I have the following HTML code:
Hello from the year 2020!
As of v2.9.9 of libxml, this behavior has been fixed in xmllint itself.
However, if you're using anything older than that, and don't want to build libxml from source just to get the fixed xmllint, you'll need one of the other workarounds here. As of this writing, the latest CentOS 8, for example, is still using a version of libxml (2.9.7) that behaves the way the OP describes.
As I gather from this SO answer, it's theoretically possible to feed a command into the --shell option of older (<2.9.9) versions of xmllint, and this will produce each node on a separate line. However, you end up having to post-process it with sed or grep to remove the visual detritus of shell mode's (human-oriented) output. It's not ideal.
XMLStarlet, if available, offers another solution, but you do need to use xmlstarlet fo to format your HTML fragment into valid XML before using xmlstarlet sel to extract nodes:
echo '
' \
| xmlstarlet fo -H -R \
| xmlstarlet sel -T -t -v '//textarea[@name="command"]' -n
If the Attempt to load network entity message from the second xmlstarlet invocation annoys you, just add 2>/dev/null at the very end to suppress it (at the risk of suppressing other messages printed to standard error).
The XMLStarlet options explained (see also the user's guide):
fo -H -R — format the output, expecting HTML input, and recovering as much bad input as possible
root node, making the fragment in the OP's example valid XMLsel -T -t -v //xpath -n — select nodes based on XPath //xpath
-T) instead of XML-t) that returns the value (-v) of the node rather than the node itself (allowing you to forgo using text() in the XPath expression)-n)Edit(s): Removed half-implemented xmllint --shell solution because it was just bad. Added an XMLStarlet example that actually works with the OP's data.