My first attempt ever at reading footpedals just polls HID input reports and then manually worked out the bitfields.

Coming back to it, and *actually reading*, you can just... read the button values via the HID API.

Well... that makes my job easier.

The circle for whom this is actually useful/relevant is, like, me. But that's also my target audience so...

Still, actually trying to do things "properly" turns out to be... easier? than hacking it yourself??

(within reason)

kind of like we've been in this place before

hey wait a minute

*'11111', sorry. Currently, it's 2 ops/ms. 8192 seeds takes just under 4s.

As a comparison, using a log function yields 8388608 seeds in .6s

oh I'm scared for '1111'

woof, and it just gets... worse
starting from seed '11', our ops/ms drops to 544. At seed '111', it plummets to 66 ops/ms

starting from seed 0, using the skip method results in, at best, about 695 ops/ms

working from a much higher seed using the log method gives 13247 ops/ms

oh I didn't quite realise *how* serious the speedup is - it's super dependent on the seed as well. higher seeds take longer to spin up.

At current speed, iterating the entire set of seeds would take ~5 days, which is honestly... not too shabby

Thinking about how to cache/store results moving forward.
Initial thought was chuck it all in a database, but at 1 byte per seed that's in the ballpark of ~3 TB...

Ran it overnight, found a few bugs (concurrency...) but found more 52-rank decks, also 10s of spades. Interesting.

I have no idea of what I'm doing with OpenCL so most of my improvements so far are, like, barely improvements.
I thought I broke my filter when I saw the speeds drop from 5s to .8s

oh yeah I'm here and trying to post here too ig enjoy!

dealing with Python atm and sqlite3 coalescing "missing" into `None`, and "null" into `None`...
I'm getting Java flashbacks and I don't like it :<