# Help me sort out my desktop windows!

It's really frustrating when I'm working and I have a ton of windows open that are completely or partially hidden behind other windows. To make things worse, I'm still using two 1024x768 monitors.

Your task is to take a list of coordinates that designate the area, in pixels, on a screen, that a list of windows occupy and output a truthy value if, for a 2 monitor set-up with 1024x768 resolution, all the following conditions are met:

• None of the windows would overlap,
• None of the windows would be partly off screen
• None of the windows would straddle a boundary between the two monitors (so the order in which you visualize monitors 1 and 2 is irrelevant)

And a falsey value if any of these criteria are not met.

You may assume that all windows are rectangular, and that no coordinates are negative in the input, which is in a file, in stdin, or in a list of tuples, is in the following format:

monitor-number left-x bottom-y right-x top-y
monitor-number left-x bottom-y right-x top-y
monitor-number left-x bottom-y right-x top-y


Where each line contains the ID of the monitor (which can either be in [1, 2] or [0, 1] depending on which is easier for your language), and the dimensions of a window.

For example, the following input will result in a truthy output because none of the windows overlap:

1 10 10 100 200
1 210 300 600 600
2 10 10 100 200


The following input will result in a falsey value because lines 1 and 3 result in an overlap:

1 10 10 50 50
2 5 5 600 200
1 20 20 80 80


The following input will result in a falsey value because it would partially off-screen on one of the monitors:

1 1000 400 1400 300


This is so the shortest solution wins.

• Can I also take input as a list of tuples which is passed to my function? Jun 18, 2017 at 20:39
• @ThreeFx sure, let me update my post. Jun 18, 2017 at 20:39
• I suppose we are assuming that the monitor with minimum ID has its right edge aligned exactly with the left edge of the other? (My 1 and 2 are in fact the other way around :)) Jun 18, 2017 at 21:38
• @JonathanAllan Sure, but either way, the criteria for truthy output is that it can't partially be on either monitor Jun 18, 2017 at 21:39
• wait - if they straddle that should yield a falsey value, even if they are not off screen? Jun 18, 2017 at 21:41

# Jelly, 33 bytes

<⁽¡ð,769FẠ
Ḋs2Zr/€p/;€Ḣµ€ẎµQL⁼LȧÇ


A monadic link taking a list of lists of the window-data and returning 0 if any windows exceed their monitor's space or overlap, or 1 otherwise.

Uses 1-indexed monitor IDs; 1,1 to 1024,768 inclusive are considered on-screen.

Try it online! (Too slow for the first two test cases! Typical golf :))

### How?

The reason this is so slow is because it creates all the pixels for all the windows, each with its monitor ID and checks for overlap using de-duplication (which is implemented using lists not sets). It also checks each pixel is on-screen rather than checking just the input corners. All in all very inefficient...

Test-case 1 succeeded (1) locally in only eleven minutes! Test-case 2 succeeded (0) in less.

<⁽¡ð,769FẠ - Link 1 check pixels are on-screen: list of lists, each being [x,y,ID]
⁽¡ð       - literal base-250 integer = 1025
,769   - pair with literal 769
<          - less than (vectorises) -> [[x<1025, y<769, ID],...]
F  - flatten into a single list
Ạ - all truthy?

Ḋs2Zr/€p/;€Ḣµ€ẎµQL⁼LȧÇ - Main link: list of lists, each being [ID,Lx,By,Rx,Ty]
µ€         - for €ach window:
Ḋ                      -   dequeue -> [Lx,By,Rx,Ty]
s2                    -   split into twos -> [[Lx,By],[Rx,Ty]]
Z                   -   transpose -> [[Lx,Rx],[By,Ty]]
/€                -   reduce €ach with:
r                  -     inclusive range -> [[Lx,...,Rx],[By,...,Ty]]
/              -   reduce with:
p               -     Cartesian product -> all pixels, [[Lx,By],[Lx,By+1],...[Lx,Ty],[Lx+1,By],... ...[Rx,Ty]]
;€            -   concatenate each pixel with the monitor ID
Ẏ        - tighten -> one list of all window's [x,y,ID] values
µ       - monadic chain separation, call that w
Q      - deduplicate (unique [x,y,ID] values)
L     - length
L   - length of w
⁼    - equal?
ȧ  - logical and

• That is... equal parts horrific and amazing. I love the solution of generating each pixel and testing, but it also makes me cringe inwardly. Jun 18, 2017 at 23:13
• Yep, welcome to the golf course :p Jun 18, 2017 at 23:13

f l=[]==[a|z@(m,a,b,c,d)<-l,y@(n,e,f,g,h)<-l,y/=z&&m==n&&e<c&&(b<h||f<d)||1024<c||768<d]


Forgot the bounds checking initially.

Takes input as a list of 5-tuples representing the windows.

### How does it work?

We pair all windows with each other and then look at all pairs (a, b) with a /= b. Since the relation is symmetric, we only have to check whether the right end of the first and left end of the second window overlap. If they do, then for them to overlap one's lower has to be higher than the other one's top or vice versa. Lastly, we do bounds checking.

If any of the conditions are true, the list will contain more than one element, so all there is left to do is to check whether the list is empty.

• You use the variable name d twice, so it fails for example for f [(2,10,10,200,2000)]. Just rename the variables inside the 5-tuple y and adjust the conditions.
– nimi
Jun 18, 2017 at 22:47
• @nimi Whoops, how did I not see that? Thanks, it should be fixed now! Jun 18, 2017 at 22:49
• Hmm, the first test case returns False: f [(1,10,10,100,200),(1,210,300,600,600),(2,10,10,100,200)].
– nimi
Jun 18, 2017 at 22:55
• @nimi I'll fix that in the morning then, it's late here. If you want to correct it, feel free to! Jun 18, 2017 at 23:02
• null is better style than []== and just as long (or short) Jun 19, 2017 at 3:19

## JavaScript (ES6), 100 bytes

a=>!a.some(([m,l,b,r,t],i)=>l<0|b<0|r>1024|t>768|a.some(([n,k,c,s,u],j)=>i!=j&m==n&l<s&b<u&r>k&t>c))


Takes input as an array of arrays. Assumes window coordinates are monitor-relative. I could save 8 bytes by assuming windows never overlap to the left of the monitor but 100 is a nice round number ;-)