Packed code misbehaving in Chrome

[nrkn]

Wondering if you had any advice or if you've come across something similar before

My packed js1k entry is misbehaving in Chrome (working in FF, Edge) - the unpacked version works fine

If you want to see the issue happening, this is the project, you can just open index.html in Chrome:

https://github.com/nrkn/js1k-2018-roguelike

As soon as you go down the stairs - > - on the first level it starts behaving strangely

Using the online packer with 1/0/0 most copies first

Code in question, before packing:

let l,t=0,e=[0,0,0,10,"@","#39f"],f=l=>Math.random()*l|0,r=(l,t)=>{for(let e=0;e<l.length;e++)if(l[e][3]&&t[0]==l[e][0]&&t[1]==l[e][1])return l[e]},h=(l,t,e,f)=>{f?l[0]<t[0]?t[0]--:l[0]>t[0]?t[0]++:l[1]<t[1]?t[1]--:l[1]>t[1]&&t[1]++:1==e?t[1]--:2==e?t[0]++:3==e?t[1]++:t[0]--},o=()=>{l=[[[e[0],e[1],3,1,".","#aaa"]],[e]];let a=f(10*t)+10,o=f(10*t)+10,i=f(2*t)+2,n=f(2*t)+2,u=f(2*t)+2,c=(t,e,h,i)=>{let n=[f(a),f(o),t,e,h,i];return r(l[0],n)&&!r(l[1],n)?(l[1].push(n),n):c(t,e,h,i)},d=(t,e)=>{if(r(l[0],e)||l[0].push([e[0],e[1],3,1,".","#aaa"]),t[0]==e[0]&&t[1]==e[1])return;let a=f(4);h(t,e,a,!f(3)),d(t,e)};for(let t=0;t<i;t++)d(l[0][f(l[0].length)],[f(a),f(o)]);c(2,1,t>8?"$":">",t>8?"#39f":"#000");for(let l=0;l<n;l++)c(1,1,"m","#f00");for(let l=0;l<u;l++)c(4,1,"¢","#f90")},i=()=>{a.width=a.width;let f=e[3]<1,h=t>9;for(let t=0;t<25;t++)for(let a=0;a<25;a++){let o=e[0]-12+a,i=e[1]-12+t,n=r(l[1],[o,i])||r(l[0],[o,i]);n||(l[1].push([o,i,5,1,"#","#aaa"]),n=r(l[1],[o,i])),a>=8&&t>=8&&a<17&&t<17&&(n[6]=1),c.fillStyle=f||h?"#39f":n[5],c.fillText(h?"$":f?0:n[6]?n[4]:" ",10*a,10*t)}f||h||(c.fillStyle="#000",c.fillText(1+t+" ¢"+e[3],0,250))},n=(a,i)=>{let n=[a[0],a[1]];h(e,n,i,1==a[2]&&f(5));let u=r(l[1],n);u&&1==a[2]&&0==u[2]&&f(2)?u[3]--:u&&0==a[2]&&1==u[2]&&f(2)?u[3]--:u&&0==a[2]&&2==u[2]?(t++,o()):u&&4==u[2]?(a[3]++,u[3]--):r(l[0],n)&&!u&&(a[0]=n[0],a[1]=n[1])};b.onkeydown=(t=>{n(e,t.which-37);for(let t=0;t<l[1].length;t++)l[1][t][3]&&1==l[1][t][2]&&n(l[1][t],f(4));i()}),o(),i()

Any ideas? I thought I could maybe work around it by using different settings but sadly none of the others get it down to 1024 bytes

[nrkn]

I made a few changes, this time it would pack to 1024 with several different combinations of RegPack settings, but all still not working properly in Chrome. Really strange! There's just something it doesn't like and can't figure out what.

[nrkn]

I was able to prevent it by wrapping it an IIFE and then golfing back the lost bytes

If it's something you want to look into regardless, I put the offensive code in this branch:

https://github.com/nrkn/js1k-2018-roguelike/tree/chrome-pack-problem

[Siorki]

Wow. You redid Nethack in 1k ?

I managed to repeat the issue in Chrome only, works well under FF. With Opera, the bug only shows up after several (at least two) flights of stairs.
Once triggered, it seems like the display is out of sync, the upper level still shows up but collision detection uses data from the lower level. Is that also what you get ?

Differences in behavior between packed and unpacked code are usually caused by variables from the unpacking routine (_ and G) being initialized before eval() is called. I'll investigate in that direction first.

[Siorki]

It seems Chrome's implementation of eval() doesn't perform as expected.
I wrapped your code into a string and evaluated it, skipping the unpacking entirely :

_='let l,t=0,e=[0,0,0,10,"@","#39f"],f=l=>Math.random()*l|0,r=(l,t)=>{for(let e=0;e<l.length;e++)if(l[e][3]&&t[0]==l[e][0]&&t[1]==l[e][1])return l[e]},h=(l,t,e,f)=>{f?l[0]<t[0]?t[0]--:l[0]>t[0]?t[0]++:l[1]<t[1]?t[1]--:l[1]>t[1]&&t[1]++:1==e?t[1]--:2==e?t[0]++:3==e?t[1]++:t[0]--},o=()=>{l=[[[e[0],e[1],3,1,".","#aaa"]],[e]];let a=f(10*t)+10,o=f(10*t)+10,i=f(2*t)+2,n=f(2*t)+2,u=f(2*t)+2,c=(t,e,h,i)=>{let n=[f(a),f(o),t,e,h,i];return r(l[0],n)&&!r(l[1],n)?(l[1].push(n),n):c(t,e,h,i)},d=(t,e)=>{if(r(l[0],e)||l[0].push([e[0],e[1],3,1,".","#aaa"]),t[0]==e[0]&&t[1]==e[1])return;let a=f(4);h(t,e,a,!f(3)),d(t,e)};for(let t=0;t<i;t++)d(l[0][f(l[0].length)],[f(a),f(o)]);c(2,1,t>8?"$":">",t>8?"#39f":"#000");for(let l=0;l<n;l++)c(1,1,"m","#f00");for(let l=0;l<u;l++)c(4,1,"¢","#f90")},i=()=>{a.width=a.width;let f=e[3]<1,h=t>9;for(let t=0;t<25;t++)for(let a=0;a<25;a++){let o=e[0]-12+a,i=e[1]-12+t,n=r(l[1],[o,i])||r(l[0],[o,i]);n||(l[1].push([o,i,5,1,"#","#aaa"]),n=r(l[1],[o,i])),a>=8&&t>=8&&a<17&&t<17&&(n[6]=1),c.fillStyle=f||h?"#39f":n[5],c.fillText(h?"$":f?0:n[6]?n[4]:" ",10*a,10*t)}f||h||(c.fillStyle="#000",c.fillText(1+t+" ¢"+e[3],0,250))},n=(a,i)=>{let n=[a[0],a[1]];h(e,n,i,1==a[2]&&f(5));let u=r(l[1],n);u&&1==a[2]&&0==u[2]&&f(2)?u[3]--:u&&0==a[2]&&1==u[2]&&f(2)?u[3]--:u&&0==a[2]&&2==u[2]?(t++,o()):u&&4==u[2]?(a[3]++,u[3]--):r(l[0],n)&&!u&&(a[0]=n[0],a[1]=n[1])};b.onkeydown=(t=>{n(e,t.which-37);for(let t=0;t<l[1].length;t++)l[1][t][3]&&1==l[1][t][2]&&n(l[1][t],f(4));i()}),o(),i()';
eval(_);

The bug still shows up.

[nrkn]

Wow. You redid Nethack in 1k ?

Well, I tried :) It's still several features short of being a proper roguelike, but 1k is a pretty tough limit. I couldn't have done it without RegPack!

Once triggered, it seems like the display is out of sync, the upper level still shows up but collision detection uses data from the lower level. Is that also what you get ?

Yeah which is weird because the lower level data should have been overwritten, and the collision function is used to decide what to draw as well as to collide movement, so it's really, really weird for one to work and not the other

The bug still shows up.

I think this indicates that it's not a RegPack issue then!

Interesting that an IIFE fixes it, maybe indicates something strange about how Chrome treats top level scoping in eval?