On Fri, 06/07 09:46, Stefan Hajnoczi wrote: > On Thu, Jun 06, 2013 at 12:01:57PM +0100, Richard W.M. Jones wrote: > > On Thu, Jun 06, 2013 at 02:25:46PM +0800, Fam Zheng wrote: > > > v7: > > > 13: Added: > > > curl: change timeout to 30 seconds > > > > I tested this against: > > > > (1) HTTP to Apache server over slow but local wifi. > > > > (2) HTTP to a remote ISO (on another continent). > > > > Test (1) is fine. > > > > Test (2) gives plenty of I/O errors. I guess that 30 seconds isn't > > sufficient here. > > > > I should note that current upstream qemu *works* in both cases. > > Although the timeout in current qemu is much shorter (5 seconds), for > > some reason this does not affect the test. > > > > I'm also confused about what problem this patch series is trying to > > fix, since upstream qemu works fine for me with the latest curl. > > One problem I've had with the upstream code is that HTTP servers which > do not support Range: headers fail in a weird way (I don't remember > exactly what happens). This series does improve this by explicitly > checking for Range: header support. > What I see is it always returns the first sectors to guest.
> But there is a lot of other refactoring going on which is reasonable > only if it works as well or better than the upstream code. > In this case it doesn't :( I can probably send a fix for range and drop the refactoring for now. -- Fam
