Luke Boswell (Oct 17 2023 at 03:50): Luke Boswell (Oct 17 2023 at 03:50):Is there a limit on the size of an expression? If I write a function which includes an expression with thousands of boolean and arithnetic operations?
I'm looking to write a function List U32 -> List [A,B,C,D] that is efficient and I figured branchless would give a fixed number of calculations per U32. The use case is mapping unicode code points into graphmeme cluster break classes.
The other idea was to create a top level Dict U32 [A,B,C,...] but not sure if that would be more efficient?
I can write a benchmark, but thought I might ask here first and save myself in case there is an obvous solution to reach for.
Brendan Hansknecht (Oct 17 2023 at 03:55):Like a when expression?
Brendan Hansknecht (Oct 17 2023 at 03:55):Just need something more concrete to understand?
Brendan Hansknecht (Oct 17 2023 at 03:59):Or a doc with some technical info on what you want to do would be good
Luke Boswell (Oct 17 2023 at 04:04):Yeah, I could explain that better. I mean like a huge sequence of comparisons, like isControlClass = (u32 > 0x0 && u32 <= 0x9) || (u32 >= 0xB && u32 <= 0xC) || (u32 == 0x61C) || ...
Luke Boswell (Oct 17 2023 at 04:07):I might be mixing operators there... but I mean boolean.
Luke Boswell (Oct 17 2023 at 04:11):I have wondered if it would be efficient to use some kind of lookup table, but that seems like a lot of memory. I've done some research on that and maybe it can be reduced by using multple lookups in sequence. The goal is U32 -> [A,B,C,...N].
Luke Boswell (Oct 17 2023 at 04:13):I can imagine creating the lookup tables by sampling every possible value from 0x0 to 0x10FFFF, and then generating the equivalent roc List [A,B,C,...].
Luke Boswell (Oct 17 2023 at 04:19):https://www.unicode.org/Public/UCD/latest/ucd/auxiliary/GraphemeBreakProperty.txt
Luke Boswell (Oct 17 2023 at 04:20):This doc maps the unicode code points to a graphmeme cluster break property. The rules for when to split up into extended graphmeme clusters are determined using these property values.
Brendan Hansknecht (Oct 17 2023 at 04:33):Ah yeah, I would do the giant Boolean expression and not waste the memory on something that big
Brendan Hansknecht (Oct 17 2023 at 04:33):That said, if it can be represented as a chain of very small lookups or a few medium lookups, that could be faster.
Brendan Hansknecht (Oct 17 2023 at 04:34):Depends mostly on the size of the lookup and what it optimized to
Brendan Hansknecht (Oct 17 2023 at 04:34):The Boolean expression should optimize just fine
Brendan Hansknecht (Oct 17 2023 at 04:35):If it is faster than the lookup mostly depends on if the lookup will be in cache and just how many instructions it becomes
Last updated: Jul 05 2025 at 12:14 UTC