A wrapper around function rbindlist() in R package data.table that tries to handle some of its limitations. Basically it preprocess the input parameter by filling in any missing columns with NAs, ensuring all column types match and are in the same order before passing it along to rbindlist.

gistfile1.r

# A wrapper around rbindlist that ensures that all columns are matched in type and
# in order before calling it.
#
smartrbindlist <- function(l) {
  if (missing(l) && !is.list(l)) {
    stop("smartrbindlist: bad or missing argument 'l")
  }
  
  mergedColClasses = list()
  i=1
  while(i <= length(l)) {
    # Validate entry types and get column names
    if (is.data.frame(l[[i]])) l[[i]] = as.data.table(l[[i]])
    if (is.data.table(l[[i]])) {
      cn = colnames(l[[i]])
    } else if (is.list(l[[i]])) {
      cn = names(l[[i]])
    } else {
      if (!is.null(l[[i]])) {
        warning(sprintf("smartrbindlist: skipping entry %i with unexpected class [%s]",
                        i, paste(class(l[[i]]), collapse=", ")))
      }
      cn = character()
    }
    if (length(cn) == 0) {
      l[[i]] = NULL
      next
    }
    
    # For all column names we've seen before check the classes for inconsistencies
    match_cn = cn %chin% names(mergedColClasses)
    for (col in cn[match_cn]) {
      if (!isTRUE(all.equal(sort(mergedColClasses[[col]]), sort(class(l[[i]][[col]]))))) {
        msg = sprintf("smartrbindlist: column %s has different classes in entry %i [%s] and its predecessors [%s]",
                      col, i, paste(class(l[[i]][[col]]), collapse=", "), 
                      paste(mergedColClasses[[col]], collapse=", "))
        stop(msg)
      }
    }
    
    # Add previously unknown column names to list of known ones.
    mergedColClasses[cn[!match_cn]] = lapply(cn[!match_cn], function(x) class(l[[i]][[x]]))
    i = i+1
  }
  if (length(l) == 0) return(data.table())
  
  # Now we do a second pass, add any missing columns and reorder fields/columns
  for (i in 1:length(l)) {
    if (is.data.table(l[[i]])) {
      cn = colnames(l[[i]])
      if (nrow(l[[i]]) == 0) {
        for (col in names(mergedColClasses)[!names(mergedColClasses) %chin% cn]) {
          l[[i]][[col]] = getNAforClass(mergedColClasses[[col]])[F]
        }
      } else {
        for (col in names(mergedColClasses)[!names(mergedColClasses) %chin% cn]) {
          l[[i]][[col]] = getNAforClass(mergedColClasses[[col]])
        }
      }
      setcolorder(l[[i]], names(mergedColClasses))
    } else if (is.list(l[[i]])) {
      cn = names(l[[i]])
      for (col in names(mergedColClasses)[!names(mergedColClasses) %chin% cn]) {
        l[[i]][[col]] = rep(getNAforClass(mergedColClasses[[col]]), length(l[[i]][[1]]))
      }
      l[[i]] = l[[i]][names(mergedColClasses)]
    }
  }
  
  return(rbindlist(l))
}

# Returns a single NA value for the given class (as returned by the class function).
# Extend this to you heart's content. :)
# If a type is not recognized, returns plain NA (logical).
#
getNAforClass <- function(cl) {
  if ("numeric" %chin% cl) {
    return(NA_real_)
  } else if ("character" %chin% cl) {
    return(NA_character_)
  } else if ("factor" %chin% cl) {
    return(as.factor(NA_character_))
  } else if ("int64" %chin% cl) {
    return(as.int64(NA_real_))
  } else if ("integer" %chin% cl) {
    return(NA_integer_)
  } else if ("POSIXct" %chin% cl) {
    return(as.POSIXct(NA_real_, origin="1970-01-01"))
  } else {
    return(NA)
  }
}

See the discussion that led me to write this here and here.