Lexical scoping is not what I expect

-----Original Message-----
From: Duncan Murdoch [mailto:murdoch.duncan at gmail.com]
Sent: Friday, June 28, 2013 11:07 AM
To: John Fox
Cc: 'S Ellison'; r-help at r-project.org
Subject: Re: [R] Lexical scoping is not what I expect

On 28/06/2013 10:54 AM, John Fox wrote:

Dear Duncan and Steve,

Since Steve's example raises it, I've never understood why it's legal

to

change the built-in global "constants" in R, including T and F. That

just

seems to me to set a trap for users. Why not treat these as reserved
symbols, like TRUE, Inf, etc.?

If we weren't allowed to change T and F, would we be allowed to change
other "constants", like functions mean or max? If we couldn't change
anything defined in the base package, would we be allowed to change
things defined in other packages like stats or utils or graphics?  I
think it's simply a matter of setting the line somewhere, and R has
chosen to set it in the most permissive place that's reasonable.  It
assumes that its users know what they are doing.

Why not allow changes to TRUE or Inf?   TRUE is a constant, whereas T
is
a variable containing TRUE.  Inf is also a constant, corresponding to a
length one vector containing that value.  Those are treated by the
parser like 2 or "hello".  It would be really bad if someone could
change the meaning of 2 (though I hear some old Fortran compilers
allowed that), but is it really so bad to allow someone to define their
own plot function, or temperature variable named T?

Duncan Murdoch

Best,
  John

-----Original Message-----
From: r-help-bounces at r-project.org [mailto:r-help-bounces at r-
project.org] On Behalf Of Duncan Murdoch
Sent: Friday, June 28, 2013 9:40 AM
To: S Ellison
Cc: r-help at r-project.org
Subject: Re: [R] Lexical scoping is not what I expect

On 28/06/2013 9:28 AM, S Ellison wrote:

I too find R's lexical scoping rules straightforward.
However, I'd say that if your code relies on lexical
scoping to find something, you should probably rewrite your

code.

Except of course that almost every function relies on lexical
scoping to some extent!

This could get messy, because a) that's true and b) it actually

leads

to some genuine risks when 'globals' get redefined or masked*.

How about I amend the assertion to "if your code relies on

lexical

scoping to find a variable you defined, you should probably rewrite
your code."

and leave it at that, subject to some common sense about whether

you

know what you're doing?

That still isn't right, because users should feel free to define
functions and call them from their other functions.

I think who defined it isn't the issue, the issue is whether it

might

change unexpectedly.  The user owns globalenv().  The package

author

owns the package namespace.  So packages should almost never read

or

write things directly from/to globalenv() (the user might change

them),

but they can create their own private environments and write there.

Where it gets a little less clear is when the user writes a

function.

I
would say functions should never write directly to globalenv(), but
it's
perfectly fine to reference constants there (like other functions
written by the user).  Referencing things there that change is the
risky
thing.

Duncan Murdoch

Steve E


*Example

sin.deg  <- function(deg) sin(deg * pi/180)
sin.deg(45)

[1] 0.7071068
	#looks about right

pi <- 3.2       #Indiana General Assembly bill #247, 1897.
sin.deg(45)

[1] 0.7173561
	#oops ...

*******************************************************************

This email and any attachments are confidential. Any

use...{{dropped:8}}

project.org/posting-

guide.html

code.