>>> >>>> >>> Pavel, I see that you are the main contributor of
> > >>>>>>> Ignite.NET.
> > >>>>>>> >>>> >>>
> > >>>>>>> >>>> >>> We should repair BinaryUtils#ReadDecimal and
> > >>>>>>> >>>> BinaryUtils#WriteDecimal.
> > >>>>>>> >>>> >>>
> > >>>>>>> >>>> >>> *ReadDecimal:
> > >>>>>>> >>>> >>> byte[] mag = ReadByteArray(stream); // including at
> least
> > >>>>>>> one sign
> > >>>>>>> >>>> bit,
> > >>>>>>> >>>> >>> which is (ceil((this.bitLength() + 1)/8))
> > >>>>>>> >>>> >>> bool neg = (mag[0] < 0);
> > >>>>>>> >>>> >>> if (scale < 0)
> > >>>>>>> >>>> >>> // TODO: a scale of -3 means the unscaled value is
> > >>>>>>> multiplied by
> > >>>>>>> >>>> 1000
> > >>>>>>> >>>> >>>
> > >>>>>>> >>>> >>> *WriteDecimal:
> > >>>>>>> >>>> >>> int sign = vals[3] < 0 ? -1 : 0;
> > >>>>>>> >>>> >>> stream.WriteInt(sign);
> > >>>>>>> >>>> >>>
> > >>>>>>> >>>> >>> Can you help with this task?
> > >>>>>>> >>>> >>>
> > >>>>>>> >>>> >>>
> > >>>>>>> >>>> >>> 2017-01-31 12:46 GMT+03:00 Igor Sapego <
> > >>>>>>> [hidden email]>:
> > >>>>>>> >>>> >>>
> > >>>>>>> >>>> >>>> Vyacheslav,
> > >>>>>>> >>>> >>>>
> > >>>>>>> >>>> >>>> I had a look at your PR and left some comments in Jira.
> > >>>>>>> >>>> >>>>
> > >>>>>>> >>>> >>>> Best Regards,
> > >>>>>>> >>>> >>>> Igor
> > >>>>>>> >>>> >>>>
> > >>>>>>> >>>> >>>> On Mon, Jan 30, 2017 at 12:52 PM, Vyacheslav Daradur <
> > >>>>>>> >>>> >>>> [hidden email]>
> > >>>>>>> >>>> >>>> wrote:
> > >>>>>>> >>>> >>>>
> > >>>>>>> >>>> >>>> > Hello. I fixed it. Please, review.
> > >>>>>>> >>>> >>>> >
> > >>>>>>> >>>> >>>> > https://issues.apache.org/jira/browse/IGNITE-3196 -
> > >>>>>>> Marshaling
> > >>>>>>> >>>> works
> > >>>>>>> >>>> >>>> wrong
> > >>>>>>> >>>> >>>> > for the BigDecimals that have negative scale
> > >>>>>>> >>>> >>>> >
> > >>>>>>> >>>> >>>>
> > >>>>>>> >>>> >>>
> > >>>>>>> >>>> >>>
> > >>>>>>> >>>> >>
> > >>>>>>> >>>> >
> > >>>>>>> >>>>
> > >>>>>>> >>>
> > >>>>>>> >>>
> > >>>>>>> >>
> > >>>>>>> >
> > >>>>>>>
> > >>>>>>
> > >>>>>>
> > >>>>>
> > >>>>
> > >>>
> > >>
> > >
> >
>
Reply | Threaded
Open this post in threaded view
|

Re: IGNITE-3196 - ready for review

dmagda
+ Vladimir Ozerov

It would be better if Vladimir Ozerov does the final review considering all the changes in .NET, C++ and Java.

Vladimir, could you do that?


Denis

> On Feb 7, 2017, at 5:04 AM, Vyacheslav Daradur <[hidden email]> wrote:
>
> +Denis
>
> >>Ok, so we agree on .NET and C++ parts, only Java part is to be reviewed.
> >>Denis, can you have a look?
>
> 2017-02-07 15:27 GMT+03:00 Pavel Tupitsyn <[hidden email] <mailto:[hidden email]>>:
> Ok, so we agree on .NET and C++ parts, only Java part is to be reviewed.
>
> Denis, can you have a look?
>
> Pavel
>
> On Tue, Feb 7, 2017 at 3:23 PM, Igor Sapego <[hidden email] <mailto:[hidden email]>> wrote:
>
> > Looks good to me.
> >
> > Best Regards,
> > Igor
> >
> > On Tue, Feb 7, 2017 at 2:55 PM, Vyacheslav Daradur <[hidden email] <mailto:[hidden email]>>
> > wrote:
> >
> > > Ok, thanks for explanations.
> > >
> > > What about this task?
> > >
> > > 2017-02-07 13:57 GMT+03:00 Igor Sapego <[hidden email] <mailto:[hidden email]>>:
> > >
> > >> But that's Ok. Since we use int8_t for bytes in C++ as well I guess
> > >> your -0x80 may have more sense than 0x80.
> > >>
> > >> Best Regards,
> > >> Igor
> > >>
> > >> On Tue, Feb 7, 2017 at 1:54 PM, Igor Sapego <[hidden email] <mailto:[hidden email]>>
> > wrote:
> > >>
> > >>> I was just curious.
> > >>>
> > >>> In C++ both constants 0x80 and -0x80 are of type 'int' and have the
> > same
> > >>> lower byte, so they give the same result. Though first number is
> > actually
> > >>> 0x00000080 when the second one is 0xFFFFFF80.
> > >>>
> > >>> So it's just made a minus sign look a little redundant and pointless to
> > >>> me
> > >>> in C++ code.
> > >>>
> > >>> Best Regards,
> > >>> Igor
> > >>>
> > >>> On Mon, Feb 6, 2017 at 10:15 PM, Vyacheslav Daradur <
> > [hidden email] <mailto:[hidden email]>
> > >>> > wrote:
> > >>>
> > >>>> Byte.MIN_VALUE = -128 = -0x80
> > >>>> Byte.MAX_VALUE = 127 = 0x7F
> > >>>>
> > >>>> It is just more evident for me.
> > >>>>
> > >>>> Maybe, I just have the Java programming style.
> > >>>>
> > >>>> In Java:
> > >>>> byte a = 100 | -0x80;  // compiled
> > >>>> byte b = 100 | 0x80;  // doesn't compile, explicit type casting is
> > >>>> neccessary (byte)(100 | 0x80)
> > >>>> System.out.println(a | -0x80); // -28
> > >>>> System.out.println(a | 0x80); // 228 - cast to int
> > >>>>
> > >>>> Is it bad style?
> > >>>>
> > >>>> 2017-02-06 20:04 GMT+03:00 Igor Sapego <[hidden email] <mailto:[hidden email]>>:
> > >>>>
> > >>>>> Vyacheslav,
> > >>>>>
> > >>>>> Overall looks good. But why do you use -0x80 instead of 0x80?
> > >>>>>
> > >>>>> Best Regards,
> > >>>>> Igor
> > >>>>>
> > >>>>> On Mon, Feb 6, 2017 at 5:36 PM, Vyacheslav Daradur <
> > >>>>> [hidden email] <mailto:[hidden email]>> wrote:
> > >>>>>
> > >>>>>> Igor,
> > >>>>>>
> > >>>>>> I didn't change the CPP code before approval approach.
> > >>>>>> I shall write directly, sorry.
> > >>>>>>
> > >>>>>> But I made CPP changes already.
> > >>>>>>
> > >>>>>> > TestEscConvertFunctionFloat
> > >>>>>> > TestEscConvertFunctionDouble.
> > >>>>>> These tests were passed
> > >>>>>> <http://ci.ignite.apache.org/viewQueued.html?itemId=445824 <http://ci.ignite.apache.org/viewQueued.html?itemId=445824>>
> > >>>>>>
> > >>>>>>
> > >>>>>>
> > >>>>>> 2017-02-06 13:20 GMT+03:00 Pavel Tupitsyn <[hidden email] <mailto:[hidden email]>>:
> > >>>>>>
> > >>>>>>> .NET changes look good to me.
> > >>>>>>>
> > >>>>>>> Pavel
> > >>>>>>>
> > >>>>>>> On Mon, Feb 6, 2017 at 1:10 PM, Igor Sapego <[hidden email] <mailto:[hidden email]>>
> > >>>>>>> wrote:
> > >>>>>>>
> > >>>>>>> > Vyacheslav, I can see two ODBC tests fail in C++ test suits that
> > >>>>>>> should
> > >>>>>>> > not:
> > >>>>>>> > - TestEscConvertFunctionFloat
> > >>>>>>> > <http://ci.ignite.apache.org/viewLog.html?buildId=444207&tab <http://ci.ignite.apache.org/viewLog.html?buildId=444207&tab>
> > >>>>>>> =buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformCppLi
> > >>>>>>> nux#testNameId-9178617718508801660>
> > >>>>>>> > - TestEscConvertFunctionDouble
> > >>>>>>> > <http://ci.ignite.apache.org/viewLog.html?buildId=444207&tab <http://ci.ignite.apache.org/viewLog.html?buildId=444207&tab>
> > >>>>>>> =buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformCppLi
> > >>>>>>> nux#testNameId5432107083822590090>
> > >>>>>>> > .
> > >>>>>>> >
> > >>>>>>> > I believe, this is because I can't see any changes in C++ Decimal
> > >>>>>>> > marshaling code.
> > >>>>>>> > Please, pay attention to file ignite\modules\platforms\cpp\
> > >>>>>>> > odbc\src\utility.cpp,
> > >>>>>>> > functions ReadDecimal and WriteDecimal.
> > >>>>>>> >
> > >>>>>>> > Best Regards,
> > >>>>>>> > Igor
> > >>>>>>> >
> > >>>>>>> > On Mon, Feb 6, 2017 at 11:21 AM, Vyacheslav Daradur <
> > >>>>>>> [hidden email] <mailto:[hidden email]>>
> > >>>>>>> > wrote:
> > >>>>>>> >
> > >>>>>>> >> Pavel, Igor
> > >>>>>>> >>
> > >>>>>>> >> Please, review it again.
> > >>>>>>> >>
> > >>>>>>> >> https://github.com/apache/ignite/pull/1473/files <https://github.com/apache/ignite/pull/1473/files>
> > >>>>>>> >>
> > >>>>>>> >> All tests
> > >>>>>>> >> <http://ci.ignite.apache.org/viewLog.html?buildId=444231&tab <http://ci.ignite.apache.org/viewLog.html?buildId=444231&tab>
> > >>>>>>> =buildResultsDiv&buildTypeId=IgniteTests_RunAll>
> > >>>>>>> >> .NET tests
> > >>>>>>> >> <http://ci.ignite.apache.org/viewLog.html?buildId=443439&tab <http://ci.ignite.apache.org/viewLog.html?buildId=443439&tab>
> > >>>>>>> =buildResultsDiv&buildTypeId=IgniteTests_IgnitePlatformNet>
> > >>>>>>> >>
> > >>>>>>> >> How about this solution?
> > >>>>>>> >>
> > >>>>>>> >> 2017-02-03 13:59 GMT+03:00 Vyacheslav Daradur <
> > >>>>>>> [hidden email] <mailto:[hidden email]>>:
> > >>>>>>> >>
> > >>>>>>> >>> 1. On my first question
> > >>>>>>> >>> I think up, if we serialize only positive numbers, we can write
> > >>>>>>> sign in
> > >>>>>>> >>> first byte, because it is positive always.
> > >>>>>>> >>> I will try to make this decision
> > >>>>>>> >>>
> > >>>>>>> >>> 2017-02-03 12:48 GMT+03:00 Pavel Tupitsyn <
> > [hidden email] <mailto:[hidden email]>
> > >>>>>>> >:
> > >>>>>>> >>>
> > >>>>>>> >>>> Vyacheslav,
> > >>>>>>> >>>>
> > >>>>>>> >>>> I see the problem now. Yes, negative scale is not supported in
> > >>>>>>> .NET.
> > >>>>>>> >>>>
> > >>>>>>> >>>> I don't think we should do the multiplication. As you
> > >>>>>>> described, this
> > >>>>>>> >>>> will
> > >>>>>>> >>>> break equality on Java side. SQL queries might be broken, etc.
> > >>>>>>> >>>> I think we should throw an exception in .NET when encountering
> > >>>>>>> negative
> > >>>>>>> >>>> decimal scale.
> > >>>>>>> >>>>
> > >>>>>>> >>>> Vladimir O, any thoughts?
> > >>>>>>> >>>>
> > >>>>>>> >>>> Pavel
> > >>>>>>> >>>>
> > >>>>>>> >>>> On Fri, Feb 3, 2017 at 12:01 PM, Vyacheslav Daradur <
> > >>>>>>> >>>> [hidden email] <mailto:[hidden email]>>
> > >>>>>>> >>>> wrote:
> > >>>>>>> >>>>
> > >>>>>>> >>>> > Hello.
> > >>>>>>> >>>> >
> > >>>>>>> >>>> > I looked and understood the code of methods ReadDecimal and
> > >>>>>>> >>>> WriteDecimal
> > >>>>>>> >>>> > on .NET platform.
> > >>>>>>> >>>> >
> > >>>>>>> >>>> > 1. At the moment remaking of this methods for my
> > >>>>>>> Java-decimal-fix is
> > >>>>>>> >>>> very
> > >>>>>>> >>>> > difficult, it needs to write new methods for
> > >>>>>>> >>>> serialization/deserialization
> > >>>>>>> >>>> > of negative decimals.
> > >>>>>>> >>>> >
> > >>>>>>> >>>> > I can make it, but there is simpler decision: to add
> > >>>>>>> additional byte
> > >>>>>>> >>>> for
> > >>>>>>> >>>> > sign.
> > >>>>>>> >>>> >
> > >>>>>>> >>>> > I need advice: difficult solution (new methods .net) Versus
> > :
> > >>>>>>> simple
> > >>>>>>> >>>> > solutions (additional byte)?
> > >>>>>>> >>>> >
> > >>>>>>> >>>> > *I don't know yet, what changes are necessary on С++
> > platform.
> > >>>>>>> >>>> >
> > >>>>>>> >>>> > 2. I see a problem with the negative scale on .NET platform.
> > >>>>>>> >>>> >
> > >>>>>>> >>>> > Now negative scale is forbidden.
> > >>>>>>> >>>> >
> > >>>>>>> >>>> > We can make:
> > >>>>>>> >>>> > if (scale < 0) return Decimal.Multiply(new decimal(lo, mid,
> > >>>>>>> hi, neg,
> > >>>>>>> >>>> 0),
> > >>>>>>> >>>> > new decimal(Math.Pow(10, -scale)));
> > >>>>>>> >>>> >
> > >>>>>>> >>>> > But there is the problem:
> > >>>>>>> >>>> > * 1 Serialize in Java; number=123456789, scale=-4
> > >>>>>>> >>>> > * 2 Deserialize in .NET; number=1234567890000, scale=0
> > >>>>>>> >>>> > * 3 Serialize in .NET; number=1234567890000, scale=0
> > >>>>>>> >>>> > * 4 Deserialize in Java; number=1234567890000, scale=0
> > >>>>>>> >>>> >
> > >>>>>>> >>>> > Logically: (1) 123456789 * 10^4 == (2)  1234567890000
> > >>>>>>> >>>> >
> > >>>>>>> >>>> > In Java (1) not equal (2), because scales are different.
> > >>>>>>> >>>> >
> > >>>>>>> >>>> > Any thougths?
> > >>>>>>> >>>> >
> > >>>>>>> >>>> > 2017-01-31 14:08 GMT+03:00 Pavel Tupitsyn <
> > >>>>>>> [hidden email] <mailto:[hidden email]>>:
> > >>>>>>> >>>> >
> > >>>>>>> >>>> >> Vyacheslav,
> > >>>>>>> >>>> >>
> > >>>>>>> >>>> >> I'm not sure I understand the code you attached.
> > >>>>>>> >>>> >>
> > >>>>>>> >>>> >> If you know how to fix the .NET part, can you just do it in
> > >>>>>>> your PR
> > >>>>>>> >>>> and
> > >>>>>>> >>>> >> run "Platform .NET" on TeamCity to verify?
> > >>>>>>> >>>> >> http://ci.ignite.apache.org/viewType.html?buildTypeId= <http://ci.ignite.apache.org/viewType.html?buildTypeId=>
> > Ignite
> > >>>>>>> >>>> >> Tests_IgnitePlatformNet
> > >>>>>>> >>>> >>
> > >>>>>>> >>>> >> Thanks,
> > >>>>>>> >>>> >>
> > >>>>>>> >>>> >> Pavel
> > >>>>>>> >>>> >>
> > >>>>>>> >>>> >> On Tue, Jan 31, 2017 at 1:35 PM, Vyacheslav Daradur <
> > >>>>>>> >>>> [hidden email] <mailto:[hidden email]>>
> > >>>>>>> >>>> >> wrote:
> > >>>>>>> >>>> >>
> > >>>>>>> >>>> >>> Pavel, I see that you are the main contributor of
> > >>>>>>> Ignite.NET.
> > >>>>>>> >>>> >>>
> > >>>>>>> >>>> >>> We should repair BinaryUtils#ReadDecimal and
> > >>>>>>> >>>> BinaryUtils#WriteDecimal.
> > >>>>>>> >>>> >>>
> > >>>>>>> >>>> >>> *ReadDecimal:
> > >>>>>>> >>>> >>> byte[] mag = ReadByteArray(stream); // including at least
> > >>>>>>> one sign
> > >>>>>>> >>>> bit,
> > >>>>>>> >>>> >>> which is (ceil((this.bitLength() + 1)/8))
> > >>>>>>> >>>> >>> bool neg = (mag[0] < 0);
> > >>>>>>> >>>> >>> if (scale < 0)
> > >>>>>>> >>>> >>> // TODO: a scale of -3 means the unscaled value is
> > >>>>>>> multiplied by
> > >>>>>>> >>>> 1000
> > >>>>>>> >>>> >>>
> > >>>>>>> >>>> >>> *WriteDecimal:
> > >>>>>>> >>>> >>> int sign = vals[3] < 0 ? -1 : 0;
> > >>>>>>> >>>> >>> stream.WriteInt(sign);
> > >>>>>>> >>>> >>>
> > >>>>>>> >>>> >>> Can you help with this task?
> > >>>>>>> >>>> >>>
> > >>>>>>> >>>> >>>
> > >>>>>>> >>>> >>> 2017-01-31 12:46 GMT+03:00 Igor Sapego <
> > >>>>>>> [hidden email] <mailto:[hidden email]>>:
> > >>>>>>> >>>> >>>
> > >>>>>>> >>>> >>>> Vyacheslav,
> > >>>>>>> >>>> >>>>
> > >>>>>>> >>>> >>>> I had a look at your PR and left some comments in Jira.
> > >>>>>>> >>>> >>>>
> > >>>>>>> >>>> >>>> Best Regards,
> > >>>>>>> >>>> >>>> Igor
> > >>>>>>> >>>> >>>>
> > >>>>>>> >>>> >>>> On Mon, Jan 30, 2017 at 12:52 PM, Vyacheslav Daradur <
> > >>>>>>> >>>> >>>> [hidden email] <mailto:[hidden email]>>
> > >>>>>>> >>>> >>>> wrote:
> > >>>>>>> >>>> >>>>
> > >>>>>>> >>>> >>>> > Hello. I fixed it. Please, review.
> > >>>>>>> >>>> >>>> >
> > >>>>>>> >>>> >>>> > https://issues.apache.org/jira/browse/IGNITE-3196 <https://issues.apache.org/jira/browse/IGNITE-3196> -
> > >>>>>>> Marshaling
> > >>>>>>> >>>> works
> > >>>>>>> >>>> >>>> wrong
> > >>>>>>> >>>> >>>> > for the BigDecimals that have negative scale
> > >>>>>>> >>>> >>>> >
> > >>>>>>> >>>> >>>>
> > >>>>>>> >>>> >>>
> > >>>>>>> >>>> >>>
> > >>>>>>> >>>> >>
> > >>>>>>> >>>> >
> > >>>>>>> >>>>
> > >>>>>>> >>>
> > >>>>>>> >>>
> > >>>>>>> >>
> > >>>>>>> >
> > >>>>>>>
> > >>>>>>
> > >>>>>>
> > >>>>>
> > >>>>
> > >>>
> > >>
> > >
> >
>

12