lattice: how to "center" a subtitle?

[...]

I am not sure where I read it and I can't find it again, but my
understanding is that expressions using bquote with lattice need to be
enclosed in as.expression() to work. That is in contrast to what happens in
base graphics.

Here is a simple example.

a <- 2
plot(1:10, a*(1:10), main = bquote(alpha == .(a)))
require(lattice)
xyplot(a*(1:10)~ 1:10, main = bquote(alpha == .(a)))
xyplot(a*(1:10)~ 1:10, main = as.expression(bquote(alpha == .(a))))

Which produces:

a <- 2
plot(1:10, a*(1:10), main = bquote(alpha == .(a)))
require(lattice)

Loading required package: lattice

xyplot(a*(1:10)~ 1:10, main = bquote(alpha == .(a)))

Error in trellis.skeleton(formula = a * (1:10) ~ 1:10, cond = list(c(1L, ?:
?object 'alpha' not found

xyplot(a*(1:10)~ 1:10, main = as.expression(bquote(alpha == .(a))))

Using expression() rather than as.expression() doesn't produce the desired
affect. Try it yourself.

As to why this is the case .....

Let's see: ?xyplot says

          'main': Typically a character string or expression describing
              the main title to be placed on top of each page. [...]

So, lattice is fairly explicit, by R standards, in requiring 'main' to
be "character" or "expression". On the other hand, ?title says

     The labels passed to 'title' can be character strings or language
     objects (names, calls or expressions), or [...]

so it additionally accepts "names" and "calls".

Now, we have

a <- 2
foo <- bquote(alpha == .(a))

foo # Looks OK

alpha == 2

mode(foo) # But

[1] "call"

is.expression(foo) # not an expression

[1] FALSE

is.expression(expression(foo)) ## YES, but

[1] TRUE

expression(foo) ## not what we want

expression(foo)

is.expression(as.expression(foo))

[1] TRUE

as.expression(foo) ## This IS what we want

expression(alpha == 2)

So I submit that lattice is behaving exactly as suggested by its documentation.

Now you would naturally argue that this is hiding behind
technicalities, and if "call" objects work for plot(), it should work
for lattice as well. But watch this:

plot(1:10, main = foo) # works perfectly

arglist <- list(1:10, main = foo)
arglist # Looks like what we want

[[1]]
 [1]  1  2  3  4  5  6  7  8  9 10

$main
alpha == 2

do.call(plot, arglist)

Error in as.graphicsAnnot(main) : object 'alpha' not found

...which I would say is "unexpected" behaviour, if not a bug.

The moral of the story is that unevaluated calls are dangerous objects
(try this one out for fun:

foo <- bquote(q(.(x)), list(x = "no"))
do.call(plot, list(1:10, main = foo))

), and carrying them around is not a good idea.

Lattice does use the do.call paradigm quite a bit, and I think it
might be quite difficult to fix it up to handle non-expression
language objects (which will still not fix the type of problem shown
above).

-Deepayan