https://bugs.kde.org/show_bug.cgi?id=417993
Eyal <eyals...@gmail.com> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |eyals...@gmail.com
--- Comment #4 from Eyal <eyals...@gmail.com> ---
For each operand, we run multiple tests with various settings of value and
vbits. A test ought to look like this:
1. Create variables aa and bb and vaa and vbb.
2. Set the vbits of aa and bb to be vaa and vbb, respectively.
3. Run that operand on aa and bb.
4. Extract the vbits that valgrind computed on the result.
And for a check, we do this:
1. Run a helper function whose input is aa,bb,vaa, and vbb.
2. That helper function computes the vbits that we expect in the output of the
operand.
Finally, we compare the expected vbits against what valgrind came up with.
In the function valgrind_execute_test, we set the vbits of aa and bb before we
run the operand. That part is fine.
But when we run the helper with the inputs aa,bb,vaa,vbb, there is no
uncertainty there. We have, in fact, extracted the uncertainty from aa and bb
into vaa and vbb and now, aa and bb have no uncertainty. They should have
vbits 0, both of them. It's now up to the helper function to use the definite
values of aa,bb,vaa, and vbb to compute the definite expected vbits of the
operand.
It would be easy to check. Just add a loop over the operands into
check_result_for_binary that does this:
opnd->vbits.bits = 0;
valgrind_set_vbits(opnd);
Maybe like 5 lines of code total? I don't have the right platform to test this
out, though... If I make a patch, would someone like to test it?
Eyal
--
You are receiving this mail because:
You are watching all bug changes.
