[Sbcl-devel] Re: much ado about zero patches

[Luke G. <lu...@bl...>]

"Perry E. Metzger" <pe...@pi...> writes:

> 2) Purge the code that does the unmap/map "trick" for zeroing large
>    blocks. I've checked several OSes and talked to a bunch of OS gurus
>    and everyone agrees it is at best slightly worse than inlined
>    memset and at worst a serious bad idea.

Why?

One nice thing is that it lazily zeros the memory as you use it,
outside the GC. That makes GC pauses shorter. On Linux 2.6 I couldn't
measure any performance problems, but perhaps it's atypical.

If I were still crusading to minimize GC pauses I would be screaming
and yelling, but fortunately I'm not :-)

-Luke