harmony-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Oliver Deakin <oliver.dea...@googlemail.com>
Subject Re: [drlvm] Thread library function table added
Date Tue, 18 Aug 2009 13:57:29 GMT
Mark Hindess wrote:
> I think we should do this.  However, this will "break" developers using
> the IBM VME.  Perhaps we could do:
>
> Index: make/properties.xml
> ===================================================================
> --- make/properties.xml (revision 805151)
> +++ make/properties.xml (working copy)
> @@ -286,7 +286,10 @@
>      <property name="exclude.module" value="nothing" />
>  
>      <!-- flags -->
> -    <property name="hy.no.thr" value="false" />
> +    <condition property="hy.no.thr" value="false">
> +        <available file="${hy.hdk}/jdk/jre/bin/default/libj9vm23.so" />
> +    </condition>
> +    <property name="hy.no.thr" value="true" />
>      <condition property="hy.skip.thr" value="true">
>          <not>
>              <equals arg1="${hy.no.thr}" arg2="false" />
>
> to mitigate this slightly?
>   

It's not very pretty, but it would at least automate this step for the 
current VME until it is updated. It would still be an issue if users 
take a first time checkout of the class libraries, build them and then 
overlay the IBM VME, as the j9vm23 library would not be present at build 
time, but at least it would save us specifying the hy.no.thr flag for 
subsequent builds. It also needs to look for j9vm23.dll on Windows, 
having an <or> condition for either library should do the trick.

Regards,
Oliver

> Regards,
>  Mark.
>
> In message <4A8A7B91.6090206@googlemail.com>, Oliver Deakin writes:
>   
>> Carrying on from this discussion, I'd like to make hy.no.thr=true the 
>> default setting and stop building the classlib thread library in the 
>> federated build.
>>
>> Please let me know if there are any objections/comments on this, 
>> otherwise I will make the change after M11 is published.
>>
>> Regards,
>> Oliver
>>
>> Oliver Deakin wrote:
>>     
>>> Mark Hindess wrote:
>>>       
>>>> In message 
>>>> <94d710af0908112202x6eb5498dy9bd1dd84dbdde0b5@mail.gmail.com>,
>>>> Sean Qiu writes:
>>>>  
>>>>         
>>>>> +1 for both of your proposal, it sounds reasonable and cool.
>>>>> Thanks, Oli.
>>>>>     
>>>>>           
>>>> I am in favour of removing the option completely and removing the
>>>> classlib thread code.  Of course, this breaks the IBM VME so perhaps we
>>>> can leave it in place - but change the default? - until a new IBM VME is
>>>> available?
>>>>   
>>>>         
>>> Yes, I think we should switch the default to hy.no.thr=true for a 
>>> transition period. There is no harm in keeping the classlib thread 
>>> code present for a while, but I think the eventual goal should be to 
>>> delete it from the repo.
>>>
>>>       
>>>>  
>>>>         
>>>>> One question here, what do you mean remove the code? svn del?
>>>>> What about we "svn move" it to some other place, such as
>>>>>
>>>>> https://svn.apache.org/repos/asf/harmony/enhanced/legacy
>>>>>     
>>>>>           
>>>> Well, we already have:
>>>>
>>>>   https://svn.apache.org/repos/asf/harmony/enhanced/classlib/archive
>>>>
>>>> but I think "svn delete" is fine since you can always checkout earlier
>>>> revisions if you need to revisit the code.
>>>>   
>>>>         
>>> Agreed.
>>>
>>> Regards,
>>> Oliver
>>>
>>>       
>>>> Regards,
>>>>  Mark.
>>>>
>>>>  
>>>>         
>>>>> 2009/8/11 Oliver Deakin <oliver.deakin@googlemail.com>:
>>>>>    
>>>>>           
>>>>>> Hi all,
>>>>>>
>>>>>> I have added an implementation of the thread library function table

>>>>>> for
>>>>>> DRLVM which can be enabled by building with the "-Dhy.no.thr=true"

>>>>>> flag
>>>>>> specified on the Ant command line. This means we no longer need to

>>>>>> build th
>>>>>>       
>>>>>>             
>>>>> e
>>>>>    
>>>>>           
>>>>>> classlib thread library in the federated build, and we also no 
>>>>>> longer need
>>>>>> to copy the DRLVM hythr library into jre/bin (although I havn't 
>>>>>> changed the
>>>>>> build to not do the copy yet). I have temporarily set the hythr 
>>>>>> library
>>>>>> version to 0.2 so that the federated build can be run with and 
>>>>>> without the
>>>>>> hy.no.thr flag set.
>>>>>>
>>>>>> This opens a couple of questions for discussion:
>>>>>> 1) Shall we set the hy.no.thr option to true as default? I 
>>>>>> personally think
>>>>>> we should - the classlib hythr library is not used in the federated

>>>>>> builds,
>>>>>> so there is no reason to continue building it.
>>>>>> 2) Shall we remove the thread library from classlib altogether? If
>>>>>> hy.no.thr=true becomes the default, I can see reasons for and 
>>>>>> against [1]
>>>>>> removing the source from classlib, but I think the reasons to 
>>>>>> remove the
>>>>>> code outweigh the reasons to keep it. My vote is to remove that source
>>>>>> module from classlib altogether.
>>>>>>
>>>>>> Ideas/comments?
>>>>>>
>>>>>> Regards,
>>>>>> Oliver
>>>>>>
>>>>>> [1] A few I can think of straight off:
>>>>>> For:
>>>>>> - Unused thread library code in classlib is unlikely to be maintained.
>>>>>> - Some classlib thread code is incorrect (x86_64 linux has some 
>>>>>> invalid
>>>>>> assembler code I believe).
>>>>>> - Shrinks the source tree footprint.
>>>>>> - All VMs will likely have their own implementation of this 
>>>>>> functionality
>>>>>> anyway.
>>>>>> Against:
>>>>>> - Raises the bar slightly for VM vendors wishing to use the Harmony

>>>>>> class
>>>>>> libraries.
>>>>>>       
>>>>>>             
>>>>
>>>>   
>>>>         
>> -- 
>> Oliver Deakin
>> Unless stated otherwise above:
>> IBM United Kingdom Limited - Registered in England and Wales with number 7415
>> 98. 
>> Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU
>>
>>     
>
>
>
>   

-- 
Oliver Deakin
Unless stated otherwise above:
IBM United Kingdom Limited - Registered in England and Wales with number 741598. 
Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU


Mime
View raw message