returning a list of nodesets

[jakejh]

Hello, I have a use case in which I want to run xml_find_all on a nodeset, but have the result be a list of nodesets (one element for each node) instead of one flattened nodeset. In this use case, it's important that I know which result node came from which original node. In my testing it's way faster to have this functionality within the xml2 package, since it can directly call C in the lapply, than to call xml_find_all in a for or foreach loop. I can imagine a couple ways to do this.

1. Extend the xml_find_all.xml_nodeset method by adding an unlist argument. This would break the convention that "xml_find_all always returns a nodeset", but seems cleaner to me. Something like below:

xml_find_all.xml_nodeset <- function(x, xpath, ns = xml_ns(x), unlist = TRUE) {
  if (length(x) == 0)
    return(xml_nodeset())

  if (isTRUE(unlist)) {
    nodes <- unlist(recursive = FALSE,
                    lapply(x, function(x)
                      .Call(xpath_search, x$node, x$doc, xpath, ns, Inf)))

    nodes <- xml_nodeset(nodes)

  } else {
    nodes <- lapply(x, function(x)
      xml_nodeset(.Call(xpath_search, x$node, x$doc, xpath, ns, Inf)))
  }

  return(nodes)
}

2. Add a separate function called xml_find_all_list, something like below:

xml_find_all_list <- function(x, xpath, ns = xml_ns(x)) {
  if (length(x) == 0)
    return(xml_nodeset())

  lapply(x, function(x)
    xml_nodeset(.Call(xpath_search, x$node, x$doc, xpath, ns, Inf)))
}

I'm happy to make a PR, just let me know how to proceed. Thanks.

[jimhester]

What about calling the argument flatten with an implementation like the following?

xml_find_all.xml_nodeset <- function (x, xpath, ns = xml_ns(x), flatten = TRUE) {
  if (length(x) == 0)
    return(xml_nodeset())

  res <- lapply(x, function(x) .Call(xpath_search, x$node, x$doc, xpath, ns, Inf))

  if (isTRUE(flatten)) {
    return(xml_nodeset(unlist(recursive = FALSE, res)))
  }

  res[] <- lapply(res, xml_nodeset)
  res
}

[jakejh]

I like the minimal redundancy. I initially considered something similar, but didn't know if calling lapply twice could slow the function down. Now that I've actually tested it though, the timings are virtually identical, at least on my xml files.

Looks like you have this under control. I'm sure you can add documentation and unit tests faster than I can, but let me know how I can help.

[jimhester]

If you want this feature please open a PR including the documentation and unit tests.