<_whitenotifier> [prjunnamed/prjcombine] wanda-phi pushed 1 commit to main [+0/-0/±14] https://github.com/prjunnamed/prjcombine/compare/e57657511a86...8648e616ac15
<_whitenotifier> [prjunnamed/prjcombine] wanda-phi 8648e61 - icecube-harvester: use `Toolchain` infrastructure.
Wanda[cis] has joined #prjcombine
<Wanda[cis]> I'm really happy with the caching layer
<Wanda[cis]> although, heh
<Wanda[cis]> now that icecube runtime is out of the picture, I can actually feel the slowness of the core bit correlation algorithm
<Wanda[cis]> I wonder if it's worth optimizing it
<Wanda[cis]> 2 minutes to rebuild the iCE65L08 database (entirely from cache)
<Wanda[cis]> not greate, not terrible
<Wanda[cis]> s/greate/great/
<Wanda[cis]> well. I guess it's profiler time.
<Wanda[cis]> ... okay how do I get a rust profiler that actually works and shows me a flamegraph that's not patent nonsense
mei[m] has joined #prjcombine
<mei[m]> i assume you've tried cargo-flamegraph?
<Wanda[cis]> yes
<Wanda[cis]> it generates something incomprehensible
<Wanda[cis]> the stack traces are batshit
<Wanda[cis]> also most of it is BTreeMap with absolutely no indication of which BTreeMap it is
<Wanda[cis]> I'm going to update absolutely everything on this machine and see if it helps.
<Wanda[cis]> it did not.
<Wanda[cis]> something is badly broken with gathering call stacks
<Wanda[cis]> ... ah. apparently it needs a secret linker option to work with mold.
<Wanda[cis]> okay. fine. --no-rosegment added to cargo config.
<Wanda[cis]> now it produces something that actually makes sense, good
cr1901 has quit [Read error: Connection reset by peer]
cr1901 has joined #prjcombine