rvest object cloned with rlang::duplicate is not properly cloned - r

Question

rvest doesn't seem to offer any way to extract text from parent object only (ignoring children). One workaround uses xml_remove(), which mutates the original object - all the way up the memory chain given R's default lazy evaluation.
I look to rlang::duplicate(), which is supposed for "modifying the copy leaves the original object intact", but the clone does not appear to be truly independent. For example:
require(rvest)
h = '<ul>
<li id="target">
text to extract
<ul><li>text to ignore</li><li>this too</li></ul>
</li>
</ul>'
doc = xml2::read_html(h)
x = html_node(doc, '#target')
html_text(x)
#> [1] "\ntext to extract\ntext to ignorethis too"
Now clone x, remove its children, and extract the text:
x2 = rlang::duplicate(x, shallow = FALSE)
children = html_children(x2)
xml2::xml_remove(children)
html_text(x2)
#> [1] "\ntext to extract\n"
That works as intended, however x has also been mutated:
html_text(x)
#> [1] "\ntext to extract\n"
Any suggestions why and how to workaround this? I do not want to start re-attaching children..

Answer 1

First of all let me say that I think yoo can solve the issue without copying the data. I'm not an expert in xpath, but I think you can use it to just select only direct text descendents, ignoring text nested in other xml nodes. I.e. the following seems to do the trick without any copy (x defined as in your question):
html_text(html_elements(x, xpath = "text()"))
# [1] "\ntext to extract\n"
That being said, I also have an answer to the question on how to make a deep copy:
The problem is that rlang::duplicate() can only copy R data structures. However, rvest builds on xml2, and xml2 builds on the C library libxml2.
When you create the xml_node object in R, the corresponding data structure is created in libxml2. On the R side, there is basically just a pointer to the libxml2 object. So rlang::duplicate() will only create a copy of that pointer, but not of the underlying data. It cannot do so, because it has no access to it as it is in a different library (that rlang doesn't know of).
The easiest way to create a copy of the underlying data seems to be to serialize and deserialze the xml. I suspect this is not very efficent though.
Example:
Read in the original data:
require(rvest)
h <- '<ul>
<li id="target">
text to extract
<ul><li>text to ignore</li><li>this too</li></ul>
</li>
</ul>'
doc <- xml2::read_html(h)
x <- html_node(doc, '#target')
Create two copies - one with rlang:duplicate() and one with xml2::xml_unserialize():
x1 <- rlang::duplicate(x, shallow = FALSE)
x2 <- xml2::xml_unserialize(xml2::xml_serialize(x, NULL))
Check that x and x1 are in fact identical, while x2 is a true copy (the memory locations you get will be of course be different to the ones shown here):
x$doc
# <pointer: 0x0000023911334ea0>
x1$doc
# <pointer: 0x0000023911334ea0>
# --> same as x
x2$doc
# <pointer: 0x00000239113377d0>
# --> different to x
Test that everything works as intented:
children <- html_children(x2)
xml2::xml_remove(children)
html_text(x2)
# [1] "\n text to extract\n "
html_text(x)
# [1] "\n text to extract\n text to ignorethis too"

Answer 2

Another potential solution (maybe a more general approach) is to use the html_children() function to obtain the text of all the child nodes and then remove that from the full text.
require(rvest)
h = '<ul>
<li id="target">
text to extract
<ul><li>text to ignore</li><li>this too</li></ul>
</li>
</ul>'
doc = xml2::read_html(h)
x = html_node(doc, '#target')
fulltext <- html_text(x)
# [1] "\ntext to extract\ntext to ignorethis too"
#find the text in the children nodes
childtext <- html_children(x) %>% html_text()
# "text to ignorethis too"
#replace the child node text with a numm
gsub(childtext, "", fulltext) %>% trimws()
#"text to extract"
#alternative using the text from the first child node
firstchild <- xml_child(x, search=1) %>% xml_text()
gsub(paste0(firstchild, ".*"), "", fulltext)
Of course, if there are additional newline "\n" or formatting character, the gsub() may break.