Make sure to timeouts do not look like normal EOF.

[adamreichold]

Closes #49

[sbstp]

Isn't this what we decided to abandon in your last merge request?

[adamreichold]

Isn't this what we decided to abandon in your last merge request?

Indeed. But now we got a bug report about this being unexpected behavior. So I bit my teeth and tried to implement something reasonable.

The alternative seems to be to tell the user to check his/her content length which does not sound very convincing. (Or to implement those checks ourselves everywhere but even this would leave a discrepancy between the io::Read-based interface and high-level ones like .text().)

[sbstp]

I care more about giving a under friendly library than a fast library, so I'm going to merge this. I don't think this will affect performance much anyway, we're doing IO after all.

[sbstp]

I'll release a patch with this fix.

[adamreichold]

I care more about giving a under friendly library than a fast library, so I'm going to merge this. I don't think this will affect performance much anyway, we're doing IO after all.

I agree. Also I would like to add that my personal reservations against this code where never about performance, it is complexity I am concerned about, i.e. mostly readability and learnability of the code base.