Project Proposal: BSD port

[Kelly O'Hair]

Good points.

A while back I changed code that was ifdef'd on the platform changed
to being ifdef'd on _LITTLE_ENDIAN, just made more sense and the code was
easier to understand.
Changing the ifdef's to specific features rather than making it a platform
thing makes sense. So I agree with your X11 comment.
And also agree that we don't want to deliver classes to a platform that
doesn't need them.

-kto

Mike Swingler wrote:
> On Aug 5, 2008, at 4:01 PM, Dmitri Trembovetski wrote:
> 
>> Rob Ross wrote:
>>
>>> I can't comment on whether it's better to have all *nix statement 
>>> variants be wrapped in #ifdefs or placed in their own 
>>> platform-specific folder. It's probably a judgment call on whether 
>>> platform-specific variations are different enough to warrant their 
>>> own directory structure.
>>
>>  I agree, but they don't need to have different roots as they are now.
>>
>>  Currently we have weird stuff similar to this:
>>    jdk/src/windows/native/sun/awt/windows/awt_Win32GraphicsDevice.cpp
>>    jdk/src/windows/classes/sun/awt/windows/Win32GraphicsDevice.java
>>
>>  It seems that it could as easily be
>>    src/native/sun/awt/windows/awt_Win32GraphicsDevice.cpp
>>    src/classes/sun/awt/windows/Win32GraphicsDevice.java
>>    ...
>>  (I'd even remove 'shared')
>>
>>  The native code for the most part mimics the java packages
>>  names, so if a package is platform-specific, it should probably
>>  have the platform in its name (sun.awt.windows in this case, or
>>  sun.awt.nix or whatever). The native code's structure would
>>  naturally follow the java package naming.
>>
>>  For cases where the package name is the same on different
>>  platform but the native code is different we could follow
>>  Kelly's approach and/or use ifdef-ing.
>>
>>  I do realize that such changes would be a huge pain though.
>>  Especially for those of us who still need to support
>>  earlier releases and port fixes from one tree to another.
> 
> I think the ideal thing to do would be to clearly discriminate in the 
> directory structure exactly what difference is between the platforms. If 
> the difference is X11, then create an x11 directory (because not all 
> *nix's are going to be based on X11). If the difference is Win32 vs. 
> Posix threading and file system operations, there should be a win32 and 
> posix directory for that sub-system. The same should be true for fonts, 
> OpenGL work, or any other system where discrimination does not cleanly 
> cleave down platform lines. #ifdef'ing should only be reserved for 
> situations where the code is largely similar, but requires just a tweak 
> one way or the other (like 32 vs. 64-bit).
> 
> Glomming all the classes together in the same pot and removing the 
> 'shared' directory will encourage everyone to just "compile all the 
> classes", and toss them into the same jar, and ship classes that make no 
> sense on for that platform. Any accessory or utility classes that don't 
> result in a runtime link error only increase the security surface area 
> of attack.
> 
> Thoughts?
> Mike Swingler
> Java Runtime Engineer
> Apple Inc.