initialization problem

[Godmar Back]

> 
> Perhaps it could be considered a bug; however, the broader issue is whether
> one might expect this behavior with other mounted file systems, such AFS or 
> WebNFS.

 If it wants to appear as a file, it better behave like a file.
Files should select for read immediately, IMO, even if the read cannot
be completed immediately.

Clearly, I might be fundamentally or philosophically wrong on this ---
maybe we should look at what CAE has to say about it.

> 
> 	> As a work-around, simply don't use select() --- anything else that will
> 	> tell you that you should do a read() should do just fine, should it not?
> 	> Such as statting it, check for IFREG or do the various ioctl's etc. pp.
> 
> We've done this now, and it appears to work.  Thanks for the suggestion.  
> The initialization of the JNI_CreateJavaVM is slow but this appears to be 
> due to all of the small READ{32,16} found in jar.c.
> 

 We know.
Somebody needs to sit down and implement a mini-stdio (don't use the
normal stdio as previously discussed on that list) that reads 8K or 
something buffers at a time and then copies data from there.  This should 
be a short exercise.  Patches are welcome!

	- Godmar