15

I'm using Emacs for quite some time and I still somehow fail to understand what is the best way to (recursively) grep for a string within a (project) directory and get the results presented either as a dired buffer or in some even more useful way. Please enlighten me.

Improve this question
  • Have you tried out using helm-projectile-grep command (if you have helm projectile installed) or projectile-grep ? – Chakravarthy Raghunandan Aug 17 '16 at 11:16
  • I do not know what helm or projectile is, but if it is a recommended tool, I would install and learn it. – Boris Stitnicky Aug 17 '16 at 11:27
  • Yes, projectile is a package that provides an easy way to do what you mentioned in the question with the command projectile-grep. Also, I HIGHLY recommend you install helm package. I can't imagine running emacs without helm and helm-projectile. You might also be interesting in helm-ag package (which provides a helm interface for silver searcher in emacs, which is supposedly faster than grep or ack). – Chakravarthy Raghunandan Aug 17 '16 at 11:31
  • 2
    So that the question is more useful and easier to find for future Google and forum searchers, perhaps consider modifying the title of the question to include the word recursively, and consider limiting the scope -- e.g., How to recursively grep file contents in a directory / subdirectories. That's just an example -- it could be something else instead. – lawlist Aug 17 '16 at 17:19
  • 1
    While the comments and answers so far have suggested a number of packages to do this in various fancy ways, I don't see from your question why simple M-x grep doesn't do what you need. It let's you give an arbitrary command line, I sometimes use find there when I need the recursion. – MAP Aug 19 '16 at 7:36
15

Well, since the original question doesn't mention rgrep, I'll go ahead and point it out here. This is the simplest option already built into Emacs, and I've found it to be more than sufficient for most things.

From the documentation,

Recursively grep for REGEXP in FILES in directory tree rooted at DIR.

The search is limited to file names matching shell pattern FILES.

So, for instance, to search all python files beneath my home directory for uses ofsome_function, I would call it with some_function, *.py, ~/ as the arguments. Results are displayed in a new buffer.

Improve this answer
  • 1
    Might want to mentioned find-grep-dired as well since OP mentioned "results presented either as a dired buffer or..." – npostavs Aug 18 '16 at 10:52
  • @npostavs I haven't used find-grep-dired myself, feel free to add it to the answer. – user2699 Aug 18 '16 at 15:12
  • I once experimented with the likes of ack and ag, on the assumption that I should be switching from rgrep to something using those, because they were supposed to be better. In the end I stayed with rgrep, because I found no benefit to changing! On the command line there's certainly a case to be made, but in Emacs rgrep does such a good job of targeting the search that there were no significant speed differences between them. (I want to say that rgrep was fractionally faster in some cases, but I don't remember for sure.) – phils Aug 31 '16 at 5:39
  • 1
    Note that next-error and previous-error can be used to navigate through the result set. I find M-g M-n and M-g M-p the most convenient of the standard bindings. – phils Aug 31 '16 at 5:50
  • find-grep-dired doesn't give me a buffer with cross-referencing links. rgrep does it for me and that last suggestion about next-error previous-error is fantastic! My only quibble is that all the messy command execution output at the start of the buffer. – robert Feb 22 '18 at 11:23
11

I happily use M-x find-grep-dired for years. It returns the results as a dired buffer, which you could use.

Improve this answer
  • I've had some difficulty with getting the cross-reference linking to work with this. It's configuration is quite nasty (find-ls-option) and after all that you end up with something that's quite verbose because it doesn't work without having the date before the filename! – robert Feb 22 '18 at 15:05
5

Solution 1 (best solution):

Install counsel (https://github.com/abo-abo/swiper/blob/master/counsel.el)

Then M-x counsel-git-grep.

No setup needed (git knows the project root and files to exclude). Both git grep and counsel is efficient.

The project need be managed by git.

counsel requires ivy-mode.

Solution 2:

This solution uses grep and works on any project. It's inferior to the Solution 1 because it's slower and need manual setup. It's also based on ivy-mode.

(defvar simple-project-root "~/.emacs.d")
(defun simple-grep ()
  (interactive)
  (unless (featurep 'ivy)
    (require 'ivy))
  (let* ((default-directory (file-name-as-directory simple-project-root))
         (keyword (read-string "Keyword:")))
    (ivy-read (format "Grep at %s:" simple-project-root)
              (split-string (shell-command-to-string (format "grep -rsnI %s ." keyword)) "\n" t)
              :action (lambda (val)
                        (let* ((lst (split-string val ":"))
                               (linenum (string-to-number (cadr lst))))
                          ;; open file
                          (find-file (car lst))
                          ;; goto line if line number exists
                          (when (and linenum (> linenum 0))
                            (goto-char (point-min))
                            (forward-line (1- linenum))))))))

You need create .dir-locals.el to setup simple-project-root, see https://www.gnu.org/software/emacs/manual/html_node/emacs/Directory-Variables.html for technical details

The code in solution 2 is just a prototype. My real implementation is much more complex. See counsel-etags-grep in https://github.com/redguardtoo/counsel-etags/blob/master/counsel-etags.el

Summary:

Those are the best two solutions I know.

If any other better solutions exist, they need at least solve below problems to be production-ready,

  • how to get the keyword to grep (for example, get keyword from selected region)

  • escape the keyword

  • if more efficient grep program exist, we should use it (ripgrep, the_silver_searcher/ag, ... etc), or else fallback the default grep

  • candidate window should use full width of screen and users can filter candidates interactively (that's why people use ivy or helm)

  • we should show the relative path in candidate window

  • able to re-use previous grepped result. So previous result should be saved. You can use ivy-resume from ivy or helm-resume from helm

  • When you save previous grepped result, the context of previous result should also be saved. For example, in the code of of Solution 2. default-directory is context. See https://github.com/abo-abo/swiper/issues/591 for more details.

  • Extended regular expression should be used because it's simpler and is already used by counsel-git-grep and the the_silver_searcher/ag.

Improve this answer
4

I'd like to add another solution to @chen bin's solution which also uses counsel (but you don't need to have a project maintained by git for this).

You need to install ag (the silver searcher) and counsel provides the command counsel-ag which searches the current directory for the entered string.

If you want to search within a project(not just the current working directory), add this to your .emacs :- 

  (defun rag/counsel-ag-project-at-point ()
    "use counsel ag to search for the word at point in the project"
    (interactive)
    (counsel-ag (thing-at-point 'symbol) (projectile-project-root)))

This function will use ag to recursively search the project root directory.

Edit: the above function defaults to the symbol at point for searching. If you do not like that behaviour, replace (thing-at-point 'symbol) with nil. And if you want the search results in it's own buffer press C-c C-o

Edit2: if you've ripgrep installed, you can replace counsel-ag with counsel-rg in the above function and it'll just as fine :)

Improve this answer
  • Is projectile-project-root useable to find Rails project root? – Boris Stitnicky Dec 7 '16 at 13:49
  • Yes, as long as it's a valid projectile project it will work. There was some package called projectile-rails I think. – Chakravarthy Raghunandan Dec 7 '16 at 14:24
  • Why did you name the function rag/... ? – Boris Stitnicky Dec 7 '16 at 15:03
  • It's a common style that people often use when writing their own functions(you can see it being done by a lot others if you browse their emacs config). All the functions I defined begin with rag/ prefix and that makes them easy to find under M-x :) – Chakravarthy Raghunandan Dec 7 '16 at 16:33
  • Ic, that's your name :-) – Boris Stitnicky Dec 8 '16 at 9:56
2

Here is an example of a custom grep function that recursively greps file contents (using a regexp for the search term) in a directory and its subdirectories, and displays the results with lines of context before and after. The built-in compile.el and grep.el libraries provide for several methods to jump to the location in the file containing the search results. There is no need to install anything else for this example to work -- all that is needed is a full version of Emacs and a computer that has the grep utility installed. The variable grep-program can be adjusted to point to a particular location of the grep utility if the default is not sufficient.

(defun recursive-grep ()
"Recursively grep file contents.  `i` case insensitive; `n` print line number;
`I` ignore binary files; `E` extended regular expressions; `r` recursive"
(interactive)
  (let* ((search-term (read-string "regex:  "))
         (search-path
           (directory-file-name (expand-file-name (read-directory-name "directory:  "))))
         (default-directory (file-name-as-directory search-path))
         (grep-command
           (concat
             grep-program
             " "
             "-inIEr --color=always -C2"
             " "
             search-term
             " "
             search-path)))
    (compilation-start grep-command 'grep-mode (lambda (mode) "*grep*") nil)))
Improve this answer
  • While the question does not seek how to simultaneously modify multiple files that are returned as results by the above grep function, I find it very helpful to use multiple-cursors and wgrep. It makes modifying multiple files in one fell swoop a breeze. Said libraries, however, would need to be installed if those features are desired. – lawlist Aug 17 '16 at 17:11
  • What's the advantage of using this over the builtin rgrep? – npostavs Aug 17 '16 at 17:38
  • 1
    @npostavs -- I found the built-in rgrep to be time-consuming to customize to my preferred regex search criteria, and I chose to hard-code everything in a custom function (which I use several times each day). If you would like to write-up an alternative answer that describes how to customize rgrep, that would be helpful to other forum readers in the future. – lawlist Aug 17 '16 at 17:41

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.