Ive noticed that functions calls like f(argumentCollection=arguments)
use a lot more cpu time then when the arguments are mentioned
explicitly. Its in the range of some 10's of ms per page call,
depending on how much functions with this construct are called.
The one from coldbox that gives me the most lag is
frameworksypertype.getPlugin.
The arguments could easily be added explicitly to the function as they
all have default values (so not if elseif etc. needed).
That test is not comparing fairly. The first example is not creating /
populating the struct inside the loop.
I tweaked the test a bit to make it a fairer comparison and wrapped it
in cfsilent so you're not timing whitespace generation(!) etc (see
below) and consistently saw:
Railo: argumentCollection passing is 10-20% slower. CF9:
argumentCollection passing is 30-50% slower.
Updated test:
<cfsilent>
<cfset n = 100000 />
<cffunction name="returnValue" access="public" returntype="String"
output="false">
<---************************************************************** --->
<cfargument name="sString" type="string" required="true" />
<cfargument name="iInteger" type="numeric" required="true" />
<cfargument name="stStruct" type="struct" required="true" />
<---************************************************************** --->
<--- Local call to get the user object from the session --->
<cfset var _l = structNew() />
<cfset var result = arguments.sString & ': [' & arguments.iInteger &
']<br />' />
<cfreturn result />
</cffunction>
<cfset s = getTickCount() />
<cfloop from="1" to="#n#" index="i">
<cfset stStruct = structNew() />
<cfset stStruct['sString'] = 'This is String No' />
<cfset stStruct['iInteger'] = i />
<cfset stStruct['stStruct'] = StructNew() />
<cfset x = returnValue(argumentCollection=stStruct) />
</cfloop>
</cfsilent>
<cfoutput>
argumentCollection took #getTickCount()-s#ms.
</cfoutput>
<cfsilent>
<cfset s = getTickCount() />
<cfloop from="1" to="#n#" index="i">
<cfset stStruct = structNew() />
<cfset stStruct['sString'] = 'This is String No' />
<cfset stStruct['iInteger'] = i />
<cfset stStruct['stStruct'] = StructNew() />
<cfset x = returnValue(stStruct.sString,stStruct.iInteger,stStruct.stStruct) />
</cfloop>
</cfsilent>
<cfoutput>
pass args ##1 took #getTickCount()-s#ms.
</cfoutput>
<cfsilent>
<cfset s = getTickCount() />
<cfloop from="1" to="#n#" index="i">
<cfset stStruct = structNew() />
<cfset stStruct['sString'] = 'This is String No' />
<cfset stStruct['iInteger'] = i />
<cfset stStruct['stStruct'] = StructNew() />
<cfset x = returnValue('This is String No',i,structNew()) />
</cfloop>
</cfsilent>
<cfoutput>
pass args ##2 took #getTickCount()-s#ms.
</cfoutput>
I forgot to mention that i found this by using Oracle JRockit mission
control. I did a JRA recording for 1 minute while refreshing a page of
the site im developing. Then i had a look at hot methods, where i
found that getPlugin is spending a relative large percentage on the
argumentCollection thing.
Btw, luis, Oracle Jrockit mission control would be a great tool to
have a look at where you can improve the performance of coldbox.
I had a page on my site that first took 3 seconds, using mission
control i made improvements to make that 80ms per page call.
Just be aware that you can't write them out explicitly if they're
optional - as in Andrew Scott's thread about renderView() - when
dealing with optional arguments, you pretty much have to use
argumentCollection to pass them through...
I just did some tests here locally, as I wanted to look directly at the most common use case, which is directly passing ‘arguments’ to another CFC as as argumentCollection.