Re: Delinearisation problem

[Michal H. <ms...@gm...>]

On Thu, Aug 11, 2011 at 09:36:53AM +0200, Michal Hocko wrote:
> On Fri, Aug 05, 2011 at 01:05:06PM +0200, Michal Hocko wrote:
> > On Fri, Aug 05, 2011 at 12:19:52PM +0200, Jozef Misutka wrote:
> > > Dne 4.8.2011 22:54, Michal Hocko napsal(a):
> > > >On Thu, Aug 04, 2011 at 10:28:50PM +0200, Michal Hocko wrote:
> > > >>On Tue, Aug 02, 2011 at 11:03:46PM +0200, Jozef Misutka wrote:
> > > >[...]
> > > >>>The original pdf is also corrupted. If you print the object 7665 0:
> > > >>>
> > > >>>...
> > > >>>/IDTree 7666 0 R
> > > >>>/K 7667 0 R
> > > >>>/ParentTree 7668 0 R
> > > >>>/ParentTreeNextKey 4718
> > > >>>/RoleMap<<
> > > >>>/1Heading /P
> > > >>>/1Heading (appx) /P<-----------------
> > > >>>/2Heading /P
> > > >>>/2Heading (appx) /P<-----------------
> > > >>>/3Heading /P
> > > >>>/4Heading /P
> > > >>>...
> > > >>>
> > > >>>Rolemap is a dictionary object so it expects key/value pairs, it gets
> > > >>>1Heading: P (ok)
> > > >>>1Heading: apps (ok)
> > > >>>but it seems strange at this point because the next key is
> > > >>>P: 2Heading (sill ok, however strange)
> > > >>>P: 2Heading (sill ok, however strange)
> > > >>>and then it gets a string instead of name object (appx)
> > > >>Not exactly. This is how we print that object. In fact the entry is
> > > >>"/1Heading (appx)" as a key and "/P" as a value. Name object cannot
> > > >>contain any spaces but it might contain numeric representation of that
> > > >>character. Xpdf code turns those numeric rep. into characters.
> > > >>
> > > >>I am wondering how we ended up with a plain space there as
> > > >>makeNamePdfValid substitutes all dangerous characters by #CODE so we
> > > >>should see /1Heading#20(appx) in the file. This sounds pretty much like
> > > >>a bug in our code.
> > > >OK, I've got it. The problem is in complexValueToString<CDict>  which
> > > >puts key blindly without any validation. The patch to fix this is
> > > >attached.
> > > 
> > > Did you really tried the patch with the specific file from my email?
> > 
> > As I wrote I had issues to build gui so I haven't tried it. But it
> > should fix the issue with the spaces in key values.
> 
> Sorry it took so long but I've been busy recently.
> I have looked at Lexer and how it handles name objects and found out
> that the patch I have sent last time is not sufficient (albeit we need it).
> The other problem is in makeNamePdfValid itself. 
> We follow a PDF specification recommendation to represent all characters
> out of range [~, !] as #hexa numbers. This is not sufficient, though
> (just consider `(', `)' and there are others).  The attached patch fixes
> that.
> 
> I've tested both patches by delinearizator and pdf_object_printer tools
> as I am still not able to build gui now.
> 
> Both patches are attached for reference.

Did you have time to look at those patches? Should I push them to the
repository?

-- 
Michal Hocko