25. Xref: Cross-referencing source code

When manipulating source code, programmers needs to know where the various symbols are defined, where they are used, and so on. This is generally available directly from their IDE and editors. But computing this information in the first place is tricky, especially for languages that support overloading of subprograms.

Some compilers like GNAT and gcc can generate this information for Ada code bases. For instance, GNAT will generate .ali files, which contain the navigation information, when compiling Ada or SPARK code.

GNATCOLL.Xref can then be used to parse and aggregate all those files into a single sqlite database, which can be conveniently used to answer queries such as “give me the declaration for this entity”, “list all places where this entity is used”, “show all subprograms that could be called in practice at this dispatching call”, “what files does this file depend on”, “show me the call graph for this application”,…

To use this package, some initialization needs to be performed first:

with GNATCOLL.Xref;     use GNATCOLL.Xref;
with GNATCOLL.SQL.Sqlite;
with GNATCOLL.Projects; use GNATCOLL.Projects;   --  1
with GNATCOLL.VFS;      use GNATCOLL.VFS;
with GNAT.Strings;

procedure Support is
   DB : Xref_Database;
   Tree : Project_Tree_Access := new Project_Tree;
   Error : GNAT.Strings.String_Access;

begin
   Tree.Load (Create ("prj.gpr"));  --  2
   Setup_DB
      (DB, Tree,
       GNATCOLL.SQL.Sqlite.Setup (Database => "testdb.db"),
       Error);  -- 3
   Free (Error);
   Parse_All_LI_Files (DB, Tree.Root_Project);   --  4
end Support;

GNATCOLL needs to be able to find the *li files. For this, it depends on project files (as supported by GNATCOLL.Projects). So the first thing to do is to parse the project (step 2).

We then need to tell GNATCOLL where the cross-reference information need to be aggregated. In this example, it will be stored in a sqlite database on the disk. By using a name “:memory:” instead, we would create a temporary in-memory database. This is in general faster, but uses more memory and needs to be recreated every time the program is restarted. We could also decide to store the information in any other database supported by GNATCOLL.SQL.Exec, for instance PostgreSQL.

Finally, in step 4 we let GNATCOLL parse all the *li files that are relevant for this project. This operation can take a while, depending on the size of the project. However, if the database already exists on the disk, it will simply be updated by parsing the files that are not already up-to-date. When all files are up-to-date, this operation is almost immediate.

At this point, we now have a database that we can start querying. Here are a few examples, but see the documentation gnatcoll-xref.ads for more types of queries. All these queries have a similar API: they return a cursor which iterates over the result returned by a SQL query. There are various kinds of cursors, depending on whether they return files, entities, or references to entities. But they all support the Has_Element, Element and Next operations, so all loops will look similar:

pragma Ada_05;   --  use object-dotted-notation
with GNATCOLL.VFS;   use GNATCOLL.VFS;

declare
    Entity : Entity_Information;
    Ref    : Entity_Reference;
    File   : Virtual_File;
    Refs   : References_Cursor;
begin
    File := Tree.Create ("source.ads");   --  5
    Ref := DB.Get_Entity ("Method", File, Line => 2);   --  6
    Entity := Ref.Entity;

    DB.References (Entity, Refs);   --  7
    while Refs.Has_Element loop
       Ref := Refs.Element;
       Put_Line (" at " & Ref.File.Display_Full_Name & ':'
          & Ref.Line'Img & ':' & Ref.Column'Img);
       Refs.Next;
    end loop;
end;

This example will print all the references to the entity that is referenced in file source.ads at line 2 (the column is unspecified).

Step 5 gets a handle on the source file. Here, we depend on the project to find the precise directory in which the source file is located. We can of course use an absolute file name instead.

Step 6 gets handle on the entity referenced on line 2 in this file. Such an entity is the starting point for most queries defined in GNATCOLL.Xref.

Finally, on step 7 and the loop below we iterate over all references, and print their location on the standard output.

Let’s do a much more complex query: we want to see all references to that entity, but also places where the entity might be called through a renames statement, or called through a dispatching call via an overriding method defined on a child tagged type (assuming this is a primitive operation of a tagged type in the first place). We also want to see all locations where a method that overrides “Method” is called:

declare
   Refs : Recursive_References_Cursor;
begin
   DB.Recursive (Entity, GNATCOLL.Xref.References'Access,
                 From_Overriding => True, From_Overridden => True,
                 From_Renames => True);
   while Refs.Has_Element loop
        ... same as before
        Refs.Next;
   end loop;
end;

As shown above, the programing pattern is always the same.

GNATCOLL.Xref provides many more subprogram to get information like the list of fields for a record type (or a C structure), the list of primitive operations or methods for a tagged object or a class, the call graph for a subprogram,…

It is also able to extract documentation for an entity from the source code, by looking at the lines of code just before or just after the declaration or the body of the entity.

26. Xref: gnatinspect

As discussed in the previous section, GNATCOLL provides an Ada API to perform cross-references queries.

There exist a few alternatives when you want to reuse that cross-reference information from other tools, or command line scripts.

You can of course access the sqlite database directly. Most programming languages have an interface to sqlite. For instance python does.

But GNATCOLL provides a command line tool dedicated to that purpose, named gnatinspect.

When it is first started on a project, this tool will refresh the xref database by parsing all the ALI files from the project. This might take a while (up to several minutes) the first time, unless of course the xref were already up-to-date because you had loaded the project in GPS first, or already run gnatinspect.

gnatinspect then displays an interactive prompt that lets you perform various queries on the database. The full list of queries is available by typing “help” at the prompt, but this documentation will demonstrate some of them.

Let’s first look at a number of command line switches that might be useful:

  • –db=ARG: this switch can be used to specify the name of the database.

    By default, gnatinspect checks in the project whether there exists an attribute IDE’Xref_Database, which should specify a file name (relative to the project’s object_dir) for the database.

    If this attribute does not exist, it defaults to “gnatinspect.db” in the project’s object directory.

    If there is no object directory defined in the project, the file is created in the project’s directory itself. You can however specify any name, including an absolute path, or a path relative to the project’s object directory.

    An alternative is to specify ‘:memory:’, which creates the database in memory. This is of course a temporary database which will disappear when gnatinspect exits, and cannot be shared with other tools.

  • –nightlydb=ARG: this switch can help speed up the initial startup of gnatinspect. The idea is that in a lot of cases, the software on which a team works is build nightly in a common setup. Running gnatinspect in that setup will create or update an xref database. Individual developers can then create their own copy of the database by starting from the contents of the nightly database (which is pointed to by the –nightlydb switch), and then gnatinspect will parse the ALI files in the user’s setup that are different from the nightly ones.

  • –runtime: by default, gnatinspect will only parse the ALI files from your project (and of course the ones from imported projects). It will not however parse the ALI files found in predefined directories, like for instance the GNAT runtime. This saves time in general. If you click on a call to one of the runtime subprograms in your own code, gnatinspect will be able to point you to its declaration. However, you will not have access to the body, because the link from declaration to body is found in the ALI files of the runtime.

  • –command=ARG: gnatinspect will update the xref database as usual, then execute a command, display its result, and exit. This can be convenient when calling gnatinspect from another tool, like Emacs or vi.

  • –file=ARG: similar to –command, but reads the commands to execute from a file. The file can contain comments (starting with ‘–’. See also the –lead switch.

  • –lang=LANG:SPEC:BODY:OBJ: specifies a naming scheme for a language. The preferred approach is to use a configuration project file (such as those generated by gprconfig for instance), that would define attributes such as Naming’Spec_Suffix, Naming’Body_Suffix and Compiler’Object_File_Suffix. However, this switch provides an alternative whereby you can specify the same values directly on the command line. For instance, the equivalent of:

    configuration project Autoconf is
       package Naming is
          for Spec_Suffix ("MyLang") use ".myl";
       end Naming;
       package Compiler is
          for Object_File_Suffix ("MyLang") use ".ali";
       end Compiler;
    end Autoconf;
    

    is to use:

    --lang=MyLang:.myl::.ali
    

    A third alternative is to have the same contents as the configuration project file above, directly in your own project file. This has the same effect, but needs to be duplicated in each of your project file.

    Given one of the above, and assuming your project file includes:

    for Languages use ("Ada", "MyLang");
    

    then any file with the .myl extension will be correctly detected by the project manager, and any .ali file with the same base name will be parsed by gnatinspect to find cross-reference information. Remember that the switch –config=autoconf.cgpr must be passed to gnatinspect if the information is provided via a config project file.

  • –lead=ARG should be used in coordination with –file, and specify lines to ignore from the file. All lines starting with the given prefix will be ignored.

  • –basenames: controls the display of file names in the output. By default, gnatinspect outputs full path information.

  • –exit: if this switch is specified, gnatinspect updates the xref database and exits immediately.

  • –project=ARG or -P ARG specifes the name of the project to load. This switch is mandatory.

  • -X VAR=VALUE is used to specify the value of scenario variables used in your project. This is similar to the homonym switch in gprbuild.

  • –symlinks should be specified if your projet uses symbolic links for files. This will ensure that the links are fully resolved as stored in the database, and thus that when a file is visible through different links, the information is appropriately coalesced in the database for that file.

  • –subdirs=ARG is similar to the homonym switch in gprbuild

  • –tracefile=ARG is used to point to a file compatible with GNATCOLL.Traces that controls the debug information generated by gnatinspect. By default, gnatinspect parses a file called ‘.gnatdebug’ in the current directory.

  • –encoding=ARG is the character encoding used for source and ALI files. By default, gnatinspect assumes they are encoded in UTF-8.

Once it has finished parsing the xref information, gnatinspect displays an interactive prompt, where a number of commands can be used to perform queries. In a lot of cases, these commands take some file information as argument (either just the file, or an entity name and the file in which it is defined).

The file names can be given as either a base name, or relative to the current directory, or even a full name. But file names are ambiguous (even when a full path is specified) when aggregate projects are used. It is valid for a given file to be part of multiple aggregate projects, and depending on the project we are considering the result of the xref queries might vary).

To remove the ambiguity, it is possible to specify the project to which the file belongs. The project is specified either as a project name (which itself could be ambiguous with aggregate projects), or as a full path.

In all commands below, whenever the parameter specifies “:file”, you can use instead “:file:project” if there are ambiguities. It is also possible not to specify the file, in which case the entity will be looked for in all sources of the project.

Here is the full list of commands supported by gnatinspect:

  • decl name:file:line:column is probably the most useful command. Given a reference to an entity, it will indicate where the entity is declared. The line and column informations are optional:

    >>> decl Func:file.adb:12
    Func:/some/path/file2.adb:20:9
    
  • body name:file:line:column is similar to decl, but will return the location of the body of the entity. When the entity is an Ada private type, its body is in fact the location of the full declaration for that type.

  • refs name:file:line:column displays all known references to the entity.

  • refs_overriding name:file:line:column displays all known references to the entity or one of its overriding entities

  • doc name:file:line:column will display documentation for the entity. The exact format for the entity might change in future versions of gnatinspect, but will in general include the type of entity, the location of its declaration, and any comment associated with it in the source code:

    >>> doc Func:file.adb
    procedure declared at /some/path/file2.adb:20:9
    
    And the comments written below Func in file2.adb
    
  • fields name:file:line:column displays the fields of an Ada record type or a C struct:

    >>> fields Rec:file.ads:20
    A:/some/path/file.ads:21
    B:/some/path/file.ads:22
    
  • child_types name:file:line:column lists all child types for this entity, for instance classes that inherit from the entity. This is the opposite of parent_types.

  • child_types_recursive name:file:line:column is similar to child_types but will also list the child types of the children. This query can be used to find a whole tagged type hierarchy (or class hierarchy in C++).

  • parent_types name:file:lin:column returns the parent types for the entity, for instance the classes or interfaces from which it derives. See also child_types.

  • methods name:file:line:column returns the list of methods (or primitive operations) for the entity.

  • method_of name:file:line:column returns the class or tagged type for which the entity is a method.

  • calls name:file:line:column lists all entities called by the entity. This includes all entities defined within the scope of the entity (so for a subprogram this will be the list of local variables, but for a package this includes all subprograms and nested packages defined within that package).

  • callers name:file:line:column lists all entities that call the entity. This information is also available from a call to ‘refs’, but ‘callers’ return the callers directly, instead of references to the original entity.

  • overrides name:file:line:column returns the entity that is overridden by the entity (generally a method from a parent class).

  • overridden name:file:line:column returns the list of entities that override the parameter (generally methods from children classes).

  • overridden_recursive name:file:line:column returns the list of entities that override the parameter (generally methods from children classes). This is recursive.

  • type name:file:line:column returns the type of the entity (variable or constant). For an enumeration literal, this returns the corresponding enumeration.

  • component name:file:line:column returns the component type of the entity (for arrays for instance).

  • literals name:file:line:column returns the valid literal values for an enumeration.

  • pointed name:file:line:column returns the type pointed to by the entity.

  • qname name:file:line:column returns the fully qualified name for the entity.

  • params name:file:line:column returns the list of parameters for the subprogram.

A number of queries are related to the source files of the project:

  • importing filename lists the files that import the file (via with statements in Ada or #include in C for instance)

  • imports filename lists the files that the file imports (via with statements in Ada or #include in C for instance). See also depends_on.

  • depends filename lists the files that the file depends on (recursively calling imports)

  • entities file lists all entities referenced or declared in the file.

Finally, some commands are not related to entities or source files:

  • refresh refreshes the contents of the xref database, by parsing all ALI files that have been changed.

  • shell Execute a shell command (an alternative is to use ‘!’ as the command).

  • scenario VARIABLE VALUE changes the value of a scenario variable, and reparse the project.

  • time command arguments executes the command as usual, and report the time it took to execute it.