API design: use pass_if() and fail_if()

[andrie]

Intent: use a similar API to the grader project, as described in rstudio/gradethis#11

In a nutshell, you should be able to write:

• initial contains the initial value
• pass_if()
• fail_if()

For both pass_if() and fail_if(), support:

• character vector, and pass if exact match
• function in the traditional form function(x)...
• function using formula form ~. == ...

```{r iris, echo=FALSE}
question_parsons(
  initial = c(
    "iris",
    "mutate(...)",
    "summarize(...)",
    "print()"
  ),
  pass_if(
    c(
      "iris",
      "mutate(...)",
      "summarize(...)"
    )
  ),
  pass_if(...),
  fail_if(
    ~length(.) < 2,
    message = "Include at least two answers"
  ),
  fail_if(
    function(x){"print()" %in% x},
    message = "You should not include print() in your answer"
  ),
  fail_if(
    function(x){x[1] != "iris"},
    message = "Your solution should start with 'iris'"
  )
)
```

[andrie]

@hadley suggests we make it possible to provide R code as the value for initial, and provides this sample code to parse the AST:

library(rlang)

parsonize <- function(x) {
  x <- enexpr(x)
  
  if (is.character(x)) {
    # Literal character string, so split up into lines, using glue's
    # triming algorithm to remove leading whitespace
    strsplit(glue::trim(x), "\n")[[1]]
  } else if (is.call(x)) {
    depipe(x)
  } else {
    abort("`x` must be a string or a function call.")
  }
}

# Returns a character vector
depipe <- function(x) {
  if (is_call(x, "+", 2) || is_call(x, "%>%", 2)) {
    c(depipe(x[[2]]), depipe(x[[3]]))
  } else {
    expr_text(x)
  }
}

parsonize("
  a + 
    b + 
    c
  ")

parsonize(a + b + c)
parsonize(a %>% b %>% c(d + e))