> On Jan 29, 2015, at 4:20 AM, Victor Stinner <[email protected]> wrote:
> 
> Should I understand that you are against the addition of a new
> connection_failed() method?

As we've been discussing this, the only unifying factor I can see is that there 
are a lot of implementation accidents that lead to vaguely bad stuff happening 
that an application might want to ... log, I guess?  The problem I'm having 
with such a method is that it doesn't seem to indicate anything that can be 
unified into a single coherent failure mode.  For example, if you have 
connection_lost, this means very different things for a UNIX socket, the end of 
a file that is being read, a failed TLS MAC, or a TCP connection getting a FIN 
or RST, but abstractly all of them indicate "the connection ended".  
connection_failed, by contrast, just seems to mean "some stuff happened to some 
sockets, maybe you'd like to know", so I'm having a lot of trouble imagining 
what kind of code a correctly-written application would put into this method 
other than logging; and even for logging, it's not clear it would know what to 
log without intimate knowledge of the implementation of every transport.

So how do you imagine this would be used?

-glyph

Reply via email to