Faster alternatives to "find" and "locate"?

Question

I will like to use "find" and locate" to search for source files in my project, but they take a long time to run. Are there faster alternatives to these programs I don't know about, or ways to speed up the performance of these programs?

Answer 1

Searching for source files in a project

Use a simpler command

Generally, source for a project is likely to be in one place, perhaps in a few subdirectories nested no more than two or three deep, so you can use a (possibly) faster command such as

(cd /path/to/project; ls *.c */*.c */*/*.c)

Make use of project metadata

In a C project you'd typically have a Makefile. In other projects you may have something similar. These can be a fast way to extract a list of files (and their locations) write a script that makes use of this information to locate files. I have a "sources" script so that I can write commands like grep variable $(sources programname).

Speeding up find

Search fewer places, instead of find / … use find /path/to/project … where possible. Simplify the selection criteria as much as possible. Use pipelines to defer some selection criteria if that is more efficient.

Also, you can limit the depth of search. For me, this improves the speed of 'find' a lot. You can use -maxdepth switch. For example '-maxdepth 5'

Speeding up locate

Ensure it is indexing the locations you are interested in. Read the man page and make use of whatever options are appropriate to your task.

   -U <dir>
          Create slocate database starting at path <dir>.

   -d <path>
          --database=<path> Specifies the path of databases to search  in.


   -l <level>
          Security  level.   0  turns  security checks off. This will make
          searchs faster.   1  turns  security  checks  on.  This  is  the
          default.

Remove the need for searching

Maybe you are searching because you have forgotten where something is or were not told. In the former case, write notes (documentation), in the latter, ask? Conventions, standards and consistency can help a lot.

Answer 2

I used the "speeding up locate" part of RedGrittyBrick's answer. I created a smaller db:

updatedb -o /home/benhsu/ben.db -U /home/benhsu/ -e "uninteresting/directory1 uninteresting/directory2"

then pointed locate at it: locate -d /home/benhsu/ben.db

Answer 3

For a find replacement, check out fd. It has a simpler / more intuitive interface than the original find command, and is quite a bit faster.

Answer 4

As of early 2021 I evaluated a few tools for similar use cases and integration with fzf:

tl;dr: plocate is a very fast alternative to mlocate in most cases. (GNU) find often still seems to be hard to beat for tools which don't require an index.

locate-likes (using indices)

I compared the commonly used mlocate and plocate. In a database of about 61 million files plocate answers specific queries (a couple of hundred results) in the order of 0.01 to 0.2 seconds and only becomes much slower (> 100 seconds) for very unspecific queries with millions of results. mlocate takes an almost constant 35 to 40 seconds to query the same database in all tested cases. Most of the time plocate is multiple orders of magnitude faster than mlocate.

find-likes (directory traversal)

It's harder to - sorry - locate a good find alternative and the result very much depends on the specific queries. Here are some results for just trying to find all files in a sample directory (~200.000 files) in descending speed order:

• GNU find (find -type f): Surprisingly, find is often the fastest tool (~ 0.01 seconds)
• plocate (plocate --regexp "^$PWD") (< 0.1 seconds)
• mlocate (mlocate --regexp "^$PWD"): Interestingly faster than the non-regexp queries (~ 5 seconds)
• ripgrep (rg --files --no-ignore --hidden): A good bit faster, but still very slow (~0.5 seconds)
• fd (fdfind --type f on Debian): Not as fast as expected (~0.7 seconds)
• zsh globs (**/* and extended versions): Generally very slow, but can be elegant in scripts (~3 seconds)

With directory structures not cached in RAM, plocate probably beats find, especially on disks with high seek times (HDDs), but does of course require an up-to-date database.

A parallelized find might give better results on systems which profit from command queuing and can request data in parallel. Depending on use case, this can be implemented in scripts, too, by running multiple instances of find on different subtrees, but performance characteristics depend on a lot of factors there.

grep-likes (looking at content)

There's a whole lot of recursive grep-like tools with different features, but most offer very limited features for finding files based on their metadata instead of their contents. Even if they are fast like ag (silversurfer), rg (ripgrep) or ugrep, they are not necessarily fast if just looking at file names.

Answer 5

A tactic that I use is to apply the -maxdepth option with find:

find -maxdepth 1 -iname "*target*"

Repeat with increasing depths until you find what you are looking for, or you get tired of looking. The first few iterations are likely to return instantaneously.

This ensures that you don't waste up-front time looking through the depths of massive sub-trees when what you are looking for is more likely to be near the base of the hierarchy.

---

Here's an example script to automate this process (Ctrl-C when you see what you want):

(
TARGET="*target*"
for i in $(seq 1 9) ; do
   echo "=== search depth: $i"
   find -mindepth $i -maxdepth $i -iname "$TARGET"
done
echo "=== search depth: 10+"
find -mindepth 10 -iname $TARGET
)

Note that the inherent redundancy involved (each pass will have to traverse the folders processed in previous passes) will largely be optimized away through disk caching.

Why doesn't find have this search order as a built-in feature? Maybe because it would be complicated/impossible to implement if you assumed that the redundant traversal was unacceptable. The existence of the -depth option hints at the possibility, but alas...

Answer 6

The Silver Searcher

You might found it useful for searching very fast the content of a huge number of source code files. Just type ag <keyword>. Here some of the output of my apt show silversearcher-ag:

• Package: silversearcher-ag
• Maintainer: Hajime Mizuno
• Homepage: https://github.com/ggreer/the_silver_searcher
• Description: very fast grep-like program, alternative to ack-grep The Silver Searcher is grep-like program implemented by C. An attempt to make something better than ack-grep. It searches pattern about 3–5x faster than ack-grep. It ignores file patterns from your .gitignore and .hgignore.

I usually use it with:

-G --file-search-regex PATTERN Only search files whose names match PATTERN.

ag -G "css$" important

Answer 7

Another easy solution is to use newer extended shell globbing. To enable:

• bash: shopt -s globstar
• ksh: set -o globstar
• zsh: already enabled

Then, you can run commands like this in the top-level source directory:

# grep through all c files
grep printf **/*.c

# grep through all files
grep printf ** 2>/dev/null

This has the advantage that it searches recursively through all subdirectories and is very fast.

Answer 8

If the file/directory you're looking for is likely to be "shallow" (ie. not that many layers deep), you can speed up your search a lot by using a breadth-first search tool like bfs.

If you know the target is not in a given directory, you can also exclude directories as shown here.