Fwd: A guide on setting up C/C++ development environment for Emacs

[Tu Hoang Do]

For C, CEDET works really well. For C++, it is a quite incomplete but
usable.  I already showed the smart completion in this screenshot:
http://tuhdo.github.io/static/auto_complete.gif

Notice in the screenshot before and after I include linux/printk.h  CEDET
realizes the change in buffer and adapt immediately. But, as I said, you
need to leave it sometimes for parsing the first time you perform smart
completion from included files. Here is another example with using CEDET
with Boost: http://tuhdo.github.io/static/c-ide/semantic-boost-demo.gif

CEDET actually understands your source code by trying to parse properly. It
works mostly with C.

please, try to M-x semantic-ia-complete-
> symbol-menu
> when you are in method implementation like this:
> void Class::method()
> {
>   // M-x semantic-ia-complete-symbol-menu should show a menu
>   // with all members in the scope. But nothing is shown.
>
}
>

For this use case, I agree this is a limitation. You need to have a prefix
for it to start completion. I think this should be less of a problem since
if you have many include files, you have a huge pool of completion
candidates that you have to narrow down anyway.


Second example. Please, try to M-x semantic-ia-complete-symbol-menu
> when you are in
> void Class::method()
> {
>   method2(
>   // M-x semantic-ia-complete-symbol-menu after the "(" will
>   // freeze Emacs to do some of stupid thinking.
>   // Terrible, horrible and sad. And nothing more.
> }
>

Parsing is an intensive task. To make CEDET work well, *you have to wait*
for it to biuld up the database for future use. For an implementation in
Pure Elisp, it is quite fast for reasonable size projects and is cross
platform. Getting Clang to work on Windows is not easy. Also, it is the
limitation of single threaded Emacs. To provides real IDE features like in
other IDEs, Emacs needs must be able to do many things at once, like
analyzing and checking. Since Emacs is single threaded, anything heavy
weight blocks it. You have to wait for a few minutes if you only include
like ten header files and your project is not too large. The way it works
is like this: for each included file, it parse the that file and if the
included file includes further files it will move deeply into those files
and parse until reaching the end.* Subsequent parsing results in less
time *because
CEDET can make use of previous parsing result. I heard that in the future
Emacs will come with cooperative threads, and it makes tasks like parsing
and syntax checking can coexist with editing.

After this, I don't want to use CEDET without any sorry.
> Sorry.
>

Sure. You always have alternatives such as other Clang solutions. But it
won't be complete either. It can complete candidates in system header files
fine, but if you want to complete for project wise, you have to give it
include paths written in absolute path.

Or, you can just use GNU Global with Emacs frontend that provides a not so
accurate for C++ but practical solution. It's really fast. I can walk the
Linux kernel with ease. You can use company-gtags to get a list of all
possible completions in your project.

Thanks.

On Wed, Aug 27, 2014 at 9:15 PM, Dmitriy Igrishin <dmitigr@gmail.com> wrote:

>
>
>
> 2014-08-27 17:18 GMT+04:00 Tu Hoang Do <solidius4747@gmail.com>:
>
> CEDET is not slow. It's reasonably fast given it is implemented in Emacs
>> Lisp. It needs time for parsing to build a database, so you see a message
>> printing this: parsing....<file> in the minibuffer. It needs parsing to
>> build up a database for smart completion. After the first time parsing,
>> completion happens instantly and you should enable Semanticdb minor mode to
>> save the parsing results, so CEDET won't have to parse the next time. CEDET
>> can handle project with the size of Emacs really well. I think even project
>> with more than a million lines of code. CEDET is the only viable solution
>> so far I found for real smart completion in Emacs. Other clang based
>> solution is pretty limited: you can only get completion from system header
>> and current directory, but not for your project. You have to add the
>> include paths to tell Clang where your project include paths are, and you
>> have to specify with absolute paths. It's really tedious.
>>
> No, it does not real smart. For example, please, try to
> M-x semantic-ia-complete-symbol-menu
> when you are in method implementation like this:
> void Class::method()
> {
>   // M-x semantic-ia-complete-symbol-menu should show a menu
>   // with all members in the scope. But nothing is shown.
> }
> Second example. Please, try to M-x semantic-ia-complete-symbol-menu
> when you are in
> void Class::method()
> {
>   method2(
>   // M-x semantic-ia-complete-symbol-menu after the "(" will
>   // freeze Emacs to do some of stupid thinking.
>   // Terrible, horrible and sad. And nothing more.
> }
>
> After this, I don't want to use CEDET without any sorry.
> Sorry.
>
> --
> // Dmitriy.
>
>