[cfe-dev] Two pass analysis framework: AST merging approach

Artem Dergachev via cfe-dev cfe-dev at lists.llvm.org
Tue Aug 9 02:09:43 PDT 2016


Tweaking the current search algorithm is a good thing to try, but it's 
clearly a matter of experiment a lot more than rational thinking. In 
fact, i've no idea why exactly are we doing depth-first nowadays, but as 
far as i remember it was much worse when we tried breadth-first.

It seems that because most path-sensitive bugs manifest themselves on a 
huge number of different paths, the approach that tries to take at least 
one of those paths [before hitting some limit] turns out to be superior.

On 8/9/16 12:03 PM, Gábor Horváth wrote:
> Hi!
>
> A side note, it might be worth to consider IDDFS instead of DFS 
> (https://en.wikipedia.org/wiki/Iterative_deepening_depth-first_search), 
> to get more optimal coverage patterns regardless of the "shape" of the 
> available code.
> What do you think?
>
> Regards,
> Gábor
>



More information about the cfe-dev mailing list