Hi Daniel, et all

While playing a bit with test case #1070 I've arrived at a point in
which there is a call to read_rewind() which seems to be
inappropriate, or at least a week read_rewind() implementation.

In order to reproduce the issue, edit data/test1070 and duplicate in
place line #18 (the one that reads "you are not supposed to be allowed
[...]") one or two hundred times. With this change the test will fail,
of course, but it doesn't matter.

Now run testcurl.pl -g 1070, set a breakpoint on transfer.c:648 and
let it run run to the breakpoint which would call read_rewind(conn,
excess).

If you step into read_rewind() you'll see that conn->read_pos is zero,
and that it is 'decreased' by an amount, resulting in a crazy
conn->read_pos value.

My question is... Do we simply prevent read_rewind() from decreasing
conn->read_pos below zero, or is there some check you think would be
necessary to add near transfer.c:635 to directly avoid the call to
read_rewind(). or whatever else.

Cheers,

-- 
-=[Yang]=-
-------------------------------------------------------------------
List admin: http://cool.haxx.se/list/listinfo/curl-library
Etiquette:  http://curl.haxx.se/mail/etiquette.html