[RFC] shared busybox as library drop-in replacement

[Rich Felker]

On Sat, Apr 08, 2006 at 03:17:41PM -0500, Rob Landley wrote:
> Translation of my question:
> 
> If someone wants to use a shared library and the shared library is not PIC 
> code, then I'm under the impression each user of the shared library gets the 
> fixups done independently and thus all executable pages they actually use are 
> dirtied for each application, which defeats the purpose of a shared library.  
> Thus for shared libraries to make sense, they're supposed to be PIC code.  Is 
> this correct?

To make sense from a memory-saving standpoint, shared libraries should
be PIC. To make sense from a disk-saving standpoint where memory is
huge and disk is small, non-PIC might make sense (and give slightly
better performance after initial load penalties for relocations).

> likely to be running at once?  How would one measure this?

Using statm..?

> > Yes, look at /proc/$pid/statm. The last number is all that counts:
> > number of dirty pages.
> 
> I know that.  And in theory, all application profiling could be done using the 
> "time" command.  This is not a test harness that gives me useful comparison 
> between different approaches with any sort of granularity.

I'm confused about what additional granularity you want.. Sorry.

Rich