slset/benchmarks, branch masterLists of symbols as sets
https://code.more-magic.net/slset/atom?h=master2025-08-19T08:17:33ZUse singular "slset" instead of plural "slsets"2025-08-19T08:17:33ZPeter Bexpeter@more-magic.net2025-08-19T08:17:33Zurn:sha1:1007f7aef69d3fe28942ad74cc002b49bd1eefaaMinor change in benchmark2025-08-18T11:39:32ZPeter Bexpeter@more-magic.net2025-08-15T11:34:27Zurn:sha1:a64d4e18ade6e0eca23806524f863852594f229d
Do not count list construction towards benchmark time.
This doesn't appear to make a big difference but it's good to pull
that out of the calculation.
Change Scheme benchmark to allow graphing by measuring time and looping2025-08-18T11:39:32ZPeter Bexpeter@more-magic.net2025-08-15T07:22:08Zurn:sha1:fa8d354bc49b0adc77e6dd6d973c8bd2d33f2769Actually use both different sets2025-08-18T11:39:32ZPeter Bexpeter@more-magic.net2025-08-05T09:01:40Zurn:sha1:098c6ba98e7cca7b0810d9283e2afd4a9ac5519b
The original implementation just diffed the same set with itself,
and needed a workaround for srfi-1's shortcut on the identical list.
Since we're comparing to other implementations with differing sets, it
makes sense to do the same here.
We actually started doing this but ended up just redefining the same
list...
Add benchmarks for Ruby, Python and Clojure2025-08-18T11:39:32ZPeter Bexpeter@more-magic.net2025-08-05T08:37:58Zurn:sha1:12a4f8e2415a767c501898d3626663f78fb3c33d
Ruby is about the same speed as slets. Clojure (babashka) and Python
are goddamn fast. Python is 10 times as fast. Clojure is twice as
slow as Python but that still outperforms Ruby and slsets.
Disable interrupts2025-08-18T11:39:32ZPeter Bexpeter@more-magic.net2025-08-05T08:27:31Zurn:sha1:565171c7b0a7389e60ec6f8593732988f43b6aef
Doesn't do that much, but it's a good idea for slsets as well because
it makes it unlikely that plist changes are ever observed, because
there's no thread switching.
Add custom hash table implementation to compare direct hash-table-as-set2025-08-18T11:39:32ZPeter Bexpeter@more-magic.net2025-08-05T08:07:40Zurn:sha1:53f99c5612daadf18e90475159c7683443033b0d
This uses hash-tables directly instead of creating the hash table from
the lset every time. It's faster than using srfi-69 because it can be
inlined better.
Add hacky cache for symbol hash2025-08-18T11:39:32ZPeter Bexpeter@more-magic.net2025-08-05T07:41:51Zurn:sha1:7f242c33130d0c6671f307e9a6224fa74765582c
Makes a bit of an impact - about 20% faster
Initial implementation of slsets egg2025-08-18T11:39:31ZPeter Bexpeter@more-magic.net2025-07-29T12:52:12Zurn:sha1:2da2aae3a1ff27f62b9d87fddf26bf1caf6ea687
This implements sets of symbols *as lists*, like in the srfi-1 lset
operations. It maintains the benefit of using lists for everything,
but not having to pay the quadratic performance penalty that srfi-1
entails. We achieve this by using plist operations to "mark" items
so we have to visit the list "set" only a fixed number of times.