Category: C
Jan 5, 2017

Normally in web application development, clicks are a matter of listeners. Either a click hit a thing or it didn’t. When using the HTML5 canvas, this isn’t the case. Images and features are rendered onto it, but you can’t attach a listener to them. Instead, you monitor where these elements visually reside versus where clicks land.

If an object is rectangular and will never rotate, the solution is easy. If click X is between object start X and object end X, and click Y likewise, then it hit. If not, it didn’t.

In this post we’ll construct a simple series of functions for determining the same thing if the rectangle has rotated, then look at how the idea can apply to more complex shapes. Nothing new or groundbreaking, and nothing language-specific — consider this a StackOverflow answer.

Visualization

For a developer who hasn’t mathed in a while, this can seem tricky at first. Once you see a few images it’s perfectly sensible.

We’ll be drawing four triangles. Each one consists of the click point and two vertices of the object. We then compare the area of the object to the combined area of the triangles. If they’re the same, the click had to be in the object. If the triangles are larger, the click couldn’t have been in the object.

Rotated Rectangle - Click hit

If the click was inside, the areas are equal.

Rotated Rectangle - Click miss

If the click was outside, the combined area of the triangles is bigger than the area of the rectangle.

How much it’s rotated doesn’t matter. Where it’s at doesn’t matter. That’s literally all there is to this.

First things first

To figure out if the click hit, we’ll need four pieces of data:

  • The coordinates of the click
  • The position of the rectangle
  • The size of the rectangle
  • The angle of rotation of the rectangle

Those last three pieces we’re just putting together so we can find the vertices. Once we know the click position and the vertex positions, figuring out if the click hit is easy.

Finding the vertices

Since you’re probably not tracking the individual locations of the vertices and are instead tracking the position of the object, you’ll want to start with a function that can find these by the object’s size and position. How this will be done is hugely variable depending on where you’re measuring from (tracking left top? center?) and if/how the object has been scaled.

If we pretend like there’s no scaling or rotation and we’re measuring from left and top, the function may initially look like this:

Since this doesn’t take into account rotation yet, we’ll need a utility function to find a point’s new location after it’s been rotated by X degrees around another point:

Modifying our previous findRectVertices  function to also use rotation may then look like this, assuming we’re rotating objects around their center:

Calculating the areas

Now that we have all the necessary points, all that’s left are three things:

  • Calculate the area of the rectangle
  • Calculate the area of the triangles
  • Compare the two

If you recall high school geometry, point two is the loaded one there. To find the area of an unknown triangle we’d use Heron’s formula, but first the distances of each side will be needed. The two utility functions may look like this:

With those functions ready, we can now write a complete function that performs the three bullet points and returns a simple True or False:

Common issues

By far the most frequent annoyance with this approach when working in the DOM is going to be getting good position numbers. So many variables can throw a wrench in how you’re measuring where the object is. If your reported object position doesn’t match up with the reported mouse or touch position, all of the above is useless.

Look for situations in which clicking near something instead of on it reports a hit, then try to identify elements that match the size of the inaccuracy. Everything from margins on the HTML or body elements, borders on divs, document scroll position, and box-sizing settings can play havoc on your ability to get clean measurements that match up with what’s plainly visible.

Going farther

This method isn’t just good for rectangles. It would be straightforward to apply to triangles, too, or for any other convex polygon. Consider how you might implement the ability to check if a click landed inside a hexagon. As a brain teaser, what would happen on a concave polygon?

Jul 20, 2016

While working on a new revision of a project, I moved from an Arduino Uno to an Arduino Mega. In theory these devices should handle SPI in the same manner, just with different pins. While MOSI, MISO, and SCK are on pins 11, 12, and 13 respectively on the Uno, those change to 51, 50, and 52 on the Mega. I initially daisy chained two LTC2400 analog-to-digital modules onto the SPI bus, just like I had them on the Uno, but immediately ran into issues.

Using SPI on the Arduino Mega

Interestingly, SPI does seem to work fine on the Mega in test cases. If you attach the bus to pins 51, 50, and 52, then run a chip select (also called slave select, CS, or SS) line to the provided pin 53, this seems to function with no issues. Pull the CS low, then read from MISO, no problems at all. It turns out that CS on 53 is actually more important than you might think when doing daisy chaining, but more on that in a moment.

PORTB and Chip Selects

Many SPI devices will use cbi() and sbi() functions to clear and set bits directly on port registers of the Arduino, which has a number of advantages like resulting in smaller code, much faster switching, and the ability to switch multiple pins at the same time. The pins on the Arduino are divided into ports logically, such as PORTA, PORTB, etc., which can be seen nicely in this diagram. If you were to want to rapidly switch pin 0 on PORTB on, you would initially set its direction bit high to make it an output, then set its port bit high to bring its voltage high.

It follows then that I/O standards on the Arduino are usually grouped into a single port, which is the case on both the Uno and the Mega. On the Mega, MOSI (pin 51) is PORTB 2, MISO (pin 50) is PORTB 3, SCK (pin 52) is PORTB 1, and SS (pin 53) is PORTB 0. This keeps everything cleanly on the same register, but with SPI it may not always work out that way. If you’re daisy chaining SPI devices, there’s only one pin 53. You’ll need to use an additional pin as an extra SS line, and that pin may not be on the same port.

Where is PORTL?

As Hao Jiang found in this post about Arduino Mega port mapping, not all ports on the Mega even seem to respond to changes in their respective registers, so unless you’re okay with using pinMode() and digitalWrite() in your SPI code, the SS options are preemptively limited. Just get ready for this to begin with–where you want that SS line may not be where you get that SS line. But that’s not the most vexing issue you’d have to debug.

The Tyranny of Pin 53

Put simply, you can’t use it if you ever want more than one SPI device. I don’t recall ever having this issue on the Uno, but on the Mega, MISO only seems to become available if pin 53 (the SS it knows) transitions to low. This leads to a few problems:

  • If one device is on 53 and another is on 49 (or any other number), you can contact the one on 53, but will receive nothing on MISO when transitioning 49 to low.
  • If you transition 53 to low at the same time, you’ll get traffic, but it will just be a collision.
  • If neither device is on 53, then neither of their SS lines when transitioned will initiate anything over MISO.

The Solution to Daisy Chaining

My theory is that in the Mega, MISO (and maybe MOSI–I wasn’t using a device that needed that line) is inextricably linked somehow with SS on pin 53, and traffic can only be received on MISO if SS also transitions to low. The solution, then, is to use different pins as SS lines (on a port register that actually works) for each daisy-chained device and when flipping their SS low, also flip 53 low even though nothing is connected to it. When flipping any SS to high, also flip 53 to high. This seemed to be a foolproof way to get data from each device.

Sep 10, 2015

I was recently commissioned to build a machine that would perform a very interesting and unusual task: Count how many times a wire had been wound around a ring by measuring its electrical properties. This task was a relatively involved one, as it included a ground-up design of the circuitry, making choices about component prices […]

Continue reading...
Fork me on GitHub