On Thu, 28 Jun 2012 14:05:10 -0500
"David C. Rankin" <drankinatty@...> wrote:

> On 06/27/2012 03:04 PM, Darrell Anderson wrote:
> > Considering the original volume of changes needed, we're about as okay as we
> > can expect, but --- we still find branding issues. For example, today I found
> > some more and I am pushing patches.
> > 
> > To resolve remaining branding issues we need the following:
> > 
> > * Some really smart person to grep the system properly looking for corrupt
> > image files (png, xcf, jpg/jpeg, whatever else might be out there). We can't
> > inspect those images for branding issues. For whatever reason, seems many of
> > the images got corrupted with CR/LF issues. We have found most of the corrupt
> > files (walch.martin found most of them --- I only pushed clean files), but I
> > don't know of a good way to search the sources for corrupt image files.
> > 
> > (Note: Corrupt image files do not block building packages, but they cause
> > usability issues that reflect badly on us).
> > 
> 
> This is tricky. Using 'file' to test even the corrupt jpeg images will return a
> valid jpeg header in most cases and miss the fact that the image is corrupt. I
> don't mind grepping or running find tde/main -type f -name "*.jpg" -execdir
> <what> '{}' \;, but I need to know what <what> will do the trick. Ideas?

Check to see if <CR> appears separately anywhere in the file, without a trailing
<LF>?  If we're seeing a binary <-> text mode corruption problem, there won't
be any separate <CR>s.  Any file that's flagged as only containing the <CRLF>
sequence should then be checked manually in an image viewer to make sure it's 
okay.

(I don't know how to do the above with grep etc., but the actual match takes
1 line in Perl.)