12:07:34ak-coramManaged to work around that ECL threading issue from the weekend: https://github.com/ak-coram/cl-duckdb/pull/10
12:08:12ak-corambut seem to have stumbled on one of my benchmarks doing a lot of consing on ECL compared to SBCL
12:09:14ak-coramSBCL conses about 3MB, but ECL above 2GB for the same function
12:14:01ak-coramhere's the code, append-row is just a loop on params with one large dispatch to a bunch of C functions via ecase: https://termbin.com/9mui
12:25:01ak-coramalso weird, I seem to be getting a major difference between running roswell vs ECL that comes with my distro (it's the same ECL-version)
12:47:25ak-coramThe difference seems to be some dependency in quicklisp vs ultralisp
13:48:30ak-coramI'm not sure what library is causing this, but using ultralisp seems to kill most of the allocations (down to about ~250MB) with ECL. Still way more compared to SBCL & CCL
13:49:22jackdanielthis seems like some serious drawback, yes
13:49:28jackdanielif you narrow the test case then it will be useful
13:49:50ak-coramis there anything besides "time" I could use to profile this?
13:50:21jackdanielno, ecl doesn't have a memory profiler
13:50:34jackdanielyou could possibly use some c-specific tool, it will work with ecl