#charts!
drowsy parcel
just in case ayone else wants to see the charts too (log scale Y axis). there's more off the high end but ti's pretty boring beyond that
notably, there does seem to be a "second peak" around 50s but it is very weak. 3 is also a bit of a weak showing relative to other low numbers, which i dont' have a particularly satisfying expanatin for yet
the 3 anomaly is much more prominent on a linear scale
oh, hmm
i wonder if ti's actualyl a 2 anomaly 🤔
valid timber
Yea the 3 is pretty odd
drowsy parcel
a peak aroudn 4, but 0 and 1 clamped to 2
that would make the shape make sense
i guess 0 wouln't happen, but 1 is feasible
valid timber
Huh, that would make sense. Since this is using the phobos compiler I could remove the max stack limitation, but I don't think I really need to since this makes a lot of sense
crystal zinc
how did you call the 2nd peak around 50. whats the intuition for that?
drowsy parcel
you mean when i had guessed and ballparked it there before any data? just having seen lots of locals-heavy main chunks
wild ass guess that got lucky
when you see lots of bytecode dumped funtions you see that count along with it and you just start to get a feel for it
valid timber
Here's the updated data with the 2 minimum stack limitation removed. Don't ask me how there's so many 0s 😂
And we can already see that you're right
drowsy parcel
0 basically ahs to be function() end 🤔
valid timber
yep
drowsy parcel
and yeah that unfolds nicely down there
3776 is a pretty small number among all the things, that could easily be a few doc files i'd think?
valid timber
Oh true! I forgot about those, they are included in several mods
drowsy parcel
that's about the only reason to have that many stubs
yeah taht makes a much more sensible shape at the bottom now
valid timber
(Oh you remade the graph with the different kind of input data, sorry!)
drowsy parcel
it's just google sheets
valid timber
I suppose with experience it's easy. I did install libre office and made a chart very quickly... but making it show what I want it to would take me time
drowsy parcel
the log scale / high end chart looks about the same of course
drowsy parcel
Here's the updated data with the 2 minimum stack limitation removed. Don't ask m...
well, from this i literally just pasted it, split ti to columns and told it to make a chart with the range still highlighted
the first one was harder since you gave me the raw data first and i had to get the counts first then chart it
but pre-summarized it's easy :p
and "just chuck it in a histrogram to see what it looks like" is a very common strategy of mine trying to get a feel for unfamiliar data :p
i'm actually surprised how clean and straight the line is even after the second peak, i expected teh chaos to set in much sooner
but i guess that's what a big data set does
valid timber
nice! I was going to share a screenshot but decided that it's not useful info 🙈
And for the record, 476 out of the 2490 files with empty functions in them contain .vscode in their path. That's about the best I got out of that
drowsy parcel
i'm actualyl more curious about the 1s now becuase all i can come up with is things of the form function() return "foo" end
the 0s are few enough that doc stubs seems to adequately explain them
but 1 is pretty hard to do 🤔
valid timber
There's actually a lot of real files with zeros in them. I've not looked at much, but one of them had an empty on version changed function, so it's also templates it seems
drowsy parcel
hmm, i guess an upval table lookup by constant string would only use 1 so you could look up a global function and call it
i would count templates to be equivalent to doc stubs tbh
valid timber
fair enough
drowsy parcel
just a differnet kind of docs
oh hmm
empty event handlers would be a 1
since they'd be declared with 1 arg and use nothing in body
maybe templates are a different group :p
valid timber
That's true. I could add param count and local count to the output too... Or even random things like instruction count. Honestly not sure how to determine what the 1s are about...
drowsy parcel
it's hard to come up with something doing useful work in 1 🤔
max stack includes the params already doesn't it?
valid timber
yea
drowsy parcel
oh you mean separate them
valid timber
exactly
drowsy parcel
my bet is on them being empty(ish) functions with 1 declared arg
placeholders that never got filled
i can't come up with anything that would work in only 1 beyond returning an arg as-is, returning a literal, or calling a global function by name with no args
valid timber
my bet is on them being empty(ish) functions with 1 declared arg
placeholders that never got filled
To determine that I'll try to use the formatter to print out just those functions so I can actually look at them. shouldn't be too hard, but unfrotunately won't be a graph :P
drowsy parcel
and i'm not even sure if returning an arg as-is would actually do it in 1, it might still copy it
valid timber
the return instruction will use single return values in place if possible afaik
drowsy parcel
but i'm pretty sure that ```lua
function()
some_global()
end
actually I know that it does since this is using Phobos which does do that (even when multiple are in sequence), but I do believe Lua itself would also do it
drowsy parcel
i think plain lua is a little less optimistic about it but i don't remember for sure off hand
crystal zinc
util has a bunch of 1 line functions like
function util.by_pixel(x,y)
return {x / 32, y / 32}
end```
and cliff rotations use a couple
could there be a lot of that elsewhere. hidding a lttile math formulas away in a 1 line function.1 register, not 1 line
valid timber
i think plain lua is a little less optimistic about it but i don't remember for ...
Confirmed, it uses it in place if it's just 1 return value, otherwise it gets put in temporaries
drowsy parcel
yeah they didn't want to try to see if the whole set was in place in order, i recall being very excited about phobos being able to pull that trick :p
drowsy parcel
1 register, not 1 line
counting registers is a much weirder measure, and 1 is an insane restriction - the lua compiler won't even bother attempting to compile a function with less than 2 normally, jan specifically modified one to get us the 0 and 1 counts for this
because even most basic operations involve at least 2 registers
and function params are reserved right off the top, so if you use the 1 register you get on that you can't use basically anything else
valid timber
Had to run afk for a bit there, but yea that's a good summary of registers.
drowsy parcel
util has a bunch of 1 line functions like
```
function util.by_pixel(x,y)
re...
just compiling this in my head, this one has 2 for the args, 1 for the newtable, plus two for the values for setlist to populate it, for a total of five, i think?
crystal zinc
ah. so even
function(x) return (x+1)
is multiple registers?
drowsy parcel
yeah it can't reuse teh one x is in, so it has to allocate another temp for x+1 to go in
oh, but ```lua
function(x)
x=x+1
return x
end
shy python
primitives are also treated like pass by reference right? then it should fit with 1
valid timber
oh, but ```lua
function(x)
x=x+1
return x
end
``` might work in 1 🤔
Yep it would. binop instructions are pretty powerful
drowsy parcel
my main question is if the complicated semantics of = allow it to do it all in one, but i'm pretty sure since x is a local here it all works
i don't htink i invoked anything fancy on the = there with no tables and no commas, so it should behave...
but yeah the only other one-register operation i've come up with is table lookup from an upval table with a contant key, aka global variable lookup
function()
local x = global_x
x=x+1
global_x = x
return x
end
🤔
valid timber
yea, = is smart (which is also why you've called it complicated, kind of related), but just to be sure I double checked
-- lua: function at @temp/test.lua:2-5
-- lua: 1 params | 0 upvals | 2 max stack
-- lua: 3 instructions | 1 constants | 0 functions
local function f(x)
-- 3 0 lua: 2f 1 ADD R(0|x) := R(0|x) + 1 [a 0] [b 0] [c 256]
x=x+1
-- 4 0 lua: 2f 2 RETURN return 1 results starting at R(0|x) [a 0] [b 2]
return x
-- 5 0 lua: 2f 3 RETURN return 0 results starting at R(0|x) [a 0] [b 1]
end
Don't even have to! 🪄
-- lua: function at @temp/test.lua:2-5
-- lua: 0 params | 1 upvals | 2 max stack
-- lua: 6 instructions | 2 constants | 0 functions
local function f()
-- 3 0 lua: 2f 1 GETTABUP R(0) := Up(0|_ENV)["x"] [a 0] [b 0] [c 256]
-- 3 0 lua: 2f 2 ADD R(0) := R(0) + 1 [a 0] [b 0] [c 257]
-- 3 0 lua: 2f 3 SETTABUP Up(0|_ENV)["x"] := R(0) [a 0] [b 256] [c 0]
x=x+1
-- 4 0 lua: 2f 4 GETTABUP R(0) := Up(0|_ENV)["x"] [a 0] [b 0] [c 256]
-- 4 0 lua: 2f 5 RETURN return 1 results starting at R(0) [a 0] [b 2]
return x
-- 5 0 lua: 2f 6 RETURN return 0 results starting at R(0) [a 0] [b 1]
end
naming it saves looking it back up the second time though
valid timber
This is true
valid timber
Alright, so looking at the functions with 1 stack (which I can't legally share unfortunately) it seems like lots of them are simply interacting with globals. Even things like game.surfaces[1].foo = true.
Or function calls to functions that are in a global table, without arguments of course.
Or a constructor function, without paramters, that creates a record only table (no list) and returns that.
Yea looking through more and more, global variables is the answer. You can even have if statements.
drowsy parcel
oh huh, i guess it woudl reuse the temp for a deep access 🤔
valid timber
yep