strsub vs. table lookup

[c h i l l a]

strtable = {
["$MyINFO"] = 1,
["$To:"] = 1,
["$Kick"] = 1,
["$ConnectToMe"] = 1,
["$RevConnectToMe"] = 1,
["$ValidateNick"] = 1,
["$Search"] =1,
["Other"] = 1,
["Other2"] = 1,

}

string = 0
table = 0

function TestStrings(line)
	local _,_,str = strfind(line,"^(%S+)")
	--print(str)
	if str then
		t1 = clock()
		if strsub(str,1,12) == "$ConnectToMe" then
		elseif strsub(str, 1, 7) == "$Search" then
		elseif strsub(str,1,5) == "$Kick" then
		elseif strsub(str,1,13) == "$ValidateNick" then
		elseif strsub(str,1,7) == "$MyINFO" then
		elseif strsub(str,1,4) == "$To:" then
		end
		t2 = clock()
		if strtable[str] then
		end
		t3 = clock()
	end
	string = string + (t2-t1)
	--print("STRSUB:   "..(t2-t1))
	--print("TABLE:   "..(t3-t2))
	table = table + (t3-t2)
end


function CreateString()
	for i = 1,1000000 do
		local word = strchar(random(65,90))
		len = random(3,30)
		for i = 1,len do
			word = word..strchar(random(97,122))
		end
		TestStrings(word)
	end
	print("STRSUB:   "..string.."     "..(string/1000000))
	print("TABLE:   "..table.."     "..(table/1000000))
end

> CreateString()
STRSUB:   3.25800000000163     3.25800000000163e-006
TABLE:   0.5000000000027285     5.000000000027285e-007
> CreateString()
STRSUB:   6.742999999995845     6.742999999995846e-006
TABLE:   0.9350000000040382     9.350000000040382e-007


pro's and contras are welcome.

[NotRabidWombat]

Your test is VERY biased towards the hash table being victorious. You create a random dc command which will most likely always be worst case scenario (ie: does not exist in the set you are searching through). So you perform 7 strsubs every single time you test that method. You should extend the test to have a set of actual results (chat, search, invalid, etc) that you randomly select and test.
The next problem is you are not taking advantage of one of the most powerful functions, strfind. You should Use strfind to grab the first word for each test and then do comparisons and table indexing.
However, you will find the hashing is MUCH faster than a switch style if/elseif/else statement.

Hashing has also been improved (collision handling I believe) in Lua 5.0. *hint* *hint* *hint*

-NotRabidWombat

[c h i l l a]

but its worst case for both methods, but I will only compare the str  from strfind, for beeing right.

Done.  Now they are both almost equal.
So i guess.. either hashing in LUA 5.0 is way faster than in 4.0, or else it actually doesn't matter.

here again the code

so you can check ;)

strtable = {
["$MyINFO"] = 1,
["$To:"] = 1,
["$Kick"] = 1,
["$ConnectToMe"] = 1,
["$RevConnectToMe"] = 1,
["$ValidateNick"] = 1,
["$Search"] =1,
["Other"] = 1,
["Other2"] = 1,

}

string = 0
table = 0
ttotal = 0

function TestStrings(line)
	tbegin = clock()
	local _,_,str = strfind(line,"^(%S+)")
	--print(str)
	if str then
		t1 = clock()
		if str == "$ConnectToMe" then
		elseif str == "$Search" then
		elseif str == "$Kick" then
		elseif str == "$ValidateNick" then
		elseif str == "$MyINFO" then
		elseif str == "$To:" then
		end
		t2 = clock()
		if strtable[str] then
			--print("yes")
		end
		t3 = clock()
	end
	string = string + (t2-t1)
	--print("STRSUB:   "..(t2-t1))
	--print("TABLE:   "..(t3-t2))
	table = table + (t3-t2)
	tend = clock()
	ttotal = ttotal + (tend-tbegin)
end

function CreateString()
	var = 0
	for _ = 1,10000 do
		for i1 = 1,10 do
			var = var + 1
			if i1 == 10 then
				TestStrings("$ConnectToMe")
			elseif i1 == 5 then
				TestStrings("$Search")
			else
				local word = strchar(random(65,90))
				len = random(3,30)
				for _ = 1,len do
					word = word..strchar(random(97,122))
				end
				TestStrings(word)
			end
		end
	end
	print("STRSUB:   "..string.."     "..(string/1000000))
	print("TABLE:   "..table.."     "..(table/1000000))
	print("TOTAL: "..ttotal)
	print(var)
	collectgarbage()
end

resuts:

> CreateString()
STRSUB:   0.3539999999999282     3.539999999999281e-007
TABLE:   0.2499999999996589     2.49999999999659e-007
TOTAL: 2.256999999999835
100000
> CreateString()
STRSUB:   0.3849999999999909     3.849999999999909e-007
TABLE:   0.280999999999608     2.80999999999608e-007
TOTAL: 2.583999999999946
100000

[plop]

chilla have you tryed it with 50+ elseif's???
here table indexing should show much more speed.
btw if your sneaky you can use str=gsub(line, 1, 3), if you look you only need the 1st 3 characters.

wombat if hashing has been improved on 5.0 i vote for your hint.
i love those thingys, i try 2 use them as much as posible on a.i..

plop

[c h i l l a]

table is getting faster. yepp.
but why strsub, when you can compare the first word,
maybe someone wants to do a test on strsub and strfind.

> CreateString()
STRSUB:   1.472999999994499     1.472999999994499e-006
TABLE:   0.2029999999995198     2.029999999995198e-007
TOTAL: 3.174999999995634
100000
> CreateString()
STRSUB:   1.658999999996013     1.658999999996013e-006
TABLE:   0.2179999999989377     2.179999999989377e-007
TOTAL: 3.671999999995023
100000

else I will stick to strfind now.

[plop]

strfind checks every character, so $RevConnectToMe whill take 16 checks and $To: only 5.
if you grab only the 1st 3 characters all whill use the same time.

plop

[c h i l l a]

sounds logical ;).
Will improve LIS with this stuff.

[NotRabidWombat]

Yes, but most scripts should use strfind to get the DC command or the command from the chat. You would need the strfind for either method, so it should be outside of the time test.

The problem with using worst case for both is hashing will always beat a switch at worst case (if it deals with collisions well).

-NotRabidWombat

[c h i l l a]

its outside the strfind..

but truth is, a good strsub, can beat the strfind. almost, half of the time can be saved as shown...

but...  I know that Ptokax, has optimised its strfind, through strsub...  in c++ way of course...

but its not as secura as I found out.. because you can

connect to ptokax with  $MyInfo   but dc protocoll  is case sensitiva so you should get dissed, well you get by LIS, but only cause it checks, for the whole MyINFO.

But to do Cases, you can improve your script by using strsub, to filterout, only the data you need or not need.

test and see for yourself what suits best, it always depends on what you need.

[plop]

Originally posted by NotRabidWombat
Yes, but most scripts should use strfind to get the DC command or the command from the chat. You would need the strfind for either method, so it should be outside of the time test.

The problem with using worst case for both is hashing will always beat a switch at worst case (if it deals with collisions well).

-NotRabidWombat

yep, would also be very boring if every1 would have 2 use the same length for a nick.
and all the commands the same length can result in some weird abreviation's. lol

if a string is always on the same place then gsub is the best 2 use.
but then again cmd = gsub(data, (strlen(user.sName)+1), -1) also works for cmd's without arguments. lol
but i think this is slower then strfind.

plop