Vectorising XPath expressions

[pvanlaake]

I regularly find myself searching XML documents for multiple unique entries among the children of a parent node. I have a handcrafted function that vectorises over the names of the entries and it thus makes separate calls to the XML library for each name; see simplified function below.

Would it be possible to vectorise at the level of the C++ code, rather than in R? I would imagine that the performance would be much better than the serialised calls I have in my code below.

# Argument `names` is a character vector
find <- function(names) {
  xp <- paste0("//entry[@id='", names, "']")
  
  # Vectorise this call to `xml_find_first()`, rather than the `lapply()` that I am using now
  nodes <- lapply(xp, function(nm) xml2::xml_find_first(my_xml_document, nm))

  # Process any nodes that were retrieved
  found <- which(sapply(nodes, class) == "xml_node")
  if (length(found)) {
    nodes <- nodes[found]
    cu <- sapply(nodes, function(n) xml2::xml_text(xml2::xml_find_first(n, "canonical_units")))
    desc <- sapply(nodes, function(n) xml2::xml_text(xml2::xml_find_first(n, "description")))
    data.frame(name = names[found], units = cu, description = desc)
  } else NULL
}

[jeroen]

I think for most cases the return type would need to change if we allow multiple xpaths, because we get multiple documents or nodesets. So there is no obvious way these calls can be generalized for vectors of xpaths.

Also it would not matter performance wise, we still need to do the same operations, and the overhead of doing this in R code is really small.