Allow control-E to complete a command

[waterhouse]

At the moment, fish's command completion appears to work like this: If, e.g., I've recently typed:
sudo port selfupdate; port echo outdated
then, if I type sudo port se, the letters lfupdate; port echo outdated appear highlighted in yellow. At this point, I can use either control-F or the right arrow to accept the completion text, making it as though I had typed it myself.

I'd like to be able to use control-E to do this too. I find it easier to type control-E than control-F in this circumstance (I use the Dvorak keyboard), and it is logically permissible because control-E has rightward-cursor-moving semantics.

To be precise: if control-E is used when the cursor is at the end of a (partially) typed command, and if there is completion text, then control-E should accept the completion text and move the cursor to the end of the (now larger) command. control-E should not be otherwise changed. (Consequently, if you type a partial command, then move the cursor left, it would take two control-E's to accept the completion.)

[zanchey]

fish allows you to change your keyboard bindings: bind \ce forward-word will do what you want. Note that you need to add it to the fish_user_keybindings function to have it loaded on startup.

[waterhouse]

I'm afraid that doesn't do what I want. First of all, going forward by a word of completed text is not the same as going to the end of the completion text if that text is more than one word. Second, and more seriously, control-E normally moves the cursor to the end of the line, and rebinding it destroys that behavior.

Currently, \ce is bound to end-of-line. The solution would be to modify the end-of-line procedure so that, if it is invoked when already at the end of a (typed) line and there is pending completion text, it will accept the completion text and move to the end of that.

[zanchey]

Sorry, I meant forward-char. I see what you mean now though, I think #344 covers the same thing.

[waterhouse]

I observe that, in reader.cpp, the code corresponding to the forward-char function uses accept_autosuggestion(true) to, well, accept, and uses a simple length comparison to detect end-of-command.

            case R_FORWARD_CHAR:
            {
                if (data->buff_pos < data->command_length())
                {
                    data->buff_pos++;
                    reader_repaint();
                }
                else
                {
                    accept_autosuggestion(true);
                }
                break;
            }

I also observe the code corresponding to R_FORWARD_WORD and R_END_OF_LINE. The fix appears straightforward. This is what I did to the R_END_OF_LINE case in reader.cpp:

            case R_END_OF_LINE:
            {
              if (data->buff_pos < data->command_length())
              {
                  while (buff[data->buff_pos] &&
                          buff[data->buff_pos] != L'\n')
                  {
                      data->buff_pos++;
                  }
              }
              else
              {
                accept_autosuggestion(true);
              }

              reader_repaint();
              break;
            }

Doing this and recompiling fish does exactly what I want.

Will this be accepted? Should I submit a diff or something to make it easier for project administrators?

[zanchey]

#91 even.

[waterhouse]

Yes, I see a trail of previous issues. #91 seems to be on essentially the same issue. Should I comment there rather than here?

[zanchey]

Either is fine. A diff would be good, or even a pull request. Change looks good though.

[waterhouse]

Diff provided (see other thread).

[ridiculousfish]

No objection from this fish. Merge or not as you see fit, zanchey.

[KamilaBorowska]

I'm fine with this change too. I don't use CTRL+E, but it doesn't look like this change will break anything (except for using CTRL+E when it would do nothing anyway).

[zanchey]

fixed on master with e3ea953

Thanks!

[waterhouse]

Excellent! Glad to have helped.