Re: [cedet-semantic] Semantic search and intellisense completion.

[Hadron <had...@go...>]

"Jamie Macbeth" <jam...@gm...> writes:

> Hi, Hadron, Eric, and others
>
> I went through some of the same difficulty with semantic intellisense
> completion a couple of months ago.  i hope that you guys don't mind me
> putting my two cents in.  please correct me on any of these points if
> i'm wrong
>
> first, semantic does require that you include the header file of
> interest in order to get completion on any symbols.  to me this seems

The file must be re-opened too which is a pain to be honest. Eric has
indicated he might change this in the future. I would sooner wait for 10
seconds when I open a C program than have to manually track and open
each header file which includes values I wish to complete on.

> to be a nice optimization.  by default semantic performs a recursive
> search through included header files to collect tags for completion,
> so if your fstab.h header file includes other header files that are
> needed to perform the completion they will be searched.

A recent correspondence suggests that some of the clever c wizardry in
the Linux headers might just not be handled properly at the moment by
semantic parser :( (What *do* the Linux app programmers use for
completion etc ?). Surely not that java horror Eclipse :-; ?

>
> second, and way more important, usually semantic needs to be able to
> find the header file itself in the filesystem before it gets it's
> tags.  it needs the full path to the header before it tries to find
> the tag table for the header in any database.  semantic stores tag
> caches on a directory by directory basis, so it wants the full path to
> the file to know which directory cache to look in.
>
> for system databases, i found that i needed to add the path in
> manually before creating the database.  i put this in my .emacs.
>
> (defun semantic-c-mode-hook ()
> (semantic-add-system-include "/path/to/header/files"))
> (add-hook 'c-mode-common-hook 'semantic-c-mode-hook)
> (add-hook 'c-mode-hook 'semantic-c-mode-hook)

I pointed this to the root of my (local) Linux header hierarchy and
still no joy.

>
> one easy way to make sure that semantic is indeed caching tags and
> creating system databases correctly is by looking for a file called
> "semantic.cache"  in the directory of interest.  it is a text file and
> you can look through it.  by the way, if you try to create a database
> where you don't have permission to write a file i think it will fail
> ... i always put this in my .emacs.
>
> (setq semanticdb-default-save-directory "/home/directory/.semanticdb")

Yup - I had that since at one stage I didnt keep a local copy.

>
> and it stores all of the caches there.
>
> finally, if the "unloaded" flag is not set in the
> semanticdb-find-default-throttle, semantic will not automatically open
> a header file that it has never seen, even if it has the path to it.

I believe from earlier comments from Eric that this is broken.

I have a situation at the moment where even if I open the file of
interest (in my local Linux header hierarchy) it is not parsing and
creating a cache file which makes me wonder if I have to add each and
every directory to the system-include?

The paths confuse me. I see no real reason to differentiate between
"system headers" and "project headers". Simply have a set of paths each
of which is the root of a hierarchy to search - and an option to follow
symbolic links. I would say that how semantic locates files is the most
difficult for a noob at the moment. I am pretty sure I read something
somewhere in the docs that it ignores certain system files too - but I
cant find it again.

>
> Jamie