Let's think big and look for a generic solution. IMHO, it is way too restrictive to insist that a word is anything matching the pattern /\k\+/ . I want a new option, 'wordpat', with a default value of '\k\+', that specifies what should be recognized as a word, for purposes of search patterns, Normal-mode commands such as w and b, and maybe other uses. (Oh, yes: Insert-mode completion.)Examples: :let &l:wordpat = '\k\+\(-\k\+\)*'
In the general, I like it! In the implementation, I don't know if there are snags that one will encounter. One might have to include the cursor position to anchor it in the search text.
Funky conditions could occur if patterns contain certain atoms (for better or worse). Could things like using \%<9c enforce that words are only contained before column 9 (makes me think of my cobol days)? Or even something like '\%>0l\%<42l\k*\%#\k*' enforce that keywords are only found in the first 41 lines of your file? Or can keywords only be duplicate-part words like "mahimahi" using a pattern like '\(\k\+\)\1'? Or keywords are only ever preceeded by "int" like 'int\_s\+\zs\k+'? Or alter the behavior of the "*" and "#" commands to find the word that *preceeds* the word under the cursor with something like '\k\+\ze\K\+\k*\%#'
All these seem like sensible (for cases where "sensible" may be a subset of as "pathological") potential use-cases for such a thing.
-tim
