Extracting function names (was: [SVN] Bundle commit 636 (C))

[Allan Odgaard]

On Apr 22, 2005, at 0:44, Chris Thomas wrote:

> • Add a primitive declaration.function, partly based on the one Allan 
> seems to have left in there. We'll have to see whether this is good 
> enough for now or not. It's very much not perfect. You can't be 
> perfect unless you embed a C interpreter, but you can do better than 
> this. However, I think the only way to do a better job without a C 
> intepreter is to maintain the curlybracket nesting level.

Yes, one option for handling curly braces could be something like (root 
level patterns):
  1. match = function prototypes
  2. begin = "(namespace|class|struct).*{", end = "}", patterns = 
include source.c
  3. begin = {, end = }, patterns = all the normal stuff
  4. include patterns from previous rule

So rule 2 is recursive, rule 3 is also recursive but doesn't include 
from the root level (like rule 4), thus ditching rule 1.

The underlying system can handle this, but I haven't written the 
handler for the include directive yet, to support arbitrary paths -- 
most likely I'll have that written for beta 7 (main problem is just how 
to specify the path, the scopes are space-separated, but to remain 
backwards compatible, I led the include still allow one to specify the 
name of the syntax instead of the scopeName, which makes it ambiguous, 
since the name can have spaces -- actually though, I do consider 
requiring the UUID in the include statement, but maybe that's a bit too 
unfriendly).

One problem with the above is that if the C++ language includes C, it 
will lose the new rules it declares (inside curly braces). There also 
still is a problem with C++ initializer lists.

Another option is to have the C parser as a plugin, but I doubt many 
would write actual parsers even if I did add an API for this ;)

A third option is to only have functions optionally grabbed using the 
scope system. This will most likely be how it ends: i.e. I'll add a 
tool to TM to query documents using the scope system (or allow it 
through AS), and by default to grab the function names it'll just run: 
“tm_query 'declaration.function name'”, but it could instead be a grep 
or ctags.