summaryrefslogtreecommitdiff
path: root/infra
diff options
context:
space:
mode:
authortsepez <tsepez@chromium.org>2016-09-20 11:28:25 -0700
committerCommit bot <commit-bot@chromium.org>2016-09-20 11:28:25 -0700
commit393fe4943226846a9b99878406d0bf75f31bb643 (patch)
tree1a70376a4c9caf54fe309782877d9f3033b51cb6 /infra
parent9972ff99285cea12a20026136e98c1e635a15010 (diff)
downloadpdfium-393fe4943226846a9b99878406d0bf75f31bb643.tar.xz
Assert that dictionary can own the objects it is given.
Upon indirect object holder destruction, all indirect objects are destroyed -- currently by order of increasing object number -- but ideally without ordering constraints. So currently, we can get away with a dictionary pointing directly at an indirect object with a higher number. It gets destroyed first, invoking Release() on its subordinates, which skips destroying them if they are indirect objects. But we don't want to rely on this artifact of destruction order. Should it happen to be reversed, the dictionary would invoke Release() on freed memory. Interestingly, CPDF_Array skirts the issue by replacing any indirect objects it is given with references. Not clear whether we should do the same thing for dictionaries, or remove it from arrays. The technique certainly complicates understanding ownership. The one violation found is in the unittest that broke the previous CL which tried to use unique_ptrs in indirect object holder. Review-Url: https://codereview.chromium.org/2353093002
Diffstat (limited to 'infra')
0 files changed, 0 insertions, 0 deletions