Меню:


This article doesn't pretend to be detailed description on how to setup Emacs to be complete development environment. I just tried to provide a small description on "How to setup CEDET to work with C, C++ & Java", although most of this description will be also applicable for other languages, supported by CEDET1.

What is CEDET?

The CEDET package is a collection of libraries, that implement different commands, but all of them have common goal — provide functionality for work with source code written in different programming languages (please, take into account that not all of these packages are included into CEDET bundled with GNU Emacs):

CEDET's versions

We need to understand that there are several versions of CEDET, that differ from each other by installation & customization methods:

This article describes new version of CEDET, including version bundled with GNU Emacs. If you want to use standalone release 1.1, or earlier, then you need to read previous version of this article.

Installation of standalone CEDET's version

Right now (October, 2012) there is no released version of CEDET with "new" activation method, so you need to take current snapshot, or get version from repository. Unpack source code, if this is necessary, change to directory & perform following command that will compile everything:

make clean-all && make

Customization

If you're using standalone CEDET's version, then you need to load it with following command:

(load-file "~/emacs/cedet-bzr/cedet-devel-load.el")

But if you're using CEDET bundled with GNU Emacs, then everything will be already loaded on start.

Semantic's customization

All standalone versions until release 1.1 (including) had activation method that was different from method for CEDET bundled into GNU Emacs. In new versions activation of package is performed by adding of symbols for selected sub-modes into special list, and then activating semantic-mode. While in the "old" versions, functionality was activated by calling one of the functions, each of them activated some specific set of features. So in the "new" version, it's enough to do following:

(semantic-mode 1)

And set of enabled features will depend on what will be put into list semantic-default-submodes, that can contain following symbols (it's better to populate this list before enabling semantic-mode):

global-semanticdb-minor-mode
enables global support for Semanticdb;
global-semantic-mru-bookmark-mode
enables automatic bookmarking of tags that you edited, so you can return to them later with the semantic-mrub-switch-tags command;
global-cedet-m3-minor-mode
activates CEDET's context menu that is bound to right mouse button;
global-semantic-highlight-func-mode
activates highlighting of first line for current tag (function, class, etc.);
global-semantic-stickyfunc-mode
activates mode when name of current tag will be shown in top line of buffer;
global-semantic-decoration-mode
activates use of separate styles for tags decoration (depending on tag's class). These styles are defined in the semantic-decoration-styles list;
global-semantic-idle-local-symbol-highlight-mode
activates highlighting of local names that are the same as name of tag under cursor;
global-semantic-idle-scheduler-mode
activates automatic parsing of source code in the idle time;
global-semantic-idle-completions-mode
activates displaying of possible name completions in the idle time. Requires that global-semantic-idle-scheduler-mode was enabled;
global-semantic-idle-summary-mode
activates displaying of information about current tag in the idle time. Requires that global-semantic-idle-scheduler-mode was enabled.

Following sub-modes are usually useful when you develop and/or debug CEDET:

global-semantic-show-unmatched-syntax-mode
shows which elements weren't processed by current parser's rules;
global-semantic-show-parser-state-mode
shows current parser state in the modeline;
global-semantic-highlight-edits-mode
shows changes in the text that weren't processed by incremental parser yet.

This approach allows to make Semantic's customization more flexible, as user can switch on only necessary features. You can also use functions with the same names to enable/disable corresponding sub-modes for current Emacs session. And you can also enable/disable these modes on the per-buffer basis (usually this is done from hook): names of corresponding variables you can find in description of global-semantic-* functions.

To enable more advanced functionality for name completion, etc., you can load the semantic/ia with following command:

(require 'semantic/ia)

After loading of this package, you'll get access to commands, described below.

System header files

To normal work with system-wide libraries, Semantic should has access to system include files, that contain information about functions & data types, implemented by these libraries.

If you're using GCC for programming in C & C++, then Semantic can automatically find directory, where system include files are stored. Just load semantic/bovine/gcc package with following command:

(require 'semantic/bovine/gcc)

You can also explicitly specify additional directories for searching of include files (and these directories also could be different for specific modes). To add some directory to list of system include paths, you can use the semantic-add-system-include command — it accepts two parameters: string with path to include files, and symbol, representing name of major mode, for which this path will be used. For example, to add Boost header files for C++ mode, you need to add following code:

(semantic-add-system-include "~/exp/include/boost_1_37" 'c++-mode)

Although I want to say, that customization for Boost support is more complex, and requires to specify where Semantic can find files with constant's definitions, etc.

Semantic's work optimization

To optimize work with tags, you can use several techniques:

(setq-mode-local c-mode semanticdb-find-default-throttle
                 '(project unloaded system recursive))

Semantic extracts syntactic information when Emacs is idle. You can customize the semantic-idle-scheduler-idle-time variable to specify idle time (in seconds), if you don't want to use default value (1 second).

Integration with imenu

The Semantic package can be integrated with the imenu package. This lead to creation of a menu with a list of functions, variables, and other tags. To enable this feature you need to add following code into your initialization file:

(defun my-semantic-hook ()
  (imenu-add-to-menubar "TAGS"))
(add-hook 'semantic-init-hooks 'my-semantic-hook)

Customization of Semanticdb

To enable Semanticdb you need to add the global-semanticdb-minor-mode symbol into the semantic-default-submodes list. And you can specify some customization variables to control behaviour of Semanticdb — for example, where to save data, etc. These variables could be set via semanticdb customization group.

Semanticdb can also use databases generated by external utilities: gtags (from GNU Global), exubertant ctags, ebrowse & cscope. To activate this functionality you can use following code (please, note that these commands may fail if you have no utilities installed, or have an incorrect versions of them — that's why they a wrapped into when):

;; if you want to enable support for gnu global
(when (cedet-gnu-global-version-check t)
  (semanticdb-enable-gnu-global-databases 'c-mode)
  (semanticdb-enable-gnu-global-databases 'c++-mode))

;; enable ctags for some languages:
;;  Unix Shell, Perl, Pascal, Tcl, Fortran, Asm
(when (cedet-ectag-version-check t)
  (semantic-load-enable-primary-exuberent-ctags-support))

EDE's customization

If you plan to use projects, then you need to enable corresponding mode, implemented by the EDE package:

(global-ede-mode t)

There are several types of projects supported by EDE, and I want to describe here only some of them.

Using EDE for C & C++ projects

For correct work of Semantic with С & C++ code it's recommended to use the EDE package (it allows to work with projects, etc.). For these languages, EDE package defines special project type: ede-cpp-root-project, that provides additional information to Semantic, and this information will be used to analyze source code of your project.

To define a project, you need to add following code:

(ede-cpp-root-project "Test"
                :name "Test Project"
                :file "~/work/project/CMakeLists.txt"
                :include-path '("/"
                                "/Common"
                                "/Interfaces"
                                "/Libs"
                               )
                :system-include-path '("~/exp/include")
                :spp-table '(("isUnix" . "")
                             ("BOOST_TEST_DYN_LINK" . "")))

For the :file parameter you can use any file at root directory of your project. This file isn't parsed — it's used only as an anchor to search all other files in project.

To search include files, Semantic uses directories from two lists, that could be specified for project. The :system-include-path parameter is used to specify list of full paths where lookup for "system" include files will be performed. Another parameter — :include-path specifies the list of directories, that will be used to search of "local" include files (if names are starting with /, this means, that path is specified relative to project's root directory). Instead of specifying paths as lists, you can also provide function, that will perform search of include files in your project. You can read about it in the EDE manual.

Another parameter, that could be specified in project's declaration is a list of definitions, that will be used during code preprocessing. The :spp-table parameter allows to specify list of pairs, consisting from symbol's name & value, defined for given symbol. In our example above, we defined two symbols — isUnix and BOOST_TEST_DYN_LINK, that will be passed to preprocessor, and this will allow to perform proper parsing of the code.

User, if required, can redefine some variables for files inside project. This could be done by specifying the :local-variables parameter with value that is a list of pairs in form symbol name/value, and these values will be set for files in project.

Preprocessing of source code

More information about definitions for C/C++ preprocessor you can find in documentation for the semantic-lex-c-preprocessor-symbol-map variable. You can obtain list of preprocessor symbols, defined for file with source code, by using the semantic-lex-spp-describe command. And then use these results to set :spp-table parameter or semantic-lex-c-preprocessor-symbol-map variable.

Many libraries store all macro definitions in one or more include files, so you can use these definitions as-is. To do this you need to specify these files in the semantic-lex-c-preprocessor-symbol-file variable, and when CEDET will perform analysis, then values from these files will be used. By default, this variable has only one value — file with definitions for C++ standard library, but you can add more data there. As example, I want to show CEDET's configuration for work with Qt4:

(setq qt4-base-dir "/usr/include/qt4")
(semantic-add-system-include qt4-base-dir 'c++-mode)
(add-to-list 'auto-mode-alist (cons qt4-base-dir 'c++-mode))
(add-to-list 'semantic-lex-c-preprocessor-symbol-file (concat qt4-base-dir "/Qt/qconfig.h"))
(add-to-list 'semantic-lex-c-preprocessor-symbol-file (concat qt4-base-dir "/Qt/qconfig-dist.h"))
(add-to-list 'semantic-lex-c-preprocessor-symbol-file (concat qt4-base-dir "/Qt/qglobal.h"))

After you'll add these lines to initialization file, you should be able to use names completion for classes, defined in Qt4 library. Example you can see on the picture below:

Using EDE for Java projects

Semantic includes a parser for source code written in Java, so name completion for source code always worked, and the main problem was to get name completions for classes from JDK, or other libraries that are used in project. For compiled code, Semanticdb can get information about name by using javap on the list libraries in the CLASSPATH. To make it working, you need to load the semantic/db-javap package:

(require 'semantic/db-javap)

The path to the JDK's main library (rt.jar on Linux & Windows, and classes.jar on Mac OS X) is usually detected automatically by the cedet-java-find-jdk-core-jar function, although you can change its behaviour by setting JAVA_HOME environment variables, or some other parameters.

If you're using Maven to build your projects, then CLASSPATH will be calculating automatically by running Maven in the root of your project (also for multi-module projects). And it isn't necessary to specify project's root manually — EDE will find it automatically by searching for the pom.xml file. I need to mention that first call to name completion functions could be relatively slow — EDE should run Maven and collect information about libraries that are used in the project. But after first run, this information is cached, and next completions will be performed faster.

If you aren't using Maven, then you can either specify all used libraries in the semanticdb-javap-classpath variable, or use the ede-java-root-project class, that is similar to ede-cpp-root-project that was described above. To use this type of project, you need to add something like to you initialization file:

 (ede-java-root-project "TestProject"
         :file "~/work/TestProject/build.xml"
         :srcroot '("src" "test")
         :localclasspath '("/relative/path.jar")
         :classpath '("/absolute/path.jar"))

As for C/C++, you need to specify name of the project, point to existing file at the project's root directory, and some additional options:

:srcroot
list of directories with source code. Directory names are specified relatively of project's root (in this example this is src & test);
:classpath
list of absolute file names for used libraries;
:localclasspath
list of file names for used libraries, relative to project's root.

When Semantic finds such project, it can use provided information for name completion.

Work with Semantic

From user's point of view, Semantic provides several major features — names completions, retrieving information about tags (variables, functions, etc.), and navigation in source code. Some of these features are implemented by semantic/ia package, while other are implemented by Senator, and Semantic's kernel.

Some of commands have no standard key bindings, so it's better to select key bindings, that are comfortable to you, and bind commands to them, like this (only for standalone CEDET):

(defun my-cedet-hook ()
  (local-set-key [(control return)] 'semantic-ia-complete-symbol)
  (local-set-key "\C-c?" 'semantic-ia-complete-symbol-menu)
  (local-set-key "\C-c>" 'semantic-complete-analyze-inline)
  (local-set-key "\C-cp" 'semantic-analyze-proto-impl-toggle))
(add-hook 'c-mode-common-hook 'my-cedet-hook)

I want to mention, that Semantic's development is pretty active, and if something doesn't work, or works wrong, then please, send examples of code to the cedet-devel mailing list — the CEDET's authors usually answers pretty fast.

Names completion

Text completion for names of functions, variables & classes is pretty often used feature when you work with source code2. There are two packages inside Semantic that implement this functionality — semantic/ia and Senator (it doesn't included into GNU Emacs). You need to take into account, that in the new versions it's recommended to use Semantic only as source of information, and perform names completion using other packages, such as auto-complete. You see example below.

Commands, implemented by semantic/ia use the semantic-analyze-possible-completions function to create a list of all possible names completion, and this function takes into account many parameters (plus it can be augmented by user's code to provide more precise list of names). At the same time, commands from Senator package use simpler methods to create a list of all possibles completions (usually they use information only about definitions in the current file), and this sometime lead to wrong names completion.

If you execute the semantic-ia-complete-symbol command when you're typing code, then this will lead to completion of corresponding name — name of function, variable, or class member, depending on the current context. If there are several possible variants, then this name will be completed to most common part, and if you'll call this command second time, then buffer with all possible completions will be shown. User can also use the semantic-ia-complete-symbol-menu command — it also performs analysis of current context, but will display list of possible completions as a graphical menu, from which the required name should be selected3. Besides this, there is semantic-ia-complete-tip command, that displays list of possible completions as tooltip.

As was mentioned above, the Senator package, also provides commands for names completion. It work very fast, but with less precision (as they use few parameters during computation of variants for completions). The senator-complete-symbol command (C-c , TAB) completes name for current tag, and insert first found completion as result. If it inserts wrong name, then you can insert second name from completion list by repeating this command, and so on. If there are a lot of the possible variants, or you want to see full list of functions and variables for some class, then it's better to use the senator-completion-menu-popup command (C-c , SPC) — it displays list of all possible completions as a graphical menu.

Besides these commands, user can use special mode (only for some languages) — semantic-idle-completions-mode (or enable it globally by adding global-semantic-idle-completions-mode symbol into semantic-default-submodes list) — in this mode names completions are shown automatically if user stops its work for a some time (idle time). By default, only first possible completion is shown, and user can use the TAB key to navigate through list of possible completions.

For C-like languages, user can use the semantic-complete-self-insert command, bound to the . and/or > keys, as this shown below:

(defun my-c-mode-cedet-hook ()
 (local-set-key "." 'semantic-complete-self-insert)
 (local-set-key ">" 'semantic-complete-self-insert))
(add-hook 'c-mode-common-hook 'my-c-mode-cedet-hook)

Evaluation of this code will lead to execution of the semantic-complete-self-insert command when user will press . or > after variables, that are instances of some data structure, and displaying a list of possible completions for given class or structure.

If you're programming in C & C++, then you can also get name completions using information from Clang (versions 2.9 & above). To do this, you need to load the semantic/bovine/clang package, and call the semantic-clang-activate function. After that, Semantic will start to call Clang, and use its code analyzer to calculate list of possible names completions.

Names completion with auto-complete package

The auto-complete package was developed to automatically complete text using information from different sources: predefined dictionaries, text from current buffer, external programs (GNU Global, etags, ...), etc. Semantic could be also used as source of information.

This package is available in the GNU Emacs's package repository (execute M-x package-list-packages to get list of available packages), but you can also install it via el-get or manually. Installation & customization are described in user's manual, so I won't cover these parts.

To use information from Semantic for names completion, you need to add ac-source-semantic or ac-source-semantic-raw into ac-sources list (this list allows you to control which data sources will be used for current buffer, so you can change it as you want). ac-source-semantic-raw differs from ac-source-semantic that for it the filtering of information isn't performed.

So your setup can be performed following way — instead of bounding keys for semantic-ia-complete-symbol-menu, semantic-ia-complete-symbol & other functions, you can simply add new names completion sources, and after that use auto-complete bindings to get names completion:

(defun my-c-mode-cedet-hook ()
  (add-to-list 'ac-sources 'ac-source-gtags)
  (add-to-list 'ac-sources 'ac-source-semantic))
(add-hook 'c-mode-common-hook 'my-c-mode-cedet-hook)

Getting information about tags

The semantic/ia package provides several commands, that allow to get information about classes, functions & variables (including documentation from Doxygen-style comments). Currently following commands are implemented:

semantic-ia-show-doc
shows documentation for function or variable, whose names is under point. Documentation is shown in separate buffer. For variables this command shows their declaration, including type of variable, and documentation string (if it's available). For functions, prototype of the function is shown, including documentation for arguments and returning value (if comments are available);
semantic-ia-show-summary
shows documentation for name under point, but information is shown in the mini-buffer, so user will see only variable's declaration or function's prototype;
semantic-ia-describe-class
asks user for a name of the class, and return list of functions & variables, defined in given class, plus all its parent classes.

Navigation in source code

One of the most useful commands for navigation in source code is the semantic-ia-fast-jump command, that allows to jump to declaration of variable or function, whose name is under point. You can return back by using the semantic-mrub-switch-tag command (C-x B), that is available when you enable the semantic-mru-bookmark-mode minor mode.

Semantic also provides two additional commands for jumping to function or variable: defined in current file — semantic-complete-jump-local (C-c , j), or defined in current project — semantic-complete-jump (C-c , J). Both commands allow to enter name of function or variable (including local variables inside functions) and jump to given definition (you can use name completion when entering the name).

The main difference between semantic-ia-fast-jump & semantic-complete-jump commands is that the first properly handles complex names, like this::that->foo(), while the second, can find only simple names, like foo.

The semantic-analyze-proto-impl-toggle command allows to switch between function's declaration and its implementation in languages, that allow to have separate declaration and implementation of functions. Another useful command is semantic-decoration-include-visit, that allows to jump to include file, whose name is under cursor.

Senator provides several commands for navigation in source code. This is senator-next-tag (C-c , n) and senator-previous-tag (C-c , p) commands, that move cursor to next or previous tag. There is also the senator-go-to-up-reference command (C-c , u), that moves cursor to the "parent" tag (for example, for class member function, "parent" tag is class declaration).

Search for places where function is called

Semantic also has very useful command — semantic-symref, that allows to find places, where symbol (whose name is under point) is used in your project. If you want to find use of symbol with arbitrary name, then you can use the semantic-symref-symbol command, that allows to enter name of the symbol to lookup.

If references to given name weren't found in corresponding database (GNU Global, etc.), then these commands will try to find them using the find-grep command. As result of execution of these commands, a new buffer with results will be created, and user can jump to found places:

Source code folding

As Semantic has almost complete syntactic information about source code, this allows it to implement folding functionality, similar to functionality implemented by hideshow package. To enable this feature, you need to perform customization of the global-semantic-tag-folding-mode variable. When you'll enable it, this will lead to displaying of small triangles at the fringle field, and you will able to fold and unfold pieces of code by clicking on them (this should work not only for source code, but also for comments, and other objects).

Senator also has similar functionality, but it's usually used for top-level objects — functions, class declarations, etc. You can fold piece of code with the senator-fold-tag command (C-c , -), and unfold it with senator-unfold-tag (C-c , +).

More Senator's commands

The Senator package provides number of commands for work with tags, that allow user to cut or copy tag, and insert it in another place. To cut current tag (usually this is declaration of some function, or its implementation) the senator-kill-tag command (C-c , C-w) should be used. You can insert complete tag with standard key binding C-y, while the senator-yank-tag command (C-c , C-y) inserts only tag declaration, without body. Another useful command is senator-copy-tag (C-c , M-w), that copies current tag — this is very handy when, for example, you want to insert declaration of function into include file.

Senator allows to change behaviour of standard search commands (re-search-forward, isearch-forward and other), when you work with source code, such way, so they will perform search only in the given tags. To enable this mode you can use the senator-isearch-toggle-semantic-mode command(C-c , i), and with the senator-search-set-tag-class-filter command (C-c , f) you can limit search to given tag types — function for functions, variable for variables, etc.

You can also perform tags search without enabling this mode — you just need to call corresponding command: senator-search-forward or senator-search-backward.

Work with Srecode

The Srecode package allows user to define code templates, but it differs from other packages, that provide insertion of templates, because list of available templates can vary depending on the current context. For example, insertion of get/set functions should happen only when you inside class declaration. Or, insertion of function's declaration, may happen only outside of other function.

The main command, that is used to insert templates, is the srecode-insert, that is bound to the C-c / / keys. This command will ask user for template's name (you can enter it using name completion). List of available templates will vary, depending on the current context. If you want to insert the same template once again, then you can use the srecode-insert-again command (C-c / .).

Templates that are defined by user, can also use they own key bindings. They can use lower-case symbols from range C-c / [a..z], and user can specify in template's definition, which key will be assigned to it. For example, for C++ you can use the C-c / c key binding to insert class declaration.

Key bindings, that use upper-case symbols, are reserved for templates & commands defined in Srecode. For example, C-c / G (srecode-insert-getset) inserts pair of functions get/set for some class member variable, while C-c / E (srecode-edit) is used to edit templates. List of these commands isn't constant, so you need to look into documentation to find actual list of commands.

Besides templates, supplied with CEDET, user can define their own templates, and store them in the ~/.srecode directory, where CEDET will find them automatically. You can read about template's creation in the Srecode manual, that comes together with other documentation in the CEDET distribution.

Additional packages

Together with CEDET the number of additional packages is supplied. Some of them are located in the contrib subdirectory, that you need to add to library search list.

The eassist package

The eassist package provides several commands, that use information obtained from Semantic. By default, these commands have no predefined key bindings, so you need to select them yourself.

The eassist-list-methods command, executed in the file with source code, will show you a list of functions, defined in current buffer, and will allow you to perform quick jump to selected function.

If you develop code in C and/or C++ languages, then the eassist-switch-h-cpp command, could be very useful to you — it jumps between header file and file, that contains implementation (if they have same names, but different extensions).


1. You can use following configuration files as a base for your configuration: standalone versions up to 1.1, standalone version after 1.1 release & bundled with GNU Emacs.

2. There is also semantic-complete-analyze-inline command, that shows list of all possible completions in separate window, that is often handy than graphical menu

3. If name completion works improperly, then try to analyze why this happens, and send bug report only after this. Information about debugging you can find in Semantic User Guide in section Smart Completion Debugging

Last change: 12.03.2014 07:58

blog comments powered by Disqus