From 70f9dc9acf47b2ffb98c89fce17dd939c84cd27d Mon Sep 17 00:00:00 2001 From: Jay Berkenbilt Date: Mon, 31 May 2010 17:12:22 +0000 Subject: current notes git-svn-id: svn+q:///qpdf/trunk@969 71b93d88-0707-0410-a8cf-f5a4172ac649 --- TODO | 35 +++++++++++++++++++++++++++++++++++ 1 file changed, 35 insertions(+) (limited to 'TODO') diff --git a/TODO b/TODO index fc41d948..2d7fb1ed 100644 --- a/TODO +++ b/TODO @@ -1,3 +1,38 @@ +Bug +=== + + * Running + + valgrind --leak-check=full .../qpdf --check memory-leak.pdf + + shows a number of QPDFObjects that were not deallocated. They all + seem to be objects allocated by dereference() while in + flattenScalarReferences. + + I'm not sure at this time how it's possible that this memory is + leaked because everything is based on PointerHolder. I've already + ruled out the way the cache is handled...the objects in the cache + overlap with the leaked objects, but they are not exactly the + leaked objects. + + Most files don't exhibit leaks, but all files that have outlines + do. I don't know whether it's because of circular object + references. + + General debugging: put a break point on 'stop_here()' and run qpdf + --check memory-leak.pdf in gdb. For simplicity, configure with + --disabled-shared CFLAGS=-g CXXFLAGS=-g. This will cause gdb to + stop at the point of allocation of all the objects that are leaked. + I'm still not sure which of these is the one that valgrind is + complaining about, and I'm not really even sure that they're + legitimate, though it's hard to imagine that they aren't. + + +Next +==== + + * Work in contrib/vc6. See email from stronghorse@tom.com + 2.2 === -- cgit v1.2.3-54-g00ecf