On Wed, 4 Aug 2010, m0n0 wrote:

>> Does FreeMiNT have a track record of a fine socket/network layer
>> otherwise?
>
> I'm not absolutly sure what you mean, but if you wan't to know if other
> network apps work correctly, yes they do. Including curl command line client

... and curl is using nothing but libcurl to do its job, so clearly libcurl
works (sometimes).

>> It does sound odd, and I'm sorry to say that I mostly suspect the
>> OS/network layer to not be fully socket compatible to the extent that
>> libcurl uses but that's just an early guess.
>
> As I saw the TCP log I was thinking the same for the first moment, but
> then why:
>
> - does the error just trigger by an chance of 70%, not always?

Well, many bugs are hard to explain before the reason is found. I would expect
due to timing or by pure chance.

> - does not trigger with my hardware, but with 2-3 different hardware setups.

And you have otherwise identical drivers used, OS revisions, libcurl version
and your app version?

Anyway, even if they are, I don't see how that necessary says that the bug is
in libcurl and not in any of the other parts! ;-) Although I of course don't
rule out any libcurl bugs. We fix more than one every week so I'm painfully
aware of their existance!

> - What is libcurl doing that could not be supported that good by the OS?

I have no idea. But libcurl works rather fine on quite a lot of platforms so
when something simple (assuming it is) breaks on a rather esoteric platform my
gut reaction is to suspect the area I'm not sure is that proven. But as I
said, it was only a guess and should not be taken very seriously until we have
a lot more data and knowledge.

> So my hope was that someone could tell me what this expire cleared message
> means...

It means Curl_expire() was called with a zero argument to cancel the current
timeout. It is perfectly normal and expected and happens in lots of
situations. I don't think that's a very useful point to start digging on this.

I think the main focus now has to be to get someone who's able to do some
actual debugging (or printf sprinkling at least) to be able to repeat the
problem.

-- 
  / daniel.haxx.se
-------------------------------------------------------------------
List admin: http://cool.haxx.se/list/listinfo/curl-library
Etiquette:  http://curl.haxx.se/mail/etiquette.html