Re: [cp-patches] PATCH: logfile rotation, native methods

[Mark Wielaard]

Hi Casey,

On Thu, 2005-07-07 at 23:17 -0700, Casey Marshall wrote:
> One feature this implementation requires, in order to function  
> properly, are file locks, which to my dismay are still not  
> implemented. On that point, I'd like to ask what the requirements on  
> JNI code are for Classpath, if any. I know there are at least two  
> "standard" APIs in Classpath for native functionality -- the  
> TARGET_NATIVE macro nonsense, and the VM* classes (I get the feeling  
> that the latter is preferred now). Should file locks (and, say,  
> memory-mapped files) use one of these? Or would it be acceptable to  
> simply write the JNI code directly, surrounded by appropriate #ifdef's?

The VM* classes and the TARGET_NATIVE macros are on two different
"layers" of portability. The VM* classes make it possible to replace our
java/POSIX/C/JNI default implementation from vm/reference with a version
that uses CNI, C# or something completely different if the runtime
writer wants that. The TARGET_NATIVE macros provide (an admittedly
awkward) way to write "portable" C code. I think you should feel free to
ignore them for new code since they make the code unreadable and
debugging a nightmare. And they don't provide us any extra portability
in practice since there is only one implementation. Please write clean
(were possible autoconf-ed) GNU/Posix JNI C implementations for any VM*
class in vm/reference. More "requirements" (goals) are here:
http://www.gnu.org/software/classpath/docs/hacking.html#SEC10

Cheers,

Mark

signature.asc
Description: This is a digitally signed message part