Can we avoid requiring .data$ to suppress R CMD check NOTEs?

[hadley]

R CMD check, calls codetools::checkUsageEnv, which calls codetools::checkUsage, which calls codetools::collectUsage with a bunch of arguments set (that's where all the work happens). The key function appears to be codetools:::checkUsageEnterGlobal():

function (type, n, e, w) 
{
    if (type == "function") {
        if (exists(n, envir = w$globalenv, mode = "function")) {
            def <- get(n, envir = w$globalenv, mode = "function")
            if (typeof(def) == "closure") 
                checkCall(def, e, function(m) w$signal(m, w))
            else {
                isBuiltin <- typeof(def) == "builtin"
                checkPrimopCall(n, e, isBuiltin, function(m) w$signal(m, 
                  w))
            }
        }
        else if (!suppressVar(n, w$suppressUndefined)) 
            w$signal(paste("no visible global function definition for", 
                sQuote(n)), w)
    }
    else if (type == "variable") {
        if (!exists(n, w$globalenv) && !suppressVar(n, w$suppressUndefined)) 
            w$signal(paste("no visible binding for global variable", 
                sQuote(n)), w)
    }
    else if (type == "<<-") {
        if (!exists(n, w$globalenv)) 
            w$signal(paste("no visible binding for '<<-' assignment to", 
                sQuote(n)), w)
    }
}

But I think that's too far down the tree in order to be able to special case tidy-evaluation using functions.

The handlers for special base R functions are stored in environment codetools:::collectUsageHandler, so one possible last resort hack would be to register handlers these on load if R CMD check is running.