This command loads binary code from a file into the application's address space and calls an initialization procedure in the package to incorporate it into an interpreter. fileName is the name of the file containing the code; its exact form varies from system to system but on most systems it is a shared library, such as a .so file under Solaris or a DLL under Windows. packageName is the name of the package, and is used to compute the name of an initialization procedure. interp is the path name of the interpreter into which to load the package (see the interp manual entry for details); if interp is omitted, it defaults to the interpreter in which the load command was invoked.
Once the file has been loaded into the application's address space, one of two initialization procedures will be invoked in the new code. Typically the initialization procedure will add new commands to a Tcl interpreter. The name of the initialization procedure is determined by packageName and whether or not the target interpreter is a safe one. For normal interpreters the name of the initialization procedure will have the form pkg_Init, where pkg is the same as packageName except that the first letter is converted to upper case and all other letters are converted to lower case. For example, if packageName is foo or FOo, the initialization procedure's name will be Foo_Init.
If the target interpreter is a safe interpreter, then the name of the initialization procedure will be pkg_SafeInit instead of pkg_Init. The pkg_SafeInit function should be written carefully, so that it initializes the safe interpreter only with partial functionality provided by the package that is safe for use by untrusted code. For more information on Safe-Tcl, see the safe manual entry.
The initialization procedure must match the following prototype:
typedef int Tcl_PackageInitProc(Tcl_Interp *interp);The interp argument identifies the interpreter in which the package is to be loaded. The initialization procedure must return TCL_OK or TCL_ERROR to indicate whether or not it completed successfully; in the event of an error it should set the interpreter's result to point to an error message. The result of the load command will be the result returned by the initialization procedure.
The actual loading of a file will only be done once for each fileName in an application. If a given fileName is loaded into multiple interpreters, then the first load will load the code and call the initialization procedure; subsequent loads will call the initialization procedure without loading the code again. For Tcl versions lower than 8.5, it is not possible to unload or reload a package. From version 8.5 however, the unload command allows the unloading of libraries loaded with load, for libraries that are aware of the Tcl's unloading mechanism.
The load command also supports packages that are statically linked with the application, if those packages have been registered by calling the Tcl_StaticPackage procedure. If fileName is an empty string, then packageName must be specified.
If packageName is omitted or specified as an empty string, Tcl tries to guess the name of the package. This may be done differently on different platforms. The default guess, which is used on most UNIX platforms, is to take the last element of fileName, strip off the first three characters if they are lib, and use any following alphabetic and underline characters as the module name. For example, the command load libxyz4.2.so uses the module name xyz and the command load bin/last.so {} uses the module name last.
If fileName is an empty string, then packageName must be specified. The load command first searches for a statically loaded package (one that has been registered by calling the Tcl_StaticPackage procedure) by that name; if one is found, it is used. Otherwise, the load command searches for a dynamically loaded package by that name, and uses it if it is found. If several different files have been loaded with different versions of the package, Tcl picks the file that was loaded first.
load [file join [pwd] mylib.DLL]
If the same file is loaded by different fileNames, it will be loaded into the process's address space multiple times. The behavior of this varies from system to system (some systems may detect the redundant loads, others may not).
#include <tcl.h> #include <stdio.h> static int fooCmd(ClientData clientData, Tcl_Interp *interp, int objc, Tcl_Obj *const objv[]) { printf("called with %d arguments\n", objc); return TCL_OK; } int Foo_Init(Tcl_Interp *interp) { if (Tcl_InitStubs(interp, "8.1", 0) == NULL) { return TCL_ERROR; } printf("creating foo command"); Tcl_CreateObjCommand(interp, "foo", fooCmd, NULL, NULL); return TCL_OK; }
When built into a shared/dynamic library with a suitable name (e.g. foo.dll on Windows, libfoo.so on Solaris and Linux) it can then be loaded into Tcl with the following:
# Load the extension switch $tcl_platform(platform) { windows { load [file join [pwd] foo.dll] } unix { load [file join [pwd] libfoo[info sharedlibextension]] } } # Now execute the command defined by the extension foo