On Sat, 21 Apr 2007, Yang Tse wrote:

> Now the funny thing is that test #56 fails again on both IRIX 6.2 MIPS C 6.2
> autobuilds.
>
> Perhaps the more or less frequent flushing of the output stream has
> something to do with it? Lurking bug actually in the library or in the
> testsuite server?
>
> Just wanted to share this out loud in case someone had any clue.

I'll add my thoughts, but no solution:

In this failure
http://curl.haxx.se/auto/log.cgi?id=20070421013235-28086#prob1, we can see how
the 'check-generated' file is corrupt after the test has run.

That file should not have that long 'aaaa' string anywhere, since it is passed
to the server in the User-Agent field and the comparison of the protocol parts
is supposed to be done with User-Agent stripped off (as specified within the
<strip> tag in the tests/data/test56 file).

The 'server.input' file shows that extra set of 'aaaaa' so it _looks_ as if
either curl actually sends a badly broken request, or the test suite's http
server logs the request badly. The server is meant to accept requests up to
150000 bytes so it can hold this.

But why does it only happen on this operating system/setup? Could it be a bug
somewhere in the socket receiving/sending that this triggers?

-- 
  Commercial curl and libcurl Technical Support: http://haxx.se/curl.html