Request: filename related functions in readtags expressions

[AmaiKinono]

Currently I don't know what exactly is needed, but there are 2 motivations behind this request, I hope they can guide us.

resolve relative file name

When finding definitions, the user may want to exclude file-local symbols that's not in current file. The predicate "not in current file" is actually not a simple one. Here's a one generated by Citre (if you are interested, it's in here)

(or (and $input (eq? $input "/path/to/test.el"))
    (and $input (eq? $input "test.el"))
    (and $input ((string->regexp "(^|/)..?/test\\.el$"
                                 :case-fold false)
                 $input)))

Suppose the current file is /path/to/test.el, and TAG_PROC_CWD is /path/to/, then the input field of this file should either be test.el (when its relative) or /path/to/test.el (when its absolute). So where does the regexp matching against (^|/)..?/test\\.el$ comes from?

This is because a user reported that if you tag the current dir by:

$ ctags -R ./

You get tags like this:

citre-put-property      ./citre-common.el

I soon realized that the input field can look like anything, e.g., you can feed a path like ../../path/to/ to the ctags command, and it could mean the same as /path/to.

At the end, what I did is: when there's a /./ or /../ in the input field, and it ends with test.el, we think it is the file /path/to/test.el. This is not correct, but is the best I can do for now.

Since there are infinite possibilities, direct matching against the input field simply can't solve the problem. We must somehow transform $input to a standard representation, then match against it.

Sort by nearness

This is a very nice feature I found in GNU Global. I'll paste its documentation here:

-N, --nearness[=start]
       Use Nearness sort method (sorting by closest from start) for the output.
       By default, alphabetical sort method is used.
       This option is effective for the tag search command, -P command
       and -g command. As an exception, -g command ignores this
       option when files are specified by arguments.
       The nearness is defined by how many parent directories to go up to reach
       the target. The result of nearness sort is concatenation of the following
       ([0]-[n]) in this order. The default of start is the current directory.
       
       [0] If the start is a file, output of local search in the file.
       [1] Output of local search in the start directory except for [0].
       [2] Output of local search in the parent directory except for [0]-[1].
       [3] Output of local search in the grandparent directory except for [0]-[2].
       ... (repeat until the project root directory)
       [n] Output of local search in the project root directory except for [0]-[n-1].
       
       In each directory, they are sorted by alphabetical order.

Simply puts it, file that are "near" to the current file is put above those "far" from the current file. This is handy when finding references, but I think readtags could also benefit from this, since:

1. uctags supports some reference tags, like references to C macros.
2. Normally we structure our project so that closely related code are put in the same file, code that are not very closely related are put in the same directory, then the same module, then different modules... So when there are multiple definitions found for a symbol, the "nearer" it is to the current file, the more possible its the definition we want.

[masatake]

(NOTE: https://en.wikipedia.org/wiki/Edit_distance)

[masatake]

I assume "resolve relative file name" can be solved easily.
I will work on after working on OpenAPI parser.
Sorry to be late to respond.
I have been thinking about "Sort by nearness". This one looks hard.

[AmaiKinono]

I have been thinking about "Sort by nearness". This one looks hard.

Maybe it's not that hard ;) Here's a prototype in Elisp:

(defun common-prefix-length (str1 str2)
  (pcase (compare-strings str1 nil nil
                          str2 nil nil)
    ('t (length str1))
    ((and result (guard (< result 0))) (- -1 result))
    ((and result (guard (> result 0))) (- result 1))))

(defun file-nearness (start to)
  (let* ((start (expand-file-name start))
         (to (expand-file-name to))
         (common-prefix-length (common-prefix-length start to))
         (rest-part (substring start common-prefix-length))
         (separators 0))
    (if (= common-prefix-length (length start) (length to))
        0
      (dotimes (i (length rest-part))
        (when (memq (aref rest-part i) '(?/ ?\\))
          (setq separators (1+ separators))))
      (1+ separators))))

(file-nearness "~/path/to/abc.c" "~/path/to/abc.c")
;; => 0
(file-nearness "~/path/to/abc.c" "~/path/to/a.h")
;; => 1
(file-nearness "~/path/to/abc.c" "~/path/to/foo/abc.c")
;; => 1
(file-nearness "~/path/to/abc.c" "~/path/too/abc.c")
;; => 2
(file-nearness "~/path/to/abc.c" "~/foo")
;; => 3

[masatake]

I misunderstood the issue; it seems that we have to add an operator.

[masatake]

I implemented -c option for resolve relative file name.

[masatake]

"Sort by nearness" is not solved yet.