Completion of sentences stored in a file

Question

I know this question is similar to completion of a sentence rather than word but I ask anyway since I suggest another way of storing what is supposed to be auto-completed.

Suppose I stored some sentences in a file. How could I, in the current buffer, complete the sentences from this file just by typing the beginning (or, better, any part) of them?

Edit: To be more explicit, here is the context: I'm in the process of grading (maths) exams and most of the positive and negative remarks I have to express (and to write down) about my students works are the sames.

Hence, suppose I have the following sentences in a sentences.txt file:

Please, don't take your dreams for reality: (a + b)² ≠ a² + b².
Please, don't take your dreams for reality: cos(a + b) ≠ cos a + cos b.
You should revise the derivatives.
No, √(a²) ≠ a.
This only works for a positive x.

I'd like to be able in my current buffer (of a file stored, say, in the same directory as sentences.txt) to type e.g.

Please

and to be prompted with the first two sentences in a drop down list, just as does autocomplete for the (single) words of the current buffer.

Answer 1

Since you don't need actual abbreviations, just call completing-read with your list of sentences. This will let you define the list of sentences however you want. Here's the simplest possible implementation:

(defun db48x/insert-canned-response ()
  (interactive)
  (let ((lines (with-temp-buffer
                 (insert-file-contents "~/temp/sentences.txt")
                 (split-string (buffer-string) "\n" t))))
    (insert (ido-completing-read "Response: " lines))))
(global-set-key (kbd "C-c k") 'db48x/insert-canned-response)

I actually used ido-completing-read because it's fancier than the default completing-read; it shows you a preview of what you're about to select, and it lets you narrow the choices by substring match rather than just by prefix match. For example, this will let you type 'cos' to select the second response from your examples, which has the same prefix as the first response.

This implementation has obvious deficiencies, but they probably don't matter in practice. If your list of responses gets large enough or your hard drive is slow enough that it becomes a problem, you can ask a new question about it.