On Wed, Feb 05, 2014 at 10:13:35PM +0000, Steve Holme wrote:
> However, after I pushed three commits [3], [4], and [5] last night, I noticed
> that Test 165 broke - as can be seen in the auto builds from the past 22 hours
> [6].
>
> Unfortunately, as I only compile for Windows, don't run Valgrind, nor have
> access to libIDN I am struggling to see where the problem is and why I broke
> this test with my commits.
>
> From the Valgind Memcheck log, and my limited knowledge of them, it looks like
> the buffer allocated in idna_to_ascii() in libIDN which is used to transform
> the non-ascii based URL is not long enough - by one byte - so generates a read
> error but if this was the case why has my rework suddenly broken things /
> flagged this?
>
> As such I am a little lost and would appreciate a little help and assistance
> with tracking this breakage down.

It probably has nothing to do with your changes. I'm able to reproduce it on my
machine with libidn 1.26, but I don't see how it could be a curl problem. The
libidn call is very simple, and when I run it under gdb, the arguments appear
fine. I suspect your commits just shifted memory usage around enough that
some latent problem in libidn is coming to light. Or, it might just be a false
positive. Debugging further would take digging into libidn.

>>> Dan
-------------------------------------------------------------------
List admin: http://cool.haxx.se/list/listinfo/curl-library
Etiquette: http://curl.haxx.se/mail/etiquette.html
Received on 2014-02-05