[PATCH 0 of 1 ] Wrong distance calculation in revlog causes huge manifests

[Noel Grandin]

On 2012-07-12 19:39, Matt Mackall wrote:
> Actually, I was (too subtly, as usual) referring to the fact that 
> every operating system has a different AIO interface, if they support 
> it at all, not suggesting that a wrapper library was an acceptable 
> solution. Most operating systems' AIO implementations are an 
> afterthought and not worth the trouble of using.

I respectfully disagree.
MS-Windows has a pretty decent implementation, since it was designed by 
the same people who built VMS, which used async calls pervasively.
The linux implementation (specifically the glibc posix AIO interface, 
not the kernel O_DIRECT API) is not ideal, but it's still pretty usable 
for modest amounts of async IO.
I have no idea what the MacOS implementation is like.

Disclaimer: http://www.peralex.com/disclaimer.html