Circuit ORAM recursion broken by changes in Obliv-C and misc

[samee]

The current HEAD is broken on even circuit ORAM tests:

$ make build/test/testOramAccess
$ build/test/testOramAccess -- 1234 ckt --auto=512 &
$ build/test/testOramAccess localhost 1234 ckt --auto=512
[...stacktrace...]

The issue also exists with bench/search. Rebuilding with make CFLAGS=-g allows valgrind to look into the problem. bench/search was more time-tractable with valgrind:

$ make CFLAGS=-g build/bench/search
$ valgrind build/bench/search bench -t ckt --size=257 --axcount=6 :1234 &
$ valgrind build/bench/search bench -t ckt localhost:1234

Logs attached

Xiao noted in email that the code is significantly different from the one in published version. While we chase down regressions, I'll try to upload the original version on a separate branch in case it helps anyone.

[samee]

Actually, I don't need to upload it. Jack's fork of the previous code is still public. Anybody interested can try this version: https://bitbucket.org/jackdoerner/absentminded-crypto-kit

Xiao and I will try to figure out what the intention behind this change was, and why it led us astray.

[jackdoerner]

Indeed. I was unaware there were updates. Now that I know, I'll wait to merge them in until this issue is resolved. A thought: you and I are both providing libraries that provide ORAM in Obliv-C, and our codebases are mostly-identical. Someday we may want to do some sort of reconciliation between them.

[samee]

Patch applied. Will close this bug after testing a bit more thoroughly. See if it is working now.