WLS: false positive bug

[dexter1]

FYI: While testing the WLS smartsearch routine i ported in lifesrc to be able to run on the command-line, i received a false positive when searching for a 2c/5 spaceship in life rule. After a day of debugging the code and finding nothing, i ran WLS (Both K and J versions) and could confirm the false positive:

wlsk.png (50.57 KiB) Viewed 2161 times

This is reproducible with g5, r14, c8, row-mirror, -2 x cells right, no smartsearch strategy. It doesn't matter what search strategy one chooses, or which direction one wants to search.

I haven't found an earlier post about this specific bug in WLS. I'm unsure if this is easy or even worthwhile to fix. My guess is that is has to do with the large speed and the fact that there is only a one-cell empty border surrounding the searchbox: The 2-cell translation might skip over the empty border in the transition from g-1 generation slice to the 0 generation slice.
I'll poke around in my command-line port on Linux (since i'm more accustomed to code development and profiling over there), and see/try if i can find a cause/fix/workaround.

[Sokwe]

My guess is that is has to do with the large speed and the fact that there is only a one-cell empty border surrounding the searchbox

As I recall when I looked into this many years ago this was the issue, although I can't say much more than that. As a result, I habitually surrounded my WLS searches with a two-to-three cell border of dead cells, which seemed to resolve the matter. The same should be done in JLS when using nonzero translation, as not doing so sometimes results in a strange error that causes the search to stop.

[HartmutHolzwart]

In some sense this bug is also a feature that sometimes leads to puffer engines. Especially in short wide searches.