In the early 1970s, I had university a co-op job at a place where one of the main languages used was like a glorified calculator. It provided two stacks, and all variables and operators were a single character. Depending upon context each letter could be overloaded and interpreted as a scalar (x+2), a vector (x@2), a function (x:2), or whatever.
(I'm guessing about the syntax; it could have been quite different.)

I'm pretty sure it was either in-house, or provided by some external company that wrote and maintained it for them.

The interesting thing for me is that all arrays were powers of 2 in length. I no longer remember the declaration syntax, but I imagine that each array was stored with an appropriate bit mask, so that x@y was a reference to address x+(y&mask).

All arrays were thus circular. This meant that there was no possibility of having indexes out of range.
It also meant that x@-2 would naturally be the second from the end of the array, without requiring any different processing than for x@2.
(Remember, this was at a time when an extra instruction or two was considered expensive.)

There were times when this behaviour came in handy (though I really can't remember any specific examples).

I'm wondering if any "real" languages ever provided arrays like this.

  • 2
    $\begingroup$ While it's not an answer to the question as described, a related fun fact is that the growth factor in many languages' implementations of dynamic arrays is 2, so if the initial size of the array is a power of two, it will always have a power of two capacity. $\endgroup$ Jun 30, 2023 at 1:21
  • 2
    $\begingroup$ @RydwolfPrograms, I remember implementations of malloc() , perhaps the original BSD, where the allocated blocks were a power of 2, but the first 4 bytes were used to store the size of the block for use by free() . This meant that the common practice of buffer=malloc(1024) actually used 2048 bytes. 1024 was an obvious choice, but in fact it was the worst possible choice. $\endgroup$ Jun 30, 2023 at 2:25
  • $\begingroup$ @RydwolfPrograms, I should add that the documentation for standard stdio functions also encouraged this use of inefficient buffer size. e.g. setvbuf(…, malloc(BUFSIZ), …, BUFSIZ), where BUFSIZ is almost always defined as a power of 2. $\endgroup$ Jun 30, 2023 at 2:38
  • 2
    $\begingroup$ Amusingly, my archives record that I posted this to a newsgroup in 1984: "… any program that uses stdio can save as much as 50% in buffer space if we just redefined BUFSIZ as 1020 instead of 1024.". In a style typical of the flame-wars of the time, my blasphemy was kindly and logically refuted: "I don't get it. You're obviously just being a f___ing a__h____.". $\endgroup$ Jun 30, 2023 at 18:44
  • 1
    $\begingroup$ Historically, gpus frequently only supported textures with power-of-two sizes, though nowadays they are not so restricted. Behaviour of out-of-bounds accesses is configurable; see this graphic. $\endgroup$
    – Moonchild
    Aug 15, 2023 at 22:52


You must log in to answer this question.

Browse other questions tagged .