Am 07.07.2015 um 11:59 schrieb David Carlisle: > On 07/07/2015 10:48, Élie Roux wrote: >> >> https://github.com/sh2d/spelling > > Excellent thanks! It's always good to have some real test cases that you > can try not to break. FWIW, for those who haven't already looked at the code, the whatsit id allocation can be found in file spelling_main.lua, function __allocate_whatsit_ids(). The user ids are then used in file spelling-stage-2.lua, function __tag_word() to wrap a "word" in a node list in whatsit nodes. I haven't look at that (experimental) code for a while, but I think instead of wrapping words in whatsit nodes I'll rather tag every node belonging to a word with an attribute in the future. Because adding whatsit nodes to a node list does have side effects. Best regards, Stephan Hennig