Why we should not enable a slow DFA by default

[Richard (Rikki) Andrew Cattermole]

On 19/12/2025 9:40 PM, Richard (Rikki) Andrew Cattermole wrote:
> Out of curiosity I wanted to know how many functions were @safe before 
> and after the extra 2 passes took place.
> 
> Before: 2900 out of 6595
> 
> After: 106 out of 1103
> 
> That is a pretty big reduction.

I managed to get the number of functions failing to analyze to zero.

Had to adjust my check to account for extern functions without bodies, 
and functions in non-root modules that have not been analyzed.

I also had to disable my workaround for not having a way to mark 
functions as CTFE only.

Either way, the good news is one pass is all that is needed to offer 
higher support for interprocedural analysis.