cscope or ctags why choose one over the other?

Question

I primarily use vim / gvim as an editor and am looking at using a combination of lxr ( the Linux Cross Reference ) and either cscope or ctags for exploring the kernel source. However I haven't ever used either cscope or ctags and would like to hear why one might choose one over the other taking into consideration my use of vim as a primary editor.

Answer 1

ctags enables two features: allowing you to jump from function calls to their definitions, and omni completion. The first means that when you are over a call to a method, hitting g] or CTRL-] will jump to the place where that method is defined or implemented. The second feature means that when you type foo. or foo->, and if foo is a structure, then a pop-up menu with field completion will be shown.

cscope also has the first feature - using set cscopetag - but not the last. However cscope additionally adds the ability to jump to any of the places where a function is called as well.

So as far as jumping around a code base is concerned, ctags will only ever lead you towards the place where the function is implemented, whereas cscope can show you where a function is called too.

Why would you choose one over the other? Well, I use both. ctags is easier to set up, faster to run and if you only care about jumping one way it will show you less lines. You can just run :!ctags -R . and g] just works. It also enables that omni complete thing.

Cscope is great for bigger, unknown code bases. The set up is a pain because cscope needs a file containing a list of names of files to parse. Also in vim, by default there are no key bindings set up - you need to run :cscope blah blah manually.

To solve the fist problem I've got a bash script cscope_gen.sh that looks like this:

#!/bin/sh
find . -name '*.py' \
-o -name '*.java' \
-o -iname '*.[CH]' \
-o -name '*.cpp' \
-o -name '*.cc' \
-o -name '*.hpp'  \
> cscope.files

# -b: just build
# -q: create inverted index
cscope -b -q

This searches for code that I'm interested in, creates the cscope.files list and creates the database. That way I can run ":!cscope_gen.sh" instead of having to remember all the set up steps.

I map cscope search to ctrl-space x 2 with this snippet, which mitigates the other downer of cscope:

nmap <C-@><C-@> :cs find s <C-R>=expand("<cword>")<CR><CR>

There's this cscope_maps.vim plugin that sets up a bunch of similar bindings. I can never remember what all the options mean, so tend to stick to ctrl-space.

So to conclude: ctags is easier to set up and mostly works without doing much else, it's vital for omni-complete too. cscope provides more features if you have to maintain a large and mostly unknown code base, but requires more leg work.

Answer 2

I was in the same situation some months ago...

The lack of precision of ctags is a pain in a.., and i find cscope much better for all the macros related stuff (and there are a bunch of macros in the linux kernel)..

concerning the usage, that's actually straightforward...you just type cscope -R at the root of your kernel and then you've got nothing to worry about.. (I mean if you just want to explore that's perfect...)

Then, the key bindings are all based on Ctrl-\ (you can remap it if you're allergic to Ctrl), you mainly use s and g....,

Developing for the kernel, I didn't need so much the completion....

Anyway, go for cscope, this is much more convenient, accurate.

Answer 3

Hmm... You should probably use etags instead of ctags...

If you use cscope, then you can see call chains, i.e., who calls this function & which functions does this function call?

I am not sure if this can be done using etags / ctags...

That's just one feature... what about finding out the file that contains a particular function definition? This you get only in cscope.

I use both cscope and etags, they are both good for different things, especially when working with a large codebase, such as the Linux Kernel. In fact, I started using cscope and etags when I started working with the Linux Kernel / Xen.

LXR is not great, because you have to click, go over the network etc., whereas you can build the cscope and tags databases on your kernel code and do not have to go over the network (unlike lxr).

Answer 4

by a quick scan of the source code, it seems that ctag's lex is written by hand, while cscope's is generated by flex, am I correct?