Have you tested the performance of the old vs new code? Note that the tests will also need to include rendering it to HTML text.

Quick benchmark:

system.time({
  for (i in 1:5000) {
    as.character(fa("calendar"))
  }
})

Result before:

   user  system elapsed 
  0.493   0.023   0.515 

After:

   user  system elapsed 
  4.900   0.030   4.935 

Following up on this PR, I think we should merge this only if we have a compelling use case in mind for the tags object over string. It's not too farfetched to imagine a document with a few hundred icons, but in that case, this change would add a non-negligible amount of time; we'd want to make sure there's a benefit that makes the cost worthwhile.

All committers have signed the CLA.