std.net.curl: Bad timeout defaults

[Johannes Pfau]

Am Fri, 14 Sep 2012 15:54:39 +0200
schrieb "Jonas Drewsen" <jdrewsen at nospam.com>:

> On Friday, 14 September 2012 at 13:31:50 UTC, Johannes Pfau wrote:
> > I've been using std.net.curl lately and there's a small issue 
> > with the
> > default timeouts:
> >
> > dataTimeout is documented as:
> > /// Set timeout for activity on connection.
> >
> > but this is not true! timeout_ms actually is "the maximum time 
> > that you
> > allow the libcurl transfer operation to take". This timeout is 
> > enforced
> > even if there is activity. So we currently limit all CURL 
> > operations to
> > 2 Minutes (default limit), which sucks for download managers 
> > and other
> > long-running operations.
> >
> > We should probably use CURLOPT_LOW_SPEED_LIMIT and
> > CURLOPT_LOW_SPEED_TIME instead. Any comments?
> 
> This is indeed not good and the comments should be corrected. 
> CURLOPT_LOW_SPEED_LIMIT and CURLOPT_LOW_SPEED_TIME seems to be 
> good candidates for an extra option.

I think we should rather keep the comments and make dataTimeout use
CURLOPT_LOW_SPEED_LIMIT=1 and CURLOPT_LOW_SPEED_TIME. Then we could add
an extra operationTimeout option for CURLOPT_TIMEOUT.

Don't you think CURLOPT_TIMEOUT is more an operation timeout than a
dataTimeout? It already includes connect time, so I'd say we should
call it operationTimeout and not dataTimeout.