Reducing overall amount of OpenJDK source code

[Erik Trimble]

On 5/24/2010 11:17 AM, David Herron wrote:
> On Mon, May 24, 2010 at 10:27 AM, Андрей Мишанин <amishanin at swemel.ru 
> <mailto:amishanin at swemel.ru>> wrote:
>
>     Hi everyone. My name is Andrey, I'm from Russia. Our company is
>     building it's own operating system based on Solaris 10 with
>     Trusted Extensions (it is named Zircon) and we've chosen OpenJDK 7
>     as a main platform for our application development. Our current
>     goal is to create a specialized version of JDK targeting solely
>     Zircon with everything we don't need thrown out (you may think of
>     it as an embedded version of JDK, however it isn't). The main task
>     is to reduce overall amount of the source code (not the size of
>     the binary image) as later it will be analyzed and certified for
>     use in government bodies. Long story short: the more source code
>     we have to submit for analysis, the more it will cost.
>
>     So far I have managed to successfully build the latest source code
>     snapshot (dated May 13) on Zircon and verified that all of our
>     Java applications run on OpenJDK without any problems. So now I
>     need to figure out how I can remove those chunks of source code we
>     don't need. The usual suspects are:
>            - all native source code targeting Windows and Linux (yes,
>     I know, Linux and Solaris native sources have much in common but
>     still)
>            - JRE classes and namespaces we do not use at all (i.e.
>     java.sql or jaxp)
>            - some of JRE classes we do not need (for example we have
>     settled on Nimbus as a default look and feel for all our GUI
>     applications, which means we do not longer need Motif or Metal L&Fs)
>
>     So I'd be grateful if you could describe an overall strategy of
>     removing OpenJDK parts I've mentioned in the previous paragraph
>     with as little hassle as possible. Well, may be not a complete
>     strategy, but some hints and best practices will be greatly
>     appreciated :-)
>
>     P. S. I expect lots of tinkering with Makefiles and another files
>     in jdk/make to be involved, so it'd be great if you could give me
>     a link to docs or description of those.
>
>
>
> An obvious consideration is whether you want the result to be 
> certifiable as Java Compatible with the trademarks .. you'll have to 
> pass the TCK and other requirements .. those tests include whether you 
> have a complete set of API's.  This includes java.sql and jaxp ... 
> hence the direction you just laid out won't be certifiable.
>
> But .. pragmatism says that the existence and success of Android shows 
> that the world is willing to adopt a near-Java platform that doesn't 
> meet the compatibility requirements.
>
> - David Herron
> http://davidherron.com
>
>
> PS I'm surprised you didn't mention CORBA as one of the things to ditch...
>

As a side note on David's point:   I'm pretty sure that having a working 
SUBSET of the full JDK is much more acceptable than having either a 
SUPERSET or disjoint Set.  That is, if you can detail ahead of time what 
sections of the "Zircon JDK" won't work for your userbase, they'll be 
much more forgiving than if you do something like replace sections of 
the JDK with other (non-compatible) implementations/designs.

As far as the VM goes (my area), it is possible to remove a GC method 
(CMS, G1, etc.), but it's not easy.  Removing anything else from the VM 
is going to be a huge amount of work, and I'd strongly advise you not to 
try.  More importantly, in the VM, removing things without a /very/ deep 
understanding of what you are doing is very likely to severely impact 
performance.  I realize you had focused on portions of the Library 
section of the JDK, but be aware that the JDK is not just the Libraries.



-- 
Erik Trimble
Java System Support
Mailstop:  usca22-123
Phone:  x17195
Santa Clara, CA

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.org/pipermail/build-dev/attachments/20100524/51456748/attachment.htm>