[poppler] Valgrind error with tiling patterns in Poppler (and probably XPDF)

Derek B. Noonburg derekn at foolabs.com
Thu Nov 8 19:51:42 PST 2007


On  4 Nov, Brad Hards wrote:
> G'day team,
> 
> I recently [0] posted a message asking for help understanding a weird looking
> Valgrind backtrace. Thanks to a hint from Krzysztof Kowalczyk, I rebuilt
> poppler without optimisation, and got a better backtrace [1].
> 
> This comes from the PDF Reference, Fourth Edition, Version 1.5 [2]. This doesn't
> appear to occur with the compressed version [3]. The problem page is 1012 (if 
> you count zero based), 1013 (if you count 1 based), and 991 (if you ignore the 
> roman numbered intro pages) - it shows Plate 8 and Plate 9.
> 
> This bug appears to have no visual impact on my system - it is just a valgrind
> alarm. It really does look like an uninitialised variable though.

Yes, it's an uninitialized variable.  I ran into that a while back.  The
real fix is more complicated (Gfx::doTilingPatternFill had some issues)
-- it will be in the next Xpdf release.  Your patch is probably ok to
avoid the uninitialized data access.  As I recall, the uninit value
never gets used (i.e., never affects what shows up on the screen), but
it does propagate through a few levels.

- Derek



More information about the poppler mailing list