You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@harmony.apache.org by Spark Shen <sm...@gmail.com> on 2006/08/03 03:18:14 UTC
[classlib][luni] Enhance the functionalities provided by BitSet
Hi All:
IMHO, EnumSet provides a set view of enum types. According to its spec,
"Enum sets are represented internally as bit vectors".
To avoid to re-invent the wheel, it could be more reasonable to utilize
java.util.BitSet class to support the implementation of EnumSet.
While current API methods provided by BitSet is not enough, I need to
extend the function provided by BitSet. There are 2 options:
1. Implement these extension methods in BitSet directly as package
private methods
2. Construct a package private utility class in
org.apache.harmony.luni.util package, and implement extensions in this
implementation class.
I prefer the first option, since there is not many classes need these
extension. If no objections, I will adopt the first schema.
Best regards.
--
Spark Shen
China Software Development Lab, IBM
---------------------------------------------------------------------
Terms of use : http://incubator.apache.org/harmony/mailing.html
To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
For additional commands, e-mail: harmony-dev-help@incubator.apache.org
Re: [classlib][luni] Enhance the functionalities provided by BitSet
Posted by Paulex Yang <pa...@gmail.com>.
Spark Shen wrote:
> Richard Liang 写道:
>>
>>
>> Spark Shen wrote:
>>> Hi All:
>>>
>>> IMHO, EnumSet provides a set view of enum types. According to its
>>> spec, "Enum sets are represented internally as bit vectors".
>>> To avoid to re-invent the wheel, it could be more reasonable to
>>> utilize java.util.BitSet class to support the implementation of
>>> EnumSet.
>>>
>> Agree :-)
>>
>>> While current API methods provided by BitSet is not enough, I need
>>> to extend the function provided by BitSet. There are 2 options:
>>> 1. Implement these extension methods in BitSet directly as package
>>> private methods
>>> 2. Construct a package private utility class in
>>> org.apache.harmony.luni.util package, and implement extensions in
>>> this implementation class.
>>>
>>> I prefer the first option, since there is not many classes need
>>> these extension. If no objections, I will adopt the first schema.
>>>
>> IMHO, it depends on the relationship between the extended functions
>> and the whole class. Would you please provide more detail
>> information? Thanks a lot.
> 1. It can be simply detected that on RI, EnumSet.noneOf() or
> EnumSet.allOf() method will return a concrete EnumSet. This concrete
> EnumSet provides its own iterator.
> All these iterator related methods are not implemented in BitSet right
> now.
>
> 2. To support the implementation of EnumSet, more utility methods used
> to set/unset bits may need to be added.
>
> All these requirements are EnumSet specific, do they need to be
> implemented in a separate class.
If so (the iterator or so has no value for other classes), I agree to
use a specific subclass of BitSet for EnumSet's requirement.
>
> Best regards
>>
>> Richard.
>>> Best regards.
>>>
>>
>
>
--
Paulex Yang
China Software Development Lab
IBM
---------------------------------------------------------------------
Terms of use : http://incubator.apache.org/harmony/mailing.html
To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
For additional commands, e-mail: harmony-dev-help@incubator.apache.org
Re: [classlib][luni] Enhance the functionalities provided by BitSet
Posted by Oleg Khaschansky <ol...@gmail.com>.
> That is a good idea. But some guru says developers are used to optimize
> code at inappropriate spot. :-) . Kidding
Microbenchmark may help to make right decision :)
On 8/4/06, Spark Shen <sm...@gmail.com> wrote:
> Oleg Khaschansky 写道:
> >> Do you mean to write a specific BitSet using one long type as its
> >> internal storage to deal
> >> with <=64 elements enums?
> >
> > No, I suggest to implement a specific EnumSet using one long type. My
> > opinion is that a specific <=64 BitSet wouldn't be very useful.
> Thank you for your suggestion.
>
> You mean to implement a specific EnumSet
> without wrapping BitSet when elements in enum is <=64.
> That is a good idea. But some guru says developers are used to optimize
> code at inappropriate spot. :-) . Kidding
>
> I will consider your suggestion seriously..
>
> Best regards
> >
> > On 8/4/06, Spark Shen <sm...@gmail.com> wrote:
> >> Oleg Khaschansky 写道:
> >> > Yes, I mean exactly one long type, not array. And I think that for
> >> > this case the code in the BitSet and EnumSet implementation would be
> >> > pretty simplier and then faster.
> >> Do you mean to write a specific BitSet using one long type as its
> >> internal storage to deal
> >> with <=64 elements enums?
> >> > But the <=64 bit BitSet looks not
> >> > very useful, so I don't suggest to optimize it. At the same time <=64
> >> > elemnts enums are, probably, used much more often and EnumSet could be
> >> > optimized in this way. It is just a suggestion, hope that it could be
> >> > useful.
> >> Thank you very much for your suggestion.
> >>
> >> Best regards.
> >> >
> >> > On 8/3/06, Spark Shen <sm...@gmail.com> wrote:
> >> >> Oleg Khaschansky 写道:
> >> >> > Hi all!
> >> >> >
> >> >> > I'd like to add to this discussion that, IMHO, implementation of
> >> the
> >> >> > EnumSet
> >> >> > based on the BitSet class, probably, would be relatively slow.
> >> >> I can not fully agree with you here.
> >> >>
> >> >> IIRC, BitSet already uses long array to store
> >> >> bits instead of Byte[] in our implementation. Every bit in any long
> >> >> of the
> >> >> long array corresponds to a enum type element.
> >> >> IMHO, this is the most effcient way to store bits. The only down
> >> side of
> >> >> using BitSet is that there are extra method invokations.
> >> >> > I'd suggest
> >> >> > to handle extra case when the number of elements in the enum is
> >> <= 64
> >> >> > and it
> >> >> > is possible to represent the set with one long value. For this case
> >> >> > all the
> >> >> > EnumSet API could be implemented using fast bitwise operations.
> >> >> Since the long array storage, no matter the number of elements in an
> >> >> enum type, fast bitwise operations can
> >> >> be applied. Do you mean to do extra optimization when number of
> >> elements
> >> >> in enum type is less than 64. Under
> >> >> this condition, may be we can use one long type instead of a long
> >> array.
> >> >> But will this kind of optimization bring any bonus
> >> >> is unknown right now.
> >> >>
> >> >> Correct me if I am wrong.
> >> >>
> >> >> Best regards
> >> >> >
> >> >> > Thank you.
> >> >> > --
> >> >> > Oleg Khaschansky,
> >> >> > Intel MPD
> >> >> >
> >> >> > On 8/3/06, Spark Shen <sm...@gmail.com> wrote:
> >> >> >>
> >> >> >> Richard Liang 写道:
> >> >> >> >
> >> >> >> >
> >> >> >> > Spark Shen wrote:
> >> >> >> >> Hi All:
> >> >> >> >>
> >> >> >> >> IMHO, EnumSet provides a set view of enum types. According
> >> to its
> >> >> >> >> spec, "Enum sets are represented internally as bit vectors".
> >> >> >> >> To avoid to re-invent the wheel, it could be more reasonable to
> >> >> >> >> utilize java.util.BitSet class to support the implementation of
> >> >> >> EnumSet.
> >> >> >> >>
> >> >> >> > Agree :-)
> >> >> >> >
> >> >> >> >> While current API methods provided by BitSet is not enough, I
> >> >> need to
> >> >> >> >> extend the function provided by BitSet. There are 2 options:
> >> >> >> >> 1. Implement these extension methods in BitSet directly as
> >> package
> >> >> >> >> private methods
> >> >> >> >> 2. Construct a package private utility class in
> >> >> >> >> org.apache.harmony.luni.util package, and implement
> >> extensions in
> >> >> >> >> this implementation class.
> >> >> >> >>
> >> >> >> >> I prefer the first option, since there is not many classes need
> >> >> these
> >> >> >> >> extension. If no objections, I will adopt the first schema.
> >> >> >> >>
> >> >> >> > IMHO, it depends on the relationship between the extended
> >> functions
> >> >> >> > and the whole class. Would you please provide more detail
> >> >> information?
> >> >> >> > Thanks a lot.
> >> >> >> 1. It can be simply detected that on RI, EnumSet.noneOf() or
> >> >> >> EnumSet.allOf() method will return a concrete EnumSet. This
> >> concrete
> >> >> >> EnumSet provides its own iterator.
> >> >> >> All these iterator related methods are not implemented in BitSet
> >> >> right
> >> >> >> now.
> >> >> >>
> >> >> >> 2. To support the implementation of EnumSet, more utility methods
> >> >> used
> >> >> >> to set/unset bits may need to be added.
> >> >> >>
> >> >> >> All these requirements are EnumSet specific, do they need to be
> >> >> >> implemented in a separate class.
> >> >> >>
> >> >> >> Best regards
> >> >> >> >
> >> >> >> > Richard.
> >> >> >> >> Best regards.
> >> >> >> >>
> >> >> >> >
> >> >> >>
> >> >> >>
> >> >> >> --
> >> >> >> Spark Shen
> >> >> >> China Software Development Lab, IBM
> >> >> >>
> >> >> >>
> >> >> >>
> >> >> >>
> >> ---------------------------------------------------------------------
> >> >> >> Terms of use : http://incubator.apache.org/harmony/mailing.html
> >> >> >> To unsubscribe, e-mail:
> >> harmony-dev-unsubscribe@incubator.apache.org
> >> >> >> For additional commands, e-mail:
> >> >> harmony-dev-help@incubator.apache.org
> >> >> >>
> >> >> >>
> >> >> >
> >> >>
> >> >>
> >> >> --
> >> >> Spark Shen
> >> >> China Software Development Lab, IBM
> >> >>
> >> >>
> >> >>
> >> >> ---------------------------------------------------------------------
> >> >> Terms of use : http://incubator.apache.org/harmony/mailing.html
> >> >> To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
> >> >> For additional commands, e-mail:
> >> harmony-dev-help@incubator.apache.org
> >> >>
> >> >>
> >> >
> >> > ---------------------------------------------------------------------
> >> > Terms of use : http://incubator.apache.org/harmony/mailing.html
> >> > To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
> >> > For additional commands, e-mail: harmony-dev-help@incubator.apache.org
> >> >
> >> >
> >>
> >>
> >> --
> >> Spark Shen
> >> China Software Development Lab, IBM
> >>
> >>
> >>
> >> ---------------------------------------------------------------------
> >> Terms of use : http://incubator.apache.org/harmony/mailing.html
> >> To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
> >> For additional commands, e-mail: harmony-dev-help@incubator.apache.org
> >>
> >>
> >
> > ---------------------------------------------------------------------
> > Terms of use : http://incubator.apache.org/harmony/mailing.html
> > To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
> > For additional commands, e-mail: harmony-dev-help@incubator.apache.org
> >
> >
>
>
> --
> Spark Shen
> China Software Development Lab, IBM
>
>
>
> ---------------------------------------------------------------------
> Terms of use : http://incubator.apache.org/harmony/mailing.html
> To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
> For additional commands, e-mail: harmony-dev-help@incubator.apache.org
>
>
---------------------------------------------------------------------
Terms of use : http://incubator.apache.org/harmony/mailing.html
To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
For additional commands, e-mail: harmony-dev-help@incubator.apache.org
Re: [classlib][luni] Enhance the functionalities provided by BitSet
Posted by Spark Shen <sm...@gmail.com>.
Oleg Khaschansky 写道:
>> Do you mean to write a specific BitSet using one long type as its
>> internal storage to deal
>> with <=64 elements enums?
>
> No, I suggest to implement a specific EnumSet using one long type. My
> opinion is that a specific <=64 BitSet wouldn't be very useful.
Thank you for your suggestion.
You mean to implement a specific EnumSet
without wrapping BitSet when elements in enum is <=64.
That is a good idea. But some guru says developers are used to optimize
code at inappropriate spot. :-) . Kidding
I will consider your suggestion seriously..
Best regards
>
> On 8/4/06, Spark Shen <sm...@gmail.com> wrote:
>> Oleg Khaschansky 写道:
>> > Yes, I mean exactly one long type, not array. And I think that for
>> > this case the code in the BitSet and EnumSet implementation would be
>> > pretty simplier and then faster.
>> Do you mean to write a specific BitSet using one long type as its
>> internal storage to deal
>> with <=64 elements enums?
>> > But the <=64 bit BitSet looks not
>> > very useful, so I don't suggest to optimize it. At the same time <=64
>> > elemnts enums are, probably, used much more often and EnumSet could be
>> > optimized in this way. It is just a suggestion, hope that it could be
>> > useful.
>> Thank you very much for your suggestion.
>>
>> Best regards.
>> >
>> > On 8/3/06, Spark Shen <sm...@gmail.com> wrote:
>> >> Oleg Khaschansky 写道:
>> >> > Hi all!
>> >> >
>> >> > I'd like to add to this discussion that, IMHO, implementation of
>> the
>> >> > EnumSet
>> >> > based on the BitSet class, probably, would be relatively slow.
>> >> I can not fully agree with you here.
>> >>
>> >> IIRC, BitSet already uses long array to store
>> >> bits instead of Byte[] in our implementation. Every bit in any long
>> >> of the
>> >> long array corresponds to a enum type element.
>> >> IMHO, this is the most effcient way to store bits. The only down
>> side of
>> >> using BitSet is that there are extra method invokations.
>> >> > I'd suggest
>> >> > to handle extra case when the number of elements in the enum is
>> <= 64
>> >> > and it
>> >> > is possible to represent the set with one long value. For this case
>> >> > all the
>> >> > EnumSet API could be implemented using fast bitwise operations.
>> >> Since the long array storage, no matter the number of elements in an
>> >> enum type, fast bitwise operations can
>> >> be applied. Do you mean to do extra optimization when number of
>> elements
>> >> in enum type is less than 64. Under
>> >> this condition, may be we can use one long type instead of a long
>> array.
>> >> But will this kind of optimization bring any bonus
>> >> is unknown right now.
>> >>
>> >> Correct me if I am wrong.
>> >>
>> >> Best regards
>> >> >
>> >> > Thank you.
>> >> > --
>> >> > Oleg Khaschansky,
>> >> > Intel MPD
>> >> >
>> >> > On 8/3/06, Spark Shen <sm...@gmail.com> wrote:
>> >> >>
>> >> >> Richard Liang 写道:
>> >> >> >
>> >> >> >
>> >> >> > Spark Shen wrote:
>> >> >> >> Hi All:
>> >> >> >>
>> >> >> >> IMHO, EnumSet provides a set view of enum types. According
>> to its
>> >> >> >> spec, "Enum sets are represented internally as bit vectors".
>> >> >> >> To avoid to re-invent the wheel, it could be more reasonable to
>> >> >> >> utilize java.util.BitSet class to support the implementation of
>> >> >> EnumSet.
>> >> >> >>
>> >> >> > Agree :-)
>> >> >> >
>> >> >> >> While current API methods provided by BitSet is not enough, I
>> >> need to
>> >> >> >> extend the function provided by BitSet. There are 2 options:
>> >> >> >> 1. Implement these extension methods in BitSet directly as
>> package
>> >> >> >> private methods
>> >> >> >> 2. Construct a package private utility class in
>> >> >> >> org.apache.harmony.luni.util package, and implement
>> extensions in
>> >> >> >> this implementation class.
>> >> >> >>
>> >> >> >> I prefer the first option, since there is not many classes need
>> >> these
>> >> >> >> extension. If no objections, I will adopt the first schema.
>> >> >> >>
>> >> >> > IMHO, it depends on the relationship between the extended
>> functions
>> >> >> > and the whole class. Would you please provide more detail
>> >> information?
>> >> >> > Thanks a lot.
>> >> >> 1. It can be simply detected that on RI, EnumSet.noneOf() or
>> >> >> EnumSet.allOf() method will return a concrete EnumSet. This
>> concrete
>> >> >> EnumSet provides its own iterator.
>> >> >> All these iterator related methods are not implemented in BitSet
>> >> right
>> >> >> now.
>> >> >>
>> >> >> 2. To support the implementation of EnumSet, more utility methods
>> >> used
>> >> >> to set/unset bits may need to be added.
>> >> >>
>> >> >> All these requirements are EnumSet specific, do they need to be
>> >> >> implemented in a separate class.
>> >> >>
>> >> >> Best regards
>> >> >> >
>> >> >> > Richard.
>> >> >> >> Best regards.
>> >> >> >>
>> >> >> >
>> >> >>
>> >> >>
>> >> >> --
>> >> >> Spark Shen
>> >> >> China Software Development Lab, IBM
>> >> >>
>> >> >>
>> >> >>
>> >> >>
>> ---------------------------------------------------------------------
>> >> >> Terms of use : http://incubator.apache.org/harmony/mailing.html
>> >> >> To unsubscribe, e-mail:
>> harmony-dev-unsubscribe@incubator.apache.org
>> >> >> For additional commands, e-mail:
>> >> harmony-dev-help@incubator.apache.org
>> >> >>
>> >> >>
>> >> >
>> >>
>> >>
>> >> --
>> >> Spark Shen
>> >> China Software Development Lab, IBM
>> >>
>> >>
>> >>
>> >> ---------------------------------------------------------------------
>> >> Terms of use : http://incubator.apache.org/harmony/mailing.html
>> >> To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
>> >> For additional commands, e-mail:
>> harmony-dev-help@incubator.apache.org
>> >>
>> >>
>> >
>> > ---------------------------------------------------------------------
>> > Terms of use : http://incubator.apache.org/harmony/mailing.html
>> > To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
>> > For additional commands, e-mail: harmony-dev-help@incubator.apache.org
>> >
>> >
>>
>>
>> --
>> Spark Shen
>> China Software Development Lab, IBM
>>
>>
>>
>> ---------------------------------------------------------------------
>> Terms of use : http://incubator.apache.org/harmony/mailing.html
>> To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
>> For additional commands, e-mail: harmony-dev-help@incubator.apache.org
>>
>>
>
> ---------------------------------------------------------------------
> Terms of use : http://incubator.apache.org/harmony/mailing.html
> To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
> For additional commands, e-mail: harmony-dev-help@incubator.apache.org
>
>
--
Spark Shen
China Software Development Lab, IBM
---------------------------------------------------------------------
Terms of use : http://incubator.apache.org/harmony/mailing.html
To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
For additional commands, e-mail: harmony-dev-help@incubator.apache.org
Re: [classlib][luni] Enhance the functionalities provided by BitSet
Posted by Tim Ellison <t....@gmail.com>.
Oleg Khaschansky wrote:
>> Do you mean to write a specific BitSet using one long type as its
>> internal storage to deal
>> with <=64 elements enums?
>
> No, I suggest to implement a specific EnumSet using one long type. My
> opinion is that a specific <=64 BitSet wouldn't be very useful.
Well it would be useful in this instance, and wouldn't do any harm to
the general use of BitSet, so why not?
Regards,
Tim
--
Tim Ellison (t.p.ellison@gmail.com)
IBM Java technology centre, UK.
---------------------------------------------------------------------
Terms of use : http://incubator.apache.org/harmony/mailing.html
To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
For additional commands, e-mail: harmony-dev-help@incubator.apache.org
Re: [classlib][luni] Enhance the functionalities provided by BitSet
Posted by Oleg Khaschansky <ol...@gmail.com>.
> Do you mean to write a specific BitSet using one long type as its
> internal storage to deal
> with <=64 elements enums?
No, I suggest to implement a specific EnumSet using one long type. My
opinion is that a specific <=64 BitSet wouldn't be very useful.
On 8/4/06, Spark Shen <sm...@gmail.com> wrote:
> Oleg Khaschansky 写道:
> > Yes, I mean exactly one long type, not array. And I think that for
> > this case the code in the BitSet and EnumSet implementation would be
> > pretty simplier and then faster.
> Do you mean to write a specific BitSet using one long type as its
> internal storage to deal
> with <=64 elements enums?
> > But the <=64 bit BitSet looks not
> > very useful, so I don't suggest to optimize it. At the same time <=64
> > elemnts enums are, probably, used much more often and EnumSet could be
> > optimized in this way. It is just a suggestion, hope that it could be
> > useful.
> Thank you very much for your suggestion.
>
> Best regards.
> >
> > On 8/3/06, Spark Shen <sm...@gmail.com> wrote:
> >> Oleg Khaschansky 写道:
> >> > Hi all!
> >> >
> >> > I'd like to add to this discussion that, IMHO, implementation of the
> >> > EnumSet
> >> > based on the BitSet class, probably, would be relatively slow.
> >> I can not fully agree with you here.
> >>
> >> IIRC, BitSet already uses long array to store
> >> bits instead of Byte[] in our implementation. Every bit in any long
> >> of the
> >> long array corresponds to a enum type element.
> >> IMHO, this is the most effcient way to store bits. The only down side of
> >> using BitSet is that there are extra method invokations.
> >> > I'd suggest
> >> > to handle extra case when the number of elements in the enum is <= 64
> >> > and it
> >> > is possible to represent the set with one long value. For this case
> >> > all the
> >> > EnumSet API could be implemented using fast bitwise operations.
> >> Since the long array storage, no matter the number of elements in an
> >> enum type, fast bitwise operations can
> >> be applied. Do you mean to do extra optimization when number of elements
> >> in enum type is less than 64. Under
> >> this condition, may be we can use one long type instead of a long array.
> >> But will this kind of optimization bring any bonus
> >> is unknown right now.
> >>
> >> Correct me if I am wrong.
> >>
> >> Best regards
> >> >
> >> > Thank you.
> >> > --
> >> > Oleg Khaschansky,
> >> > Intel MPD
> >> >
> >> > On 8/3/06, Spark Shen <sm...@gmail.com> wrote:
> >> >>
> >> >> Richard Liang 写道:
> >> >> >
> >> >> >
> >> >> > Spark Shen wrote:
> >> >> >> Hi All:
> >> >> >>
> >> >> >> IMHO, EnumSet provides a set view of enum types. According to its
> >> >> >> spec, "Enum sets are represented internally as bit vectors".
> >> >> >> To avoid to re-invent the wheel, it could be more reasonable to
> >> >> >> utilize java.util.BitSet class to support the implementation of
> >> >> EnumSet.
> >> >> >>
> >> >> > Agree :-)
> >> >> >
> >> >> >> While current API methods provided by BitSet is not enough, I
> >> need to
> >> >> >> extend the function provided by BitSet. There are 2 options:
> >> >> >> 1. Implement these extension methods in BitSet directly as package
> >> >> >> private methods
> >> >> >> 2. Construct a package private utility class in
> >> >> >> org.apache.harmony.luni.util package, and implement extensions in
> >> >> >> this implementation class.
> >> >> >>
> >> >> >> I prefer the first option, since there is not many classes need
> >> these
> >> >> >> extension. If no objections, I will adopt the first schema.
> >> >> >>
> >> >> > IMHO, it depends on the relationship between the extended functions
> >> >> > and the whole class. Would you please provide more detail
> >> information?
> >> >> > Thanks a lot.
> >> >> 1. It can be simply detected that on RI, EnumSet.noneOf() or
> >> >> EnumSet.allOf() method will return a concrete EnumSet. This concrete
> >> >> EnumSet provides its own iterator.
> >> >> All these iterator related methods are not implemented in BitSet
> >> right
> >> >> now.
> >> >>
> >> >> 2. To support the implementation of EnumSet, more utility methods
> >> used
> >> >> to set/unset bits may need to be added.
> >> >>
> >> >> All these requirements are EnumSet specific, do they need to be
> >> >> implemented in a separate class.
> >> >>
> >> >> Best regards
> >> >> >
> >> >> > Richard.
> >> >> >> Best regards.
> >> >> >>
> >> >> >
> >> >>
> >> >>
> >> >> --
> >> >> Spark Shen
> >> >> China Software Development Lab, IBM
> >> >>
> >> >>
> >> >>
> >> >> ---------------------------------------------------------------------
> >> >> Terms of use : http://incubator.apache.org/harmony/mailing.html
> >> >> To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
> >> >> For additional commands, e-mail:
> >> harmony-dev-help@incubator.apache.org
> >> >>
> >> >>
> >> >
> >>
> >>
> >> --
> >> Spark Shen
> >> China Software Development Lab, IBM
> >>
> >>
> >>
> >> ---------------------------------------------------------------------
> >> Terms of use : http://incubator.apache.org/harmony/mailing.html
> >> To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
> >> For additional commands, e-mail: harmony-dev-help@incubator.apache.org
> >>
> >>
> >
> > ---------------------------------------------------------------------
> > Terms of use : http://incubator.apache.org/harmony/mailing.html
> > To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
> > For additional commands, e-mail: harmony-dev-help@incubator.apache.org
> >
> >
>
>
> --
> Spark Shen
> China Software Development Lab, IBM
>
>
>
> ---------------------------------------------------------------------
> Terms of use : http://incubator.apache.org/harmony/mailing.html
> To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
> For additional commands, e-mail: harmony-dev-help@incubator.apache.org
>
>
---------------------------------------------------------------------
Terms of use : http://incubator.apache.org/harmony/mailing.html
To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
For additional commands, e-mail: harmony-dev-help@incubator.apache.org
Re: [classlib][luni] Enhance the functionalities provided by BitSet
Posted by Spark Shen <sm...@gmail.com>.
Oleg Khaschansky 写道:
> Yes, I mean exactly one long type, not array. And I think that for
> this case the code in the BitSet and EnumSet implementation would be
> pretty simplier and then faster.
Do you mean to write a specific BitSet using one long type as its
internal storage to deal
with <=64 elements enums?
> But the <=64 bit BitSet looks not
> very useful, so I don't suggest to optimize it. At the same time <=64
> elemnts enums are, probably, used much more often and EnumSet could be
> optimized in this way. It is just a suggestion, hope that it could be
> useful.
Thank you very much for your suggestion.
Best regards.
>
> On 8/3/06, Spark Shen <sm...@gmail.com> wrote:
>> Oleg Khaschansky 写道:
>> > Hi all!
>> >
>> > I'd like to add to this discussion that, IMHO, implementation of the
>> > EnumSet
>> > based on the BitSet class, probably, would be relatively slow.
>> I can not fully agree with you here.
>>
>> IIRC, BitSet already uses long array to store
>> bits instead of Byte[] in our implementation. Every bit in any long
>> of the
>> long array corresponds to a enum type element.
>> IMHO, this is the most effcient way to store bits. The only down side of
>> using BitSet is that there are extra method invokations.
>> > I'd suggest
>> > to handle extra case when the number of elements in the enum is <= 64
>> > and it
>> > is possible to represent the set with one long value. For this case
>> > all the
>> > EnumSet API could be implemented using fast bitwise operations.
>> Since the long array storage, no matter the number of elements in an
>> enum type, fast bitwise operations can
>> be applied. Do you mean to do extra optimization when number of elements
>> in enum type is less than 64. Under
>> this condition, may be we can use one long type instead of a long array.
>> But will this kind of optimization bring any bonus
>> is unknown right now.
>>
>> Correct me if I am wrong.
>>
>> Best regards
>> >
>> > Thank you.
>> > --
>> > Oleg Khaschansky,
>> > Intel MPD
>> >
>> > On 8/3/06, Spark Shen <sm...@gmail.com> wrote:
>> >>
>> >> Richard Liang 写道:
>> >> >
>> >> >
>> >> > Spark Shen wrote:
>> >> >> Hi All:
>> >> >>
>> >> >> IMHO, EnumSet provides a set view of enum types. According to its
>> >> >> spec, "Enum sets are represented internally as bit vectors".
>> >> >> To avoid to re-invent the wheel, it could be more reasonable to
>> >> >> utilize java.util.BitSet class to support the implementation of
>> >> EnumSet.
>> >> >>
>> >> > Agree :-)
>> >> >
>> >> >> While current API methods provided by BitSet is not enough, I
>> need to
>> >> >> extend the function provided by BitSet. There are 2 options:
>> >> >> 1. Implement these extension methods in BitSet directly as package
>> >> >> private methods
>> >> >> 2. Construct a package private utility class in
>> >> >> org.apache.harmony.luni.util package, and implement extensions in
>> >> >> this implementation class.
>> >> >>
>> >> >> I prefer the first option, since there is not many classes need
>> these
>> >> >> extension. If no objections, I will adopt the first schema.
>> >> >>
>> >> > IMHO, it depends on the relationship between the extended functions
>> >> > and the whole class. Would you please provide more detail
>> information?
>> >> > Thanks a lot.
>> >> 1. It can be simply detected that on RI, EnumSet.noneOf() or
>> >> EnumSet.allOf() method will return a concrete EnumSet. This concrete
>> >> EnumSet provides its own iterator.
>> >> All these iterator related methods are not implemented in BitSet
>> right
>> >> now.
>> >>
>> >> 2. To support the implementation of EnumSet, more utility methods
>> used
>> >> to set/unset bits may need to be added.
>> >>
>> >> All these requirements are EnumSet specific, do they need to be
>> >> implemented in a separate class.
>> >>
>> >> Best regards
>> >> >
>> >> > Richard.
>> >> >> Best regards.
>> >> >>
>> >> >
>> >>
>> >>
>> >> --
>> >> Spark Shen
>> >> China Software Development Lab, IBM
>> >>
>> >>
>> >>
>> >> ---------------------------------------------------------------------
>> >> Terms of use : http://incubator.apache.org/harmony/mailing.html
>> >> To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
>> >> For additional commands, e-mail:
>> harmony-dev-help@incubator.apache.org
>> >>
>> >>
>> >
>>
>>
>> --
>> Spark Shen
>> China Software Development Lab, IBM
>>
>>
>>
>> ---------------------------------------------------------------------
>> Terms of use : http://incubator.apache.org/harmony/mailing.html
>> To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
>> For additional commands, e-mail: harmony-dev-help@incubator.apache.org
>>
>>
>
> ---------------------------------------------------------------------
> Terms of use : http://incubator.apache.org/harmony/mailing.html
> To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
> For additional commands, e-mail: harmony-dev-help@incubator.apache.org
>
>
--
Spark Shen
China Software Development Lab, IBM
---------------------------------------------------------------------
Terms of use : http://incubator.apache.org/harmony/mailing.html
To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
For additional commands, e-mail: harmony-dev-help@incubator.apache.org
Re: [classlib][luni] Enhance the functionalities provided by BitSet
Posted by Oleg Khaschansky <ol...@gmail.com>.
Yes, I mean exactly one long type, not array. And I think that for
this case the code in the BitSet and EnumSet implementation would be
pretty simplier and then faster. But the <=64 bit BitSet looks not
very useful, so I don't suggest to optimize it. At the same time <=64
elemnts enums are, probably, used much more often and EnumSet could be
optimized in this way. It is just a suggestion, hope that it could be
useful.
On 8/3/06, Spark Shen <sm...@gmail.com> wrote:
> Oleg Khaschansky 写道:
> > Hi all!
> >
> > I'd like to add to this discussion that, IMHO, implementation of the
> > EnumSet
> > based on the BitSet class, probably, would be relatively slow.
> I can not fully agree with you here.
>
> IIRC, BitSet already uses long array to store
> bits instead of Byte[] in our implementation. Every bit in any long of the
> long array corresponds to a enum type element.
> IMHO, this is the most effcient way to store bits. The only down side of
> using BitSet is that there are extra method invokations.
> > I'd suggest
> > to handle extra case when the number of elements in the enum is <= 64
> > and it
> > is possible to represent the set with one long value. For this case
> > all the
> > EnumSet API could be implemented using fast bitwise operations.
> Since the long array storage, no matter the number of elements in an
> enum type, fast bitwise operations can
> be applied. Do you mean to do extra optimization when number of elements
> in enum type is less than 64. Under
> this condition, may be we can use one long type instead of a long array.
> But will this kind of optimization bring any bonus
> is unknown right now.
>
> Correct me if I am wrong.
>
> Best regards
> >
> > Thank you.
> > --
> > Oleg Khaschansky,
> > Intel MPD
> >
> > On 8/3/06, Spark Shen <sm...@gmail.com> wrote:
> >>
> >> Richard Liang 写道:
> >> >
> >> >
> >> > Spark Shen wrote:
> >> >> Hi All:
> >> >>
> >> >> IMHO, EnumSet provides a set view of enum types. According to its
> >> >> spec, "Enum sets are represented internally as bit vectors".
> >> >> To avoid to re-invent the wheel, it could be more reasonable to
> >> >> utilize java.util.BitSet class to support the implementation of
> >> EnumSet.
> >> >>
> >> > Agree :-)
> >> >
> >> >> While current API methods provided by BitSet is not enough, I need to
> >> >> extend the function provided by BitSet. There are 2 options:
> >> >> 1. Implement these extension methods in BitSet directly as package
> >> >> private methods
> >> >> 2. Construct a package private utility class in
> >> >> org.apache.harmony.luni.util package, and implement extensions in
> >> >> this implementation class.
> >> >>
> >> >> I prefer the first option, since there is not many classes need these
> >> >> extension. If no objections, I will adopt the first schema.
> >> >>
> >> > IMHO, it depends on the relationship between the extended functions
> >> > and the whole class. Would you please provide more detail information?
> >> > Thanks a lot.
> >> 1. It can be simply detected that on RI, EnumSet.noneOf() or
> >> EnumSet.allOf() method will return a concrete EnumSet. This concrete
> >> EnumSet provides its own iterator.
> >> All these iterator related methods are not implemented in BitSet right
> >> now.
> >>
> >> 2. To support the implementation of EnumSet, more utility methods used
> >> to set/unset bits may need to be added.
> >>
> >> All these requirements are EnumSet specific, do they need to be
> >> implemented in a separate class.
> >>
> >> Best regards
> >> >
> >> > Richard.
> >> >> Best regards.
> >> >>
> >> >
> >>
> >>
> >> --
> >> Spark Shen
> >> China Software Development Lab, IBM
> >>
> >>
> >>
> >> ---------------------------------------------------------------------
> >> Terms of use : http://incubator.apache.org/harmony/mailing.html
> >> To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
> >> For additional commands, e-mail: harmony-dev-help@incubator.apache.org
> >>
> >>
> >
>
>
> --
> Spark Shen
> China Software Development Lab, IBM
>
>
>
> ---------------------------------------------------------------------
> Terms of use : http://incubator.apache.org/harmony/mailing.html
> To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
> For additional commands, e-mail: harmony-dev-help@incubator.apache.org
>
>
---------------------------------------------------------------------
Terms of use : http://incubator.apache.org/harmony/mailing.html
To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
For additional commands, e-mail: harmony-dev-help@incubator.apache.org
Re: [classlib][luni] Enhance the functionalities provided by BitSet
Posted by Spark Shen <sm...@gmail.com>.
Oleg Khaschansky 写道:
> Hi all!
>
> I'd like to add to this discussion that, IMHO, implementation of the
> EnumSet
> based on the BitSet class, probably, would be relatively slow.
I can not fully agree with you here.
IIRC, BitSet already uses long array to store
bits instead of Byte[] in our implementation. Every bit in any long of the
long array corresponds to a enum type element.
IMHO, this is the most effcient way to store bits. The only down side of
using BitSet is that there are extra method invokations.
> I'd suggest
> to handle extra case when the number of elements in the enum is <= 64
> and it
> is possible to represent the set with one long value. For this case
> all the
> EnumSet API could be implemented using fast bitwise operations.
Since the long array storage, no matter the number of elements in an
enum type, fast bitwise operations can
be applied. Do you mean to do extra optimization when number of elements
in enum type is less than 64. Under
this condition, may be we can use one long type instead of a long array.
But will this kind of optimization bring any bonus
is unknown right now.
Correct me if I am wrong.
Best regards
>
> Thank you.
> --
> Oleg Khaschansky,
> Intel MPD
>
> On 8/3/06, Spark Shen <sm...@gmail.com> wrote:
>>
>> Richard Liang 写道:
>> >
>> >
>> > Spark Shen wrote:
>> >> Hi All:
>> >>
>> >> IMHO, EnumSet provides a set view of enum types. According to its
>> >> spec, "Enum sets are represented internally as bit vectors".
>> >> To avoid to re-invent the wheel, it could be more reasonable to
>> >> utilize java.util.BitSet class to support the implementation of
>> EnumSet.
>> >>
>> > Agree :-)
>> >
>> >> While current API methods provided by BitSet is not enough, I need to
>> >> extend the function provided by BitSet. There are 2 options:
>> >> 1. Implement these extension methods in BitSet directly as package
>> >> private methods
>> >> 2. Construct a package private utility class in
>> >> org.apache.harmony.luni.util package, and implement extensions in
>> >> this implementation class.
>> >>
>> >> I prefer the first option, since there is not many classes need these
>> >> extension. If no objections, I will adopt the first schema.
>> >>
>> > IMHO, it depends on the relationship between the extended functions
>> > and the whole class. Would you please provide more detail information?
>> > Thanks a lot.
>> 1. It can be simply detected that on RI, EnumSet.noneOf() or
>> EnumSet.allOf() method will return a concrete EnumSet. This concrete
>> EnumSet provides its own iterator.
>> All these iterator related methods are not implemented in BitSet right
>> now.
>>
>> 2. To support the implementation of EnumSet, more utility methods used
>> to set/unset bits may need to be added.
>>
>> All these requirements are EnumSet specific, do they need to be
>> implemented in a separate class.
>>
>> Best regards
>> >
>> > Richard.
>> >> Best regards.
>> >>
>> >
>>
>>
>> --
>> Spark Shen
>> China Software Development Lab, IBM
>>
>>
>>
>> ---------------------------------------------------------------------
>> Terms of use : http://incubator.apache.org/harmony/mailing.html
>> To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
>> For additional commands, e-mail: harmony-dev-help@incubator.apache.org
>>
>>
>
--
Spark Shen
China Software Development Lab, IBM
---------------------------------------------------------------------
Terms of use : http://incubator.apache.org/harmony/mailing.html
To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
For additional commands, e-mail: harmony-dev-help@incubator.apache.org
Re: [classlib][luni] Enhance the functionalities provided by BitSet
Posted by Oleg Khaschansky <ol...@gmail.com>.
Hi all!
I'd like to add to this discussion that, IMHO, implementation of the EnumSet
based on the BitSet class, probably, would be relatively slow. I'd suggest
to handle extra case when the number of elements in the enum is <= 64 and it
is possible to represent the set with one long value. For this case all the
EnumSet API could be implemented using fast bitwise operations.
Thank you.
--
Oleg Khaschansky,
Intel MPD
On 8/3/06, Spark Shen <sm...@gmail.com> wrote:
>
> Richard Liang 写道:
> >
> >
> > Spark Shen wrote:
> >> Hi All:
> >>
> >> IMHO, EnumSet provides a set view of enum types. According to its
> >> spec, "Enum sets are represented internally as bit vectors".
> >> To avoid to re-invent the wheel, it could be more reasonable to
> >> utilize java.util.BitSet class to support the implementation of
> EnumSet.
> >>
> > Agree :-)
> >
> >> While current API methods provided by BitSet is not enough, I need to
> >> extend the function provided by BitSet. There are 2 options:
> >> 1. Implement these extension methods in BitSet directly as package
> >> private methods
> >> 2. Construct a package private utility class in
> >> org.apache.harmony.luni.util package, and implement extensions in
> >> this implementation class.
> >>
> >> I prefer the first option, since there is not many classes need these
> >> extension. If no objections, I will adopt the first schema.
> >>
> > IMHO, it depends on the relationship between the extended functions
> > and the whole class. Would you please provide more detail information?
> > Thanks a lot.
> 1. It can be simply detected that on RI, EnumSet.noneOf() or
> EnumSet.allOf() method will return a concrete EnumSet. This concrete
> EnumSet provides its own iterator.
> All these iterator related methods are not implemented in BitSet right
> now.
>
> 2. To support the implementation of EnumSet, more utility methods used
> to set/unset bits may need to be added.
>
> All these requirements are EnumSet specific, do they need to be
> implemented in a separate class.
>
> Best regards
> >
> > Richard.
> >> Best regards.
> >>
> >
>
>
> --
> Spark Shen
> China Software Development Lab, IBM
>
>
>
> ---------------------------------------------------------------------
> Terms of use : http://incubator.apache.org/harmony/mailing.html
> To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
> For additional commands, e-mail: harmony-dev-help@incubator.apache.org
>
>
Re: [classlib][luni] Enhance the functionalities provided by BitSet
Posted by Spark Shen <sm...@gmail.com>.
Richard Liang 写道:
>
>
> Spark Shen wrote:
>> Hi All:
>>
>> IMHO, EnumSet provides a set view of enum types. According to its
>> spec, "Enum sets are represented internally as bit vectors".
>> To avoid to re-invent the wheel, it could be more reasonable to
>> utilize java.util.BitSet class to support the implementation of EnumSet.
>>
> Agree :-)
>
>> While current API methods provided by BitSet is not enough, I need to
>> extend the function provided by BitSet. There are 2 options:
>> 1. Implement these extension methods in BitSet directly as package
>> private methods
>> 2. Construct a package private utility class in
>> org.apache.harmony.luni.util package, and implement extensions in
>> this implementation class.
>>
>> I prefer the first option, since there is not many classes need these
>> extension. If no objections, I will adopt the first schema.
>>
> IMHO, it depends on the relationship between the extended functions
> and the whole class. Would you please provide more detail information?
> Thanks a lot.
1. It can be simply detected that on RI, EnumSet.noneOf() or
EnumSet.allOf() method will return a concrete EnumSet. This concrete
EnumSet provides its own iterator.
All these iterator related methods are not implemented in BitSet right now.
2. To support the implementation of EnumSet, more utility methods used
to set/unset bits may need to be added.
All these requirements are EnumSet specific, do they need to be
implemented in a separate class.
Best regards
>
> Richard.
>> Best regards.
>>
>
--
Spark Shen
China Software Development Lab, IBM
---------------------------------------------------------------------
Terms of use : http://incubator.apache.org/harmony/mailing.html
To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
For additional commands, e-mail: harmony-dev-help@incubator.apache.org
Re: [classlib][luni] Enhance the functionalities provided by BitSet
Posted by Richard Liang <ri...@gmail.com>.
Spark Shen wrote:
> Hi All:
>
> IMHO, EnumSet provides a set view of enum types. According to its
> spec, "Enum sets are represented internally as bit vectors".
> To avoid to re-invent the wheel, it could be more reasonable to
> utilize java.util.BitSet class to support the implementation of EnumSet.
>
Agree :-)
> While current API methods provided by BitSet is not enough, I need to
> extend the function provided by BitSet. There are 2 options:
> 1. Implement these extension methods in BitSet directly as package
> private methods
> 2. Construct a package private utility class in
> org.apache.harmony.luni.util package, and implement extensions in this
> implementation class.
>
> I prefer the first option, since there is not many classes need these
> extension. If no objections, I will adopt the first schema.
>
IMHO, it depends on the relationship between the extended functions and
the whole class. Would you please provide more detail information?
Thanks a lot.
Richard.
> Best regards.
>
--
Richard Liang
China Software Development Lab, IBM
---------------------------------------------------------------------
Terms of use : http://incubator.apache.org/harmony/mailing.html
To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
For additional commands, e-mail: harmony-dev-help@incubator.apache.org