How fast, on your machine
Finding a place with a Grid Point Code is arithmetic, not a request. This page runs that arithmetic on your own hardware, in your own browser, using the published package every other page here already loaded.
Nothing is sent anywhere and nothing is fetched. The measurement is of this machine at this moment, so it is not a number to compare against somebody else's. It is a number you watched happen.
Measuring needs JavaScript, which is the one thing on this site that genuinely does. Everything else here works without it.
| Operation | What it does | Rate | Each | Every batch |
|---|---|---|---|---|
encode | A latitude and longitude to ten characters | — | — | |
decode | Ten characters back to a latitude and longitude | — | — | |
isValid | Whether a string is a code that decodes | — | — | |
neighbours | The eight cells surrounding one | — | — |
Four ways a benchmark lies, and what is done about each
- The work gets deleted
- An engine that can see a result is never used may remove the call that produced it, and then the page is timing an empty loop. Every operation returns a number that is summed, and the total is printed above, a figure that could only exist if the work was done.
- The clock is too blunt
performance.now()is deliberately coarsened against timing attacks, in some browsers to tens of microseconds. Timing one operation would measure the clock, so batches are doubled until each lasts at least 50 ms and the resolution stops mattering.- The first run is not the steady state
- Code starts interpreted and is compiled once it looks worth compiling, so early iterations can be an order of magnitude slower than settled ones. Three batches are run and thrown away before any are kept.
- One number hides the spread
- A busy tab, a thermal limit or a collection pause widens the distribution, and a single figure drawn from it is a guess. Fifteen batches are timed, the median is reported rather than the best, and every batch is drawn in the last column so you can see what the median came from.
The points are the same on every machine: a thousand of them, from a seeded generator, so what differs between two runs is the hardware and not the work.