Amos Jeffries, le mer. 14 janv. 2026 11:47:31 +1300, a ecrit:
> On 14/01/2026 11:23, Samuel Thibault wrote:
> > Amos Jeffries, le mer. 14 janv. 2026 10:50:45 +1300, a ecrit:
> > > Per my earlier reply, I do advise keeping the HEAD initial request.
> > 
> > I don't see the use of it. You have no guarantee that the GET will get
> > the same server response header.
> 
> It is a bandwidth efficient way to detect and handle 3xx/4xx/5xx status.

?

This is not what httpfs uses it for. It uses it for just for
netfs_attempt_read to be able to ignore the header returned by GET.

> The way the translator is currently implemented with different TCP
> connections for HEAD and GET is almost guaranteed to hit different servers
> with potentially different response headers.

And thus you don't want to make to requests, just one.

> Moving to keep-alive on the same TCP connection ensures the same server
> supplies both HEAD and GET. Thus, flipping the probability to almost
> certainly having the same headers on both. Not quite guarantee, but much
> better than the current code.

My point is that we should just make the code *not* have to rely on such
guarantee. You can just GET, and in netfs_attempt_read drop the server
response header.

Samuel

Reply via email to