Its an artificial restriction. Method arguments are just like temps, i.e., the interpreter does not distinguish between the two; its only the compiler that restricts assignments to arguments (I'm not 100% sure, but I assume that this still holds with Eliot's new closure compiler).
I think it would be nice to be able to assign to arguments. In some cases they make for concise and better readable code (I got used to this for instance in Python and JavaScript). For example, consider a method that uses an argument in multiple places and you figure that you need to handle the case in which the argument is nil. You just add a line like: arg := arg ifNil: [ 0 ]. Without argument assignment you would need to create a temp var with a different name and change all usage of the old arg variable. Block args had been writable until the new closure compiler was introduced, which by default disallows arg assignments. I don't remember what the rationale for this change was (?). On the flip side, one has to consider code portability. For projects like Seaside this probably is a no go. But for projects that don't need to be ported I think it would be a worthwhile change. Projects that need to avoid portability troubles with other dialects should use Lint rules (stuff like this is better checked by Lint than the compiler...). I'll also note that this topic was discussed before and some people argued that argument assignments can lead to bad code quality... Cheers, Adrian On Nov 28, 2010, at 11:41 , Max Leske wrote: > +1 > > > On 28.11.2010, at 11:26, laurent laffont wrote: > >> I would like to know whether it's a design choice or a technical obstacle. >> >> I've seen: >> >> Parser>>assignment: varNode >> " var ':=' expression => AssignmentNode." >> | loc start | >> (loc := varNode assignmentCheck: encoder at: prevMark + >> requestorOffset) >= 0 >> ifTrue: [^self notify: 'Cannot store into' at: loc]. >> ..... >> >> >> TempVariableNode>>assignmentCheck: encoder at: location >> ^((self isBlockArg and: [Parser allowBlockArgumentAssignment not]) >> or: [self isMethodArg]) >> "<-- not allowed !" >> ifTrue: [location] >> ifFalse: [-1] >> >> But no explanation :( >> >> >> Laurent >> >> >> >> On Sun, Nov 28, 2010 at 10:59 AM, Max Leske <[email protected]> wrote: >> Yeah, that's true, I've noticed that too. I guess it's on purpose but I have >> no idea. AFAIK you can't assign a new value to a parameter anywhere in the >> method not only when using blocks. Maybe it's not possible because there is >> not really a 'variable' but only a reference. So assigning a new value to >> the parameter would effectively override the reference. But don't trust me >> on this.... >> >> Cheers, >> Max >> >> >> On 28.11.2010, at 10:54, laurent laffont wrote: >> >>> Sorry >>> >>> foo: n >>> ^[:i| n := n+i. ] >>> >>> vs >>> >>> foo: n >>> |s| >>> s := n. >>> ^[:i| s := s+i. ] >>> >>> >>> Laurent >>> >>> On Sun, Nov 28, 2010 at 10:51 AM, Max Leske <[email protected]> wrote: >>> In the first example you didn't declare 's': >>> >>> foo:n >>> ^ [ :i || s | s:= s+i ] >>> >>> Or did I misunderstand the question? >>> >>> Cheers, >>> Max >>> >>> >>> On 28.11.2010, at 10:30, laurent laffont wrote: >>> >>>> Hi, >>>> >>>> I can't write >>>> >>>> foo: n >>>> ^[:i| s := s+i. ] >>>> >>>> but >>>> >>>> foo: n >>>> |s| >>>> s := n. >>>> ^[:i| s := s+i. ] >>>> >>>> >>>> Why ? >>>> >>>> >>>> Cheers, >>>> >>>> Laurent Laffont >>>> >>>> Pharo Smalltalk Screencasts: http://www.pharocasts.com/ >>>> Blog: http://magaloma.blogspot.com/ >>> >>> >> >> >
