Suggestion for indexing strings

Posted by Nick Gammon on Fri 26 Nov 2010 07:56 PM — 9 posts, 32,138 views.

Nick Gammon Australia Forum Administrator Fri 26 Nov 2010 07:56 PM #0

I saw on the Lua mailing list a cool suggestion for string indexing.

The gist is to make a simpler syntax for pulling out an individual character from a string. For example, if you want to find the first character you currently do:


a = "nick"
print (string.sub (a, 1, 1)) --> n

This is a little ugly. Now if you add the following code to the initialization of your Lua code:


getmetatable ("").__index = function (str, i)

  if (type (i) == "number") then
    return string.sub (str, i, i)  -- index into str
  end -- if

  return string [i]  -- fallback (eg. string.match)
end -- function

What that does is add a metamethod for handling the indexing into any string (since all Lua strings share the same metatable).

Now you can do this:


a = "nick"
print (a [1])  --> n

[EDIT]

Modified per discussion below, so the discussion may not make a heap of sense.

Twisol USA Fri 26 Nov 2010 08:18 PM #1

You might want to use string.sub(str) instead of str:sub(), since you're incurring needless overhead by invoking the metamethod again.

Improvement allowing for a range:

getmetatable ("").__index = function (str, i)
  if (type (i) == "number") then
    return string.sub (str, i, i)  -- index into string
  elseif (type (i) == "table") then
    return string.sub (str, i[1], i[2])
  else
    return string [i]  -- fallback
  end  -- if
end -- function

[EDIT]: It's a pity you can't make the same syntax work for modifying strings.

Nick Gammon Australia Forum Administrator Fri 26 Nov 2010 08:38 PM #2

Twisol said:

You might want to use string.sub(str) instead of str:sub(), since you're incurring needless overhead by invoking the metamethod again.

I had that initially and changed it. I was trying to avoid the overhead of:

Looking up "string" in the global environment
Looking up "sub" in that table

Whereas doing str:sub hopefully only is one lookup (the metatable). Am I wrong?

As for the table, I think that might be getting as complicated as the original.


print ( a [ { 2, 3 } ] )

instead of:


print ( a:sub (2, 3) )

Twisol USA Fri 26 Nov 2010 08:48 PM #3

Nick Gammon said:
I had that initially and changed it. I was trying to avoid the overhead of:

*Looking up "string" in the global environment
*Looking up "sub" in that table

Whereas doing str:sub hopefully only is one lookup (the metatable). Am I wrong?

Well, it's one lookup plus a function call (the metamethod), which probably involves more Lua bytecode than just a lookup. And you can cache 'string' in a local. In fact, you can cache 'sub' in a local, too:

local string = string
local sub = string.sub
getmetatable ("").__index = function (str, i)
  if (type (i) == "number") then
    return sub (str, i, i)  -- index into string
  else
    return string [i]  -- fallback
  end  -- if
end -- function

Nick Gammon said:
As for the table, I think that might be getting as complicated as the original.

Touche. I was trying to go for the Ruby syntax of str[1,2], but you can't pass multiple values to the index metamethod.

Nick Gammon Australia Forum Administrator Fri 26 Nov 2010 09:22 PM #4

It seems you may have a point. I think you added in caching string.sub while I was replying, but the difference was minor as it turned out.

Test bed:


collectgarbage ( )

local a, b = "nick"
start = utils.timer ()

for i = 1, 1000000 do
  b = a [3]
end -- for

print (string.format ("Time taken = %0.2f", utils.timer () - start))

Results in speed order:


 return sub (str, i, i)         --> 0.42 (string.sub cached) 
 return string.sub (str, i, i)  --> 0.43 (string cached)   
 return string.sub (str, i, i)  --> 0.47 secs
 return str:sub (i, i)          --> 0.72 secs

So my original was about 1.5 as fast than my improved version. Caching the string.sub lookup helped, but not by much.

I'll change the original post back to the simpler version.

Nick Gammon Australia Forum Administrator Fri 26 Nov 2010 09:26 PM #5

Meanwhile I'm not sure about the fallback line. Surely if you do, say:


a = "abcd"
print (a ["x"])

... you are indexing into the key "x" in string "a"? (Not key "x" in the string table). However somewhat surprisingly it seems to work in the sense that it returns nil.

However wouldn't it always?

So perhaps the fallback should be either:



  return nil  -- fallback

-- or

  error ("Attempting non-numeric index into string")

Nick Gammon Australia Forum Administrator Fri 26 Nov 2010 09:29 PM #6

Or even:


  return ""   -- fallback

... on the basis that indexing out of range returns an empty string.

Twisol USA Fri 26 Nov 2010 09:36 PM #7

You need to leave it as string[i]. Otherwise you prevent users from doing str:gsub() or anything else in 'string' using the shorthand.

Nick Gammon Australia Forum Administrator Fri 26 Nov 2010 09:50 PM #8

Aha. That makes sense.