From: Brian Candler <b.candler@...>
Date: 2010-11-11T23:49:59+09:00
Subject: Re: Analyzer for errors in code ?

David Unric wrote in post #960444:
> My appologies I did not realized sooner the deeper conceptual
> differencies between these languages.
> Despite of that, if Ruby cann't decide if mymsg is a variable or a
> method call in advance, how does it rule out the possibility for a
> static parser to notify the 'mymsg' symbol has no assignment or method
> definiton in the source code ?

On the contrary: ruby *does* always decide (statically, at parse time, 
not at run-time) whether a bareword is a local variable or a method 
call, based on whether a previous assignment statement exists previously 
in the same scope.

e.g.

def foo
  if false
    x = 123
  end
  puts x           # prints nil - x is a local variable
  puts y           # NameError
  puts z()         # NoMethodError
end
foo

Although the assignment to x was never executed, it exists in the parse 
tree before the point where it is referenced. Therefore at that point 
bareword x is known to be a local variable. At runtime the value is nil 
because no assignment was actually executed.

y is decided statically to be a method call, as the parsetree shows, 
because no assignment was seen. Note however that the error message says 
"undefined local variable or method `y'". The programmer might have 
meant to assign to a local variable called y, or might have meant to 
define a method called y. Neither was done, but Ruby has no way to tell 
which you intended.

z() is known from the syntax definitely to be a method call, and so the 
error is "undefined method `z'". The same would be true for self.z. It 
doesn't matter whether a statement z=... exists earlier.

Unfortunately, because Ruby doesn't require you to annotate your code, 
there is no additional information which could be used to highlight an 
anomaly.

For example, in perl, scalar variables are flagged by $. If you do

   $myarg = 123;
   ...
   print $my_arg;

it can see that $my_arg is used only once, and is therefore likely an 
error. With 'use strict' you are forced to declare your variables, so 
every variable must have a matching declaration.

The non-verbose nature of Ruby, combined with the dynamic definition of 
methods at runtime means that these tests simply can't be done. Dynamic 
method definitions are very widely used - look at code which uses 
ActiveRecord for instance. The methods in an object are based on the 
columns in the database which the program connects to.

And what this means is, you need tests. Your test suite needs to execute 
each line of code in your source at least once. A code coverage tool 
like 'rcov' can help you achieve that, or if you are paranoid look at 
'heckle'.

Is this harder or more expensive than static code checking? Only if you 
take the point of view that you don't need unit tests.

-- 
Posted via http://www.ruby-forum.com/.