Re: [boost] Formal Review of IO and Toolbox extensions to Boost.GIL starts TOMORROW

Subject: Re: [boost] Formal Review of IO and Toolbox extensions to Boost.GIL starts TOMORROW
From: Phil Endecott (spam_from_boost_dev_at_[hidden])
Date: 2010-12-11 18:20:57

Domagoj Saric wrote:
> "Phil Endecott" <spam_from_boost_dev_at_[hidden]> wrote in message 
> news:1291925453226_at_dmwebmail.dmwebmail.chezphil.org...
>> Christian Henning wrote:
>>> This would assume that both png reader and writer have some bugs in
>>> common.
>>
>> No it doesn't.  Imagine a bug in the reader that caused the data to be all 
>> 0.
>
> To follow Lubomir in giving Christian more support, the 'clear-cut' "no it 
> doesn't" seems (even to me :) a bit too harsh as Christian did admit that 
> the test is not perfect and, statistically speaking (as he spoke of 'good 
> indication'), he was not so far from truth (i.e. what is the probability of 
> the reader reading all zeros and what of the reader and writer having a 
> common bug...)...

My "imagine..." line was intended to represent the whole class of faults
that would not be detected by this testing strategy.  For example, the last
bug I had in my PNG loader was that it lost the right-most pixel of odd-width
images when loading into a 16-bit destination.  This would not be 
detected by
this read-write-read-compare approach.

Regards,  Phil.