I looked in the bug list and did not immediately see anything written
up on this problem.
I encountered the following core dump while running regression tests
on libxml2-2.3.8 with xmlDoValidityCheckingDefaultValue is set to true.
The input test document does not have a <!DOCTYPE root_element_name .... >
statement and looks like this:
<?xml version="1.0"?>
<amp>
<ipsID>xxx</ipsID>
</amp>
Therefore, I would expect libxml to generate a message about this.
But instead, it coredumps.
The problem is very reproducible on our OS which is QNX.
As you can see, as I step through the code
1) it fails to find the <!DOCTYPE .... > statement.
2) It then attempts to issue a error by calling
xmlGenericError( xmlGenericErrorContext,
which calls vfprintf
Several layers deeper in the trio it core dumps.
The stack is trashed, so the only way
I can show you anything is to step through it with the debugger
and collect stack and variable data. Hopefully, I have given
you enough to reproduce the problem. If you can not reproduce
the problem, then I must assume that it has something to do with
the way I build libxml on QNX. But since everything else works
so well, I am inclinded to think that this is a real problem.
Attached is the stack and variable data I collected with the WATCOM
debugger. It is sort of a histogram, with the final stack trace near
the bottom.
I was hoping to actully solve the problem, but I have no desire to
mess in the trio code. My first thought was that there was a mismatch
between the arg list and the %s %d in the message, but that seems
to match.
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
Stanley J. Boehm, Senior Software Engineer
Center for Industrial Technology
Sunnyvale, CA 94085-2810
Phone: (408) 617-2483 E-mail: Stan Boehm philips com
FAX : (408) 617-7711
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
Attachment:
core_dump
Description: Binary data