[Insight-developers] Re: Limitation in MetaIO causing problems initkNeuralNetworkFileReader/Writer

[kent williams]

I certainly am not proposing to break existing, released code.

Looking at the MetIO classes more closely this morning, it appears that it
already supports binary data output, which doesn't have the limitations that
The ASCII record type has with respect to data size.

Like the vnl_matlab classes, it needs to take into account machine
endianness, which right now it ignores.

So in contrast to yesterday evening, I'm no longer running around with my
hair on fire ;-)


On 8/21/07 6:12 AM, "Atwood, Robert C" <r.atwood at imperial.ac.uk> wrote:

>  
> I for one would not be too happy with any such breakage, The intent as I
> understand (and agree with) is that the meta-header is human readable,
> and sticking tons of data into the tags would break this design idea,
> let alone code using it.
> 
> I use it all the time and can easily write separate (.mhd) headers by
> hand, or by little scripts, for 3rd party and legacy image formats/image
> series formats, and thus I can avoid using import filter code. My vote
> is to please maintain this design.
> 
> Robert
> 
>> -----Original Message-----
>> From: 
>> insight-developers-bounces+r.atwood=imperial.ac.uk at itk.org
>> [mailto:insight-developers-bounces+r.atwood=imperial.ac.uk at itk
>> .org] On Behalf Of Stephen R. Aylward
>> Sent: 20 August 2007 23:20
>> To: kent williams
>> Cc: Hans Johnson; ITK
>> Subject: [Insight-developers] Re: Limitation in MetaIO
>> causing problems initkNeuralNetworkFileReader/Writer
>> 
>> Seems like the metaio library is being misused.  The fields are only
>> intended for holding tag/value pairs.   Reading/writing large
>> amounts of 
>> data should done directly to the file or to an external file.
>> 
>> Look at how matrix, arrays, and images are read/written -
>> first read the 
>> tag/value pairs that specify how much data should be
>> read/written, and
>> then directly read/write the data from the file after the tag/value
>> pairs.  Also, it should support writing those arrays as ascii
>> or binary 
>> data.
>> 
>> It could be done the way you suggested, but that was not the
>> design for 
>> metaIO.
>> 
>> Hope this helps,
>> Stephen
>> 
>> kent williams wrote:
>>> I logged this as a bug: http://www.itk.org/Bug/view.php?id=5545
>>> 
>>> The bug report has more details, but there are 2
>> limitations with MetaIO, in
>>> the context of the NeuralNet I/O classes: The
>> MET_FieldRecordType is a C
>>> structure, not a C++ class, and it contains a value array
>> with a fixed size
>>> of 255.
>>> 
>>> I'm testing a very quick and dirty fix -- simply enlarging
>> the value array
>>> so that Hans can move forward with his NeuralNetwork code
>> testing*, but
>>> MetaIO maybe deserves some new scrutiny.
>>> 
>>> Given that it's been around for years in ITK, I can't think
>> of what I'd do
>>> that wouldn't break backwards compatibility. The obvious
>> thing to do would
>>> be to move MET_FieldRecordType to a C++ class, make the value array
>>> resizable, etc.  Any code depending on the current state of
>> play would
>>> break.
>>> 
>>> 
>>> * and putting a test in MET_InitWriteField so it won't
>> scribble all over
>>> memory.  There are, as I count them, 4 separate potential
>> buffer overruns in
>>> that function.
>>> 
>>> 
>> 
>> -- 
>> =============================================================
>> Stephen R. Aylward, Ph.D.
>> Chief Medical Scientist
>> Kitware, Inc. - Chapel Hill Office
>> http://www.kitware.com
>> Phone: (518)371-3971 x300
>> _______________________________________________
>> Insight-developers mailing list
>> Insight-developers at itk.org
>> http://www.itk.org/mailman/listinfo/insight-developers
>> 
> _______________________________________________
> Insight-developers mailing list
> Insight-developers at itk.org
> http://www.itk.org/mailman/listinfo/insight-developers