Max Nikulin writes: > Ideally it should be done pandoc and only if it causes incorrect > parsing of org markup. NBSP, probably, should be replaced by some > exporters, I do not think, it is a problem e.g. in HTML files.
The reason for this filter is my own comfort. Linguistics texts contains a lot of certain characters such as "/" or "*", and they are often italicized or bold. So, in order not to be more confused than necessary, I prefer that they pass as entities. In general, there are certain characters that I am more comfortable working with as entities than as literal characters (for example, a lot of zero-width combining diacritics that are used a lot in linguistics or epigraphy (and there are no fonts that include the NFC normalized version of all possible combinations: in fact, they are not in Unicode, and would have to go to the private use area). Summarizing, I prefer that these characters have their actual typographic representation only with LuaTeX. A very typical example is the character U+0323 (COMBINING DOT BELOW). It is very uncomfortable to work /in situ/, although there are fonts that usually render it well (with the 'mark' otf tag). (Naturally, I have to do, inside Org, a lot of corrections in italics later, due to the bad habit that Word users have of applying direct formatting. Interestingly only the pandoc docx reader trims the emphasis before exporting to Org or Markdown, so as not to produce things like "/ foo /". But the odt reader doesn't. I don't know if I'm missing something.