Here's one take. I attempt to make
up for lack of rigour here by
expenditure of vigour, which never
gets high grades.
You are trying to treat a scalar
as if it were a local. It is not the
same thing; if it were, one of the ideas
would be redundant. Rather, a scalar _name_
can be a local macro. (It could be something
permanent, as well.)
A local has no real use unless its name
is written down, so that Stata sees the
name and substitutes the contents.
(You _could_ always just put
something into a macro and do nothing
with it, but that would be pointless.)
As a consequence, writing down
a local as part of a command in effect
gives you its contents straight away:
local i = 1
gen mpg`i' = mpg + `i'
immediately yields
gen mpg1 = mpg + 1
It's as if defining a local macro were
like putting something in a transparent
box. That is, someone's given you
a present in a box, but the box is
transparent, so that immediately
you can see what the present is. It's
Stata 10! (Just kidding.)
You know all this. So how does a scalar
differ? With a scalar the box is
less transparent. Depending on context,
you may get first the _name_, not the contents,
which is not what you want.
tempname foo
scalar `foo' = 3
forval i = 1/`foo' {
...
}
will fail because here Stata sees the
name `foo' -- really __000002 or
some such -- when what it wants in this
context is a number -- and what you want, similarly,
is that Stata sees a number.
This would work:
forval i = 1/`=`foo'' {
...
}
in which you force Stata to evaluate
your expression (here just a single name,
but the principle is the same),
i.e. to look beyond the name and
find the contents, to open the box,
in short.
Note that appealing to -display-
muddies the distinction here, and
makes the difference between locals
and scalars with local macro names seem
smaller than it really is. It is -display-'s job
to open boxes -- that is precisely
what is designed to do. It is
a pair of scissors and a knife
and a little helper all in one.
However, -forval-
does not have the job of opening
boxes. It does none of that. Before
-forval- does anything, however,
the parser does substitute macro
references. With a local macro
that's enough. With a scalar
you need to go one step further,
because the name is not what you need.
Context is important here. With
the same scalar on the right-hand
side
gen bar = `foo' + 5
Stata again is evaluating and
so looks inside the box. This
is just what you expect from
first lessons in Stata: with
gen mpg2 = mpg + 5
Stata is obliged by the context
to look beyond the (variable)
name -mpg- and get at the contents,
the values of the variables.
-set trace on- helps you see
what's happening here.
Moral: Keep straight the difference
between names and contents, between
what is on the outside of the box
and what's inside.
P.S. -while-? Well, I have no
special story here. Evidently,
Stata was written so that the
-while- context forces evaluation.
Perhaps the designers of Stata
can spell out the logic.
Nick
[email protected]
Dimitriy V. Masterov
> Can someone please explain to me why one cannot use scalars
> to define the
> upper bound in a range of a forvalues loop. A forvalues loop
> with a local
> works, a while loop with a scalar works, a while loop with a
> local works,
> but not a forvalues loop with a scalar. There seems to be
> nothing in the
> programming manual about this issue.
>
> This is the code for the examples above:
>
> tempname m1 m2
>
> scalar `m1'=5
> local `m2'=5
>
>
> forvalues a=1(1)``m2'' {
> di "`a'"
> }
>
> local a=1
> while `a'<=``m2'' {
> di "`a'"
> local a=`a'+1
> }
>
>
> local a=1
> while `a'<=`m1' {
> di "`a'"
> local a=`a'+1
> }
>
> forvalues a=1(1)`m1' {
> di "`a'"
> }
>
> I am using Stata 8.2 SE with XP.
>
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
