[Box Backup] ./runtest.pl common debug fails

[Ben Summers]

On 16 Nov 2005, at 12:28, Urs Rau wrote:

> Ben, thanks for this quick reply.
>
> Ben Summers wrote:
>>> common: FAILED: 1 tests failed
>>>
>>> Is this expected? Where should I start looking?
>>
>>
>> This line
>>
>>> FAILURE: Condition [block != b2] failed at testcommon.cpp(181)
>>
>> tells you what happened. In fact, this is harmless, it's a check  
>> to  see
>> that realloc reallocated a block, which on your platform it   
>> didn't. The
>> test hopes the block was reallocated so it can check the  debug leak
>> checking is working.
>>
>> You omitted to tell us anything about your platform, which is not
>> particularly helpful when asking about bugs in cross-platform   
>> software.
>
> Sorry about that. This was run on a linux fedora core 2 box runing  
> under
> kernel  2.6.6-1.435, perl 5.8.3-18, zlib-1.2.1.1-2.1,  
> openssl-0.9.7f-3,
> g++ (GCC) 3.3.3 20040412 (Red Hat Linux 3.3.3-7).
>
>> However, from your test results it looks like you're good  to go on
>> whatever OS you're using.
>
> Thanks Ben. I guess I was just not expecting any test to fail on a
> redhat fc2 with relatively recent kernel. On what system would this  
> test
> not fail? Or in other words what is it, that my FC2 doesn't handle as
> expected?

Differences in clib. The test expects that when a tiny allocated  
block is resized to a very big block, its address will change as it  
is moved from one pool to another. You clib doesn't appear to work  
this way. The test is there to check that the memory leak detection  
will keep on tracking the block even if it's address is changed.

As I said, nothing to worry about. Just an incorrect assumption about  
malloc algorithms, which isn't even important to a running Box Backup  
system.

Ben