harmony-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Pavel Pervov <pmcfi...@gmail.com>
Subject Re: Question about the boot and classloading
Date Tue, 19 May 2009 13:27:42 GMT
Please, read this nice presentation [1] on osgi vs classic class loading.
[1] http://www.martinliqert.com/events/WJAX2008-ClassloadingTypeVisibilityOSGi.pdf

2009/5/19, tian galaxy <harmonymail.ty@gmail.com>:
> XiaoFeng , Mark
> Thanks for your help ,very valuable to me.
> I think I need to research more on this .Much more complex than I thought
> previously.
>
>
> Regard
> Tian
>
> 2009/5/19 Xiao-Feng Li <xiaofeng.li@gmail.com>
>
>> On Tue, May 19, 2009 at 6:41 PM, Mark Hindess
>> <mark.hindess@googlemail.com> wrote:
>> >
>> > In message
>> > <dff214c70905190147x57135ef3hf26cbf6a93ae2255@mail.gmail.com>,
>> tian
>> > galaxy writes:
>> >>
>> >> Hindess .Thanks for the help ! I realized that this far more complex
>> than I
>> >> thought before .
>> >> So I think I need a deeper look into both osgi and harmony .
>> >> Here are several things I don't qiute understand and I noted them
>> below.May
>> >> be it's because I am not quite familiar with the low level ,
>> classloading
>> >> details .
>> >> thanks!
>> >>
>> >> 2009/5/19 Mark Hindess <mark.hindess@googlemail.com>
>> >> >
>> >> > In message <
>> dff214c70905180610w407e3bd4h5baa319729c1fc55@mail.gmail.com>,
>> >> > tian galaxy writes:
>> >> > >
>> >> > > OK ,thanks .Seems the minimal class set has been defined by the
>> osgi.
>> >> >
>> >> > Not sure how this helps though.  The OSGi Minimum Execution
>> Environment
>> >> > says what is theoretically the minimum required to run OSGi.  What
we
>> >> > need to concern ourselves with is the *practical* minimum and that
>> >> > really depends on our implementation details.
>> >> >
>> >> > I don't think that we should implement OSGi in the VM.  Instead the
>> >> > existing bootstrap classloader should load the OSGi framework (with
>> >> > a unique classloader)
>> >>
>> >> If we load the osgi with the bootstrap class loader , why comes a
>> >> unique
>> >> class loader?
>> >
>> > It isn't necessary but I think it is useful to use a separate
>> > classloader for the OSGi framework so that we can isolate the classes
>> > use by the VM's OSGi framework and avoid polluting the namespace.
>> >
>> >> Do you mean the bootstrap class loader should delegate the loading to
>> >> the 'unique class loader '?
>> >
>> > Yes.
>> >
>> >> > and the system classes (with a "temporary" classloader).
>> >>
>> >> Again ,the relation between temporary and bootstrap classloder?  And
>> >> why temporary ?
>> >
>> > Temporary because until the OSGi framework is up you wont know which
>> > bundle the classes belong to and thus wont know which bundle classloader
>> > the classes should belong to.  Alternatively, I suppose you could
>> > implement minimal support for reading OSGi headers from manifests
>> > in the VM so that you could assign classes to what would become a bundle
>> > classloader when the framework is up.
>> >
>> > -Mark.
>>
>>
>> This makes sense to me. The VM simply grabs in all the stuff "as
>> usual" including the OSGi framework, then sort them out according to
>> the bundles' definitions.
>>
>> It could be VM-neutral, because it is the VM's own business to take
>> care of its loaded classes. A callback interface OSGI_ready() can
>> inform the VM to tidy up the classes loaded for OSGi (or before it is
>> ready).
>>
>> Thanks,
>> xiaofeng
>>
>>
>> >> >  Once the OSGi framework is configured the system classes need to be
>> >> > moved to the correct classloader for the OSGi bundle to which they
>> >> > belong.
>> >> >
>> >> > This means that during the initial starting-OSGi-framework phase
>> >> > the OSGi bundle import/exports will not be enforced but, other than
>> >> > implementing OSGi in the VM, I don't see how to avoid this.
>> >> >
>> >> > The multiplicity of bootstrap classloaders still needs to be hidden
>> >> > since the API demands that getClassLoader should return null for the
>> >> > bootstrap classloader(s).
>> >> >
>> >> > The biggest problem is probably how to implement this as
>> VM-independent
>> >> > API such that a VM wanting to implement OSGi had to provide only a
>> >> > minimal C/Java interface in order to gain OSGi support.
>> >> >
>> >> > Regards,
>> >> >  Mark.
>> >> >
>> >> > > I will look more closely on this problems and let you know if
there
>> some
>> >> > > progresses made.
>> >> > > Thanks  a  lot !
>> >> >
>> >> > > 2009/5/18 Pavel Pervov <pmcfirst@gmail.com>
>> >> > >
>> >> > > > I do not know osgi model in detail, but what minimal set
of
>> classes is
>> >> > > > needed to start osgi model itself? Then we could delegate
to
>> "system"
>> >> > > > bundle loader from bootstrap classloader.
>> >> > > >
>> >> > > > 2009/5/18, tian galaxy <harmonymail.ty@gmail.com>:
>> >> > > > > Yes , implement osgi loading model in there is too complex.
>> >> > > > >
>> >> > > > > We know some modules are more important than others
,
>> >> > > > > performing
>> some
>> >> > > > > fundamental , basic functions .
>> >> > > > > Is it possible to only load these more" basic " modules
with
>> >> > bootstrap
>> >> > > > class
>> >> > > > > loader and make the other modules work the osgi bundle?
>> >> > > > > Actually I don't quite understand why system classes
could only
>> be
>> >> > loaded
>> >> > > > by
>> >> > > > > bootstrap class loader?I noticed that there are some
safety
>> concerns
>> >> > > > > ,is there any reasons other than that?
>> >> > > > > Thanks !
>> >> > > > >
>> >> > > > >
>> >> > > > > 2009/5/18 Pavel Pervov <pmcfirst@gmail.com>
>> >> > > > >
>> >> > > > >> Short answer is "no". You'll have to implement osgi
loading
>> model
>> >> > > > >> inside bootstrap classloader. Is it too complex?
>> >> > > > >>
>> >> > > > >> 2009/5/18, tian galaxy <harmonymail.ty@gmail.com>:
>> >> > > > >> > Hi Pavel.
>> >> > > > >> >
>> >> > > > >> > Thanks.
>> >> > > > >> > Yes ,that's right .My question is whether there
is a way to
>> load
>> >> > the
>> >> > > > >> system
>> >> > > > >> > class by class loader other than the  bootstrap
class loader
>> ?
>> >> > > > >> > If we could load the system class by different
loaders other
>> than
>> >> > the
>> >> > > > >> > bootstrap class loader then there is a chance
to make the
>> harmony
>> >> > > > >> > runtime
>> >> > > > >> to
>> >> > > > >> > be osgi-ed.
>> >> > > > >> > I trying to work on this problem . Do you have
any
>> suggestions ?
>> >> > > > >> >
>> >> > > > >> >
>> >> > > > >> > 2009/5/18 Pavel Pervov <pmcfirst@gmail.com>
>> >> > > > >> >
>> >> > > > >> >> Tian,
>> >> > > > >> >> System classes are all loaded with bootstrap
classloader.
>> >> > > > >> >> It
>> is
>> >> > > > >> >> written in C++ and is located in
>> >> > > > >> >> working_vm/vm/vmcore/src/class_support/classloader.cpp
>> >> > > > >> >>
>> >> > > > >> >> 2009/5/17, tian galaxy <harmonymail.ty@gmail.com>:
>> >> > > > >> >> > Hi all
>> >> > > > >> >> > I have a question about the classloading
in the drlvm.
>> >> > > > >> >> >
>> >> > > > >> >> > I know that there are kernel library
inside the drlvm
>> >> > > > >> >> > such
>> as
>> >> > > > >> >> > java.lang.classloader, which enclose
the the
>> systemclassloader.
>> >> > And
>> >> > > > >> this
>> >> > > > >> >> > class is loaded and used in the process
of creating VM .
>> >> > > > >> >> > While the systemclassloader is a subtype
of
>> >> > > > >> >> > urlclassloader
>> >> > which is
>> >> > > > >> >> > in
>> >> > > > >> >> the
>> >> > > > >> >> > "luni" module.So does the JNI_CreateJavaVM
>> >> > > > >> >> > also load the urlclassloader in the
boot procedure of vm
>> ?(I
>> >> > failed
>> >> > > > >> >> > to
>> >> > > > >> >> > search related code in the source
code of drlvm).
>> >> > > > >> >> > If so , when ?Before or after the
jni available ?
>> >> > > > >> >> > Since I am trying to implement the
osgi on harmony and
>> >> > > > >> >> > the
>> osgi
>> >> > > > works
>> >> > > > >> >> with
>> >> > > > >> >> > bundles (modules) , I think I need
to know the procedure
>> of
>> >> > booting
>> >> > > > >> and
>> >> > > > >> >> > loading of class from different modules
at the start up
>> time.
>> >> > > > >> >> > Any clue about this ?Thanks
>> >> > > > >> >> >
>> >> > > > >> >> > plus,do we have the source code of
java.exe ?Where?
>> >
>> >
>> >
>>
>>
>>
>> --
>> http://people.apache.org/~xli
>>
>
>
>
> --
> Regards
> Michael Tian
>

Mime
View raw message