On Mon, 02 Feb 2004 16:34:22 -0800, Lester Gong <ljmgong@yahoo.com> wrote:
There is a bug in dynamic_bitset operator<<=().
Calling operator<<=() followed by operator>>=() may shift in non-zero bits from the left.
For example, this sequence of statements ...
dynamic_bitset<> b(std::string("0110"); b <<= 1; // gives "1100" b <<= 1; // gives "1000" b <<= 1; // gives "0000" b >>= 1; // gives "1000" !! b >>= 1; // gives "1100" !! b >>= 1; // gives "0110"
Seems that some stray bits are not cleared when shifted beyond the lhs of the bitfield.
Yep, that was a stupid bug by me :( It was fixed a while ago, when I added an assert in the dynamic_bitset destructor that checks the invariants of the objects being destroyed. Unfortunately the official dynamic_bitset has been unchanged for long time, for various reasons (there's a more recent version in the sandbox, but that's still outdated) and so the bug is still there. The good news are that a major update will be uploaded in the next days. Stay tuned :) I'll post a message to the list as soon as it will be in the main CVS repository. BTW, I would be really glad to know what you are using dynamic_bitset for.
Attached files:
test_dynamic_bitset.cc Simple program that illustrates bug test_dynamic_bitset.out Output from simple test program dynamic_bitset.hpp.patch Patch that seems to fix the bug
bitset_test.hpp.patch dyn_bitset_unit_tests2.cpp.patch Proposed changes to test programs that would catch this problem.
*Thank you* very much, Lester. That's a lot of code. I've added your name to the source. As to the test programs I'm reluctant to add test for specific sequences of functions (<<= followed by >>=, etc.), unless there's a inherent logic reason. What happened here is that an invariant was broken, and that could have caused problems whatever (well, almost) function we called after <<=. Had you called count() for instance, that would have failed as well. Genny.