Compiling and Linking

Separate compilation

TADS 3 provides "separate compilation," which means that you can arrange your program's source code into several modules, each of which is compiled separately from the others, then linked together with the others into the finished program.

The compiler includes a built-in "make" facility, which automatically recompiles the source modules that you've changed since they were last compiled. This similar to the Unix "make" tool, but TADS 3's isn't programmable like the Unix version; instead, the TADS 3 version is pre-programmed with the relationships among source, object, and image files.

To build a program using TADS 3, you run the compiler command, t3make, with all of the source modules listed as arguments. Each module is compiled separately. This saves time as you develop your game, because each time you change your source code and recompile, the compiler only has to rebuild the files that have actually changed since the last build. Most people find that as they work on a game, they tend to focus their work at any given time in a small part of the game - in one or two source files. On each rebuild, only those modified source files need to be compiled; the compiler automatically re-uses the compiled version of the rest of the files, which didn't change between builds.

To take maximum advantage of separate compilation, you should break your game up into multiple source files, and keep each file fairly small. There's no need to take this to extremes; the compiler is fairly fast, and you won't see a lot of difference in compilation time between, say, a 5k file and a 10k file (but go up to 100k or 200k you'll start to see a difference). Most people find that they want to divide their games up into multiple files anyway, just to make the navigation easier in the source code; and most games have natural ways of being divided, such as into geographical areas of the game world. Just go with what fits your game and your working style; the time savings of separate compilation will probably turn out to be a free bonus side effect.

The compilation process

When you run t3make, it first compiles each source file that has been modified since it was last compiled. The result of compiling a source module is an "object file"; a source file called "mygame.t" would have a corresponding object file called "mygame.t3o". ("t3o" stands for "T3 Object.") If a source file - or any of its included header files - has been modified since the corresponding object file was created, t3make compiles the source file. If the object file is up to date with the source (and all included headers), there's no need to recompile the source file, so t3make automatically skips it.

Note that the "object" in "object file" doesn't refer to Object Oriented Programming. Rather, this is a whole separate bit of computer jargon. In this context, "object file" refers to the file generated by the compiler, representing the compiled form of a source file. This kind of file isn't human-readable; it's a bunch of binary gibberish that's only meaningful to the compiler. It's also not something you can directly set running - it's not executable code. It's an intermediate file, halfway between the source code and the finished program. An object file contains a version of the source program that has been translated into machine code, but which usually contains references to external symbols that must be resolved by the linker before the code becomes an executable program.

After compiling all of the modified source files into object files, t3make looks to see if the image file is up to date by checking to see if any object files are newer. If at least one object file is newer than the image file, the image file must be re-linked. Of course, if any source files are compiled during this run of t3make, the image will have to be re-linked. If it is necessary to re-link, t3make loads all of the object files, resolves their mutual external references, and creates an image file.

Project Files

If you divide your program into several source files, you might find that your t3make command line becomes long enough that it's tedious to type every time you rebuild. This doesn't matter if you're using an integrated environment like TADS Workbench, but it does if you're using the command-line compiler directly.

To make things easier for command-line users, t3make can read the command options from a file rather than making you type them all every time you build. This is known as a project file; it's simply a text file that contains the file names and options that you'd normally type on the t3make command line.

Project file syntax

For the most part, a project file simply contains the same text you'd put on the t3make command line. There are a few special features, though.

Comments: a line starting with a pound sign ("#") is a comment. t3make ignores comment lines.

Quoted elements: if a file name contains a space or "#", enclose the file name in double quotes. If the name contains a double quote mark, stutter (double) it: for my "file".txt, write "my ""file"".txt".

Portable path notation: filenames that contain directory path elements should be expressed in Unix-style notation, with "/" characters separating path elements. This applies to the names of source files, and also to option arguments that specify files or directories, such as -I, -Fs, -Fy, etc.

See Universal Paths for more details.

This will look a little weird if you're using Windows and you're accustomed to "\" as the path separator, or you're used to ":" separators on Mac OS, but it makes the project file portable. You can give it to someone using a different operating system, and it'll work without any changes. t3make automatically converts the "/" notation to the local path notation on each platform.

Use default extensions: another portability concern is filename suffixes, or extensions - the ".t" at the end of a source file's name for example. You're better off omitting these for source files (.t) and libraries (.tl) - the compiler will automatically add them for you. There are a few platforms with unusual conventions that require different extension formats, so it's more portable to allow the compiler to supply the default. The -o file is an exception; for this you should specify the exact name you want.

Sample project file

Here's a sample project file.

#
# project file for calc
#

# image file
-o exe/calc.t3

# source files
-file "calc sources/calc"
-file "tok sources/tok"

Using a project file

To read options from a file, use the -f compiler option:

t3make -f calc.t3m

The options read from an options file are appended after any options on the command line, so you can mix options from a file and the command line. For example, if you wanted to use the project file above to compile the program for debugging, you could enter a command like this:

t3make -f calc.t3m -d

Default Project File

If you run t3make and do not specify any modules (in other words, your command line consists only of options), and you also don't include a -f option to specify a project file, the compiler looks for a default project file called "makefile.t3m" in the current directory. If this file is present, the compiler reads the file as though you had specified it with the -f option. This makes it very easy to build your program; if you put your build options in the file makefile.t3m, you can build simply by typing "t3make".

The normal rules for project files apply to makefile.t3m, so you can specify additional options on the command line when building with the default project file. For example, to build for debugging using the default project file, you would simply enter this:

t3make -d

The t3make command line

The typical t3make command line looks like this:

t3make -o mygame.t3 -Fo obj -Fy sym game1.t game2.t game3.t

The -o option tells the linker the name of the image file; this is the T3 executable file that you run using the TADS 3 interpreter. The -Fo option, if you include it, tells the compiler where to put object files; by default, they'll go in the same place as the source file, but if you want to store object files in a separate directory, which you might wish to do to keep your source directory uncluttered, you can use this option. Similarly, the -Fy option tells the compiler where to put "symbol" files, which are another kind of generated file that the compiler creates; as with objects, you might want to keep these files in a separate directory from the source to avoid clutter. (You don't otherwise need to worry about symbol files; they're purely for the compiler's use.) Finally, you must list the source files that make up your game.

The complete syntax of the t3make command is as follows:

t3make [ option ... ]  ]  source [ source ... ]  [ -res resourceOption ... ]  ] 

The options are special keywords that all start with a dash ("-") that control the way the compiler works; more on these in a moment. The sources are the names of the source files and library files that go into the build. The resourceOptions let you add multi-media resources into the final image file.

Options

Here's a list of the compiler options that you can specify:

Source and library files

After the options, you list all of the source modules involved in the compilation. Each source module can be a regular source file (a ".t" file), or it can be a "library" file, as explained below. The compiler attempts to infer the type of each file by checking its filename suffix: if the suffix is ".tl", the compiler assumes that the file is a library, otherwise it assumes it is a source file. If you don't use the conventional filename suffixes for your source and library files, you must explicitly tell the compiler the type each file by prefixing each file with a "-source" or "-lib" specifier. For example:

t3make -o mygame.t3 -source game1.tads -lib mylib.tadslib

Immediately following the name of a library, you can list one or more "-x" options to exclude modules that the library includes. This is explained below.

The "-source" and "-lib" type specifiers aren't actually options, in terms of the order in which they appear on the command line. They look like options, in that they start with a dash, but they can't be mixed in positionally with the regular options. They can only appear in the module list portion of the command line, which follows all of the options.

The compiler treats the "-" prefix as special everywhere in the command line. This means that if the name of one of your source files actually starts with "-", you must put a "-source" specifier immediately before that filename, even if it ends with the conventional suffix, because otherwise the compiler would be confused into thinking the filename was meant as a type specifier.

Note that you can always use the "-source" and "-lib" specifiers, even when you don't need to. If you're putting your build settings into a project makefile (a .t3m file), it's a good idea to use "-source" and "-lib" specifiers consistently for all of your sources, because it makes the project file easier to read. Also, when you use one a "-source" or "-lib" specifier, you can omit the ".t" or ".tl" suffix from the filename, because the compiler will know to automatically add the appropriate suffix using the appropriate local OS conventions. So using the specifier and omitting the suffix makes your project files more portable by eliminating any dependency on OS file naming conventions.

File inclusion order

The order of files listed in the project is important when you use "modify" or "replace". The reason is that "modify" and "replace" statements can be used to further modify a previously modified object. This makes it necessary to have an explicit, well-defined order in which modifications are applied - otherwise there would be ambiguity about which was the "final" version of an object.

The rule for carrying out "modify" and "replace" operations is that they're applied in "source order" within a single source file, and they're applied in project file order across source files. "Project file order" is simply the order in which the files are listed in the project file (the .t3m file). (It's almost as though you were to append all of your source files together into one big file, appending one after the other in the project file order. Of course, the compiler doesn't actually do this - that wouldn't work with the preprocessor, which handles each file's #define symbols separately, and it wouldn't work with certain compiler features that are scoped to individual source files. But for the purposes of "modify" and "replace", the overall effect is to put all of the object definitions into a single, linear order from the first file listed in the project to the last.)

Other than "modify" and "replace" dependencies, the order of source file inclusion doesn't matter. In particular, source code in one module can freely refer to functions and objects defined in other modules, regardless of the relative order of those modules within the project file.

It's usually not too hard to determine the correct ordering for your project files. The important thing is to put files with "modify" after the files that define the things they modify.

For nearly every project, it will work well to use an ordering like this:

The common thread here is that we order the files from most general to most specific: we start with the base system and adv3 libraries that everyone depends on, then we add some custom libraries that some people use but aren't universal, then we add the private files that you wrote just for this project.

Source Filename Uniqueness

All source files in a project must have unique "root" filenames. (The root filename is the part of the filename that doesn't include any directory path prefix.) This applies even to source files in libraries-a source file in one library can't have the same root filename as a source file in any other library, and it can't have the same root name as any source file included in the project's main source file list.

The reason that source filenames must be unique is that the compiler uses the source name to determine the name of the object (.t3o) and symbol (.t3s) files generated from the source file. Since all of these generated files for a given project are usually placed into a single output directory, if two source modules had the same filename, they'd both correspond to the same object file, so one module's object file would overwrite the other with the same name. To avoid this problem, the compiler requires that each source file's name is unique.

Multi-media resources

Following the list of source and library files, you can specify a list of multi-media resource files to "bundle" into the image file. A bundled resource is appended to the .t3 file, along with catalog information that lets the interpreter recover the file when it's needed. Bundling resource files into the .t3 file makes your game more self-contained, because the .t3 file is the only file you have to distribute - you don't have to include the resources as separate files, since their contents are stored within the .t3 file. Bundling also prevents users from casually or accidentally browsing your image and sound files, since ordinary operating system tools don't know how to dissect the .t3 file's contents.

You can specify the list of multi-media resources to bundle into your image file by placing the "-res" option after all of your source and library modules, then listing your multi-media resources. After the "-res" option, you can specify one or more resource item. Each resource item can be one of the following:

A "resource name" is the name by which the resource is known at run-time. This is the name you use to access the resource from HTML; for example, it's the name you'd use in an <IMG SRC=xxx> tag. You'd also use this name to access the resource from the File class.

There are two special things to note about resource bundling.

First, when you compile in Debug mode, the compiler only stores "links" to the resources, not copies of their contents. A link records the resource name and its corresponding local file name; at run time, the HTML renderer uses this information to load the resource from the local file. This means that a Debug build is not self-contained with respect to resources - it still needs to have the local files available while running.

The reasoning here is that Debug builds are purely for your own testing, so you'll only be running them within your build environment on your own machine, where all of the resource files started out. Since the resource files are all available anyway, embedding copies of in the .t3 file would just waste time and disk space. By skipping the resource-copy step, the Debug build can run a little faster. This is important while you're actively developing a game, because you'll probably run through the edit/compile/test cycle many times.

Second, the resources are added to the final image file after the pre-initialization phase has finished. This means that you can include resources that are created during your pre-initialization phase, which is handy for things like the "gameinfo.txt" file (which the standard library builds automatically during pre-initialization). This also means that resources you access during pre-initialization won't come from the image file, but this should be transparent to your program, since the resource loader automatically looks for the resource files in the file system anyway.

Libraries

It's often useful to take a set of source files and group them together as though they were a unit. For example, the standard Adventure Library included with TADS consists of a number of separate source files; for the most part, you don't want to have to think about which individual files are involved - you just want to include the entire Adventure Library as a unit.

The compiler has a mechanism to simplify working with groups of files like this: you can create a separate, special kind of file that lists the regular source files that make up the group, and then include only this special listing file in your compilation. The compiler automatically reads the list of files from the special file. This special file is called a "library file" (not to be confused with the more generic usage of "library" that refers to a group of files the provide reusable source code, such as the Adventure Library - a library file is a specially formatted file that specifies a group of source files to include in a compilation).

By convention, a library filename always has the suffix ".tl" (but note that the period might be replaced with another character on some platforms, and some platforms don't use filename suffixes at all).

Library headers (#include files)

The compiler automatically adds each library's directory to the include path. If you're distributing a library, this means that you can bundle all of your library's files together into a single directory, and the user will only have to specify the path to your library's install directory once, when listing the .tl file on the compiler command line.

Library Format

A library file is simply a text file that obeys a specific format. A library file is formatted with one definition per line (blank lines are ignored, as are lines starting with the comment character "#"). A definition has this format:

keyword : value

The keyword is a word specifying what kind of information the line contains, and the value is the text of the information defined. The valid keywords are:

In addition, certain directives can appear as simply keywords without associated values:

Specifying Filenames

The values for the source, library, and resource keywords are filenames, given in a portable format. You must not specify a filename extension (such as ".t") on source and library values, because the compiler adds the appropriate extension for the local platform automatically. You must use slash ("/") characters to indicate directory path separators; the compiler automatically converts the "/" characters to the appropriate local conventions. If you specify any directory paths, you must specify only relative paths that are subdirectories of the folder containing the library; you are not allowed to specify absolute paths, and you are not allowed to use ".." or other OS-specific conventions to refer to parent directories. If you follow all of these conventions, you will ensure that your library files will be portable to all operating systems, so that other people using different operating systems won't have to modify your library files for their local conventions.

When the compiler reads a library file, it converts each source, library, and resource value to local conventions. The compiler makes the following changes to each name:

Preprocessor Symbol (Macro) Substitution

Inside a library file, you can substitute the value of a preprocessor symbol defined with the -D option. This allows you to create a library that selects different sub-sections that depend on -D settings; the standard adv3 library, for example, uses this to select the language-specific library based on the setting of the LANGUAGE symbol.

To substitute the value of a preprocessor symbol, use the format "$(NAME)", where NAME is the name of the preprocessor symbol to substitute. For example, to substitute the value of the LANGUAGE variable, put "$(LANGUAGE)" where you want the value to be substituted:

library: $(LANGUAGE)/$(LANGUAGE).tl

In this example, the LANGUAGE variable's value is substituted into this line of text twice. If the command line contained the option "-D LANGUAGE=en_us", then the compiler would read the line above as though it were written like this:

library: en_us/en_us.tl

The compiler substitutes all "$(NAME)" sequences before interpreting each line of text from the library file.

If a dollar sign appears in a library file, but isn't part of a complete "$(NAME)" sequence, the compiler displays an error message. To use a dollar sign literally in a library file (for example, if you want to refer to a file that has a dollar sign in its name), write two dollar signs: the compiler converts each "$$" sequence to a single dollar sign.

Any symbols used in "$(NAME)" constructions must be defined with -D options, either on the command line or in the project (.t3m) file, before the library is listed in the command line or project file. The compiler displays an error message if any symbol is used in a library file but isn't defined with a -D option.

Example library file

Here's an example of a library file:

name: Calculator Library
source: display
source: keypad
source: arith
source: sci
source: trig

This library has the display name "Calculator Library", and includes five source files: display.t, keypad.t, arith.t, sci.t, and trig.t. Assuming you called this library "calc.tl", you would include the library in a compilation like so:

t3make -d -lib calc.tl mygame.t

Because the library name ends in ".tl", you don't need to include a "-lib" specifier before it - the compiler infers that the file is a library from the suffix. (You always can include a specifier for any file, but you don't need to unless the file has a non-standard suffix for its type.)

Exclusions

Sometimes, a library includes non-essential files that you don't need in your program. For example, a library might have some extended functionality that it includes for those programs that want it, but which can be omitted by programs that don't use it. Including extraneous code is generally not harmful, but it does unnecessarily increase the size of your program's compiled image file and its run-time memory needs, so it's best to avoid adding code you're not using.

The obvious way of eliminating unneeded code would be to edit the library file itself to remove the modules you don't need. This isn't ideal, though, because it would remove those modules from other projects you're working on that might need the extra code. To solve this problem, you could simply create a copy of the library file and remove the unneeded modules from it. This creates another problem, though: if you got the library file from someone else, and they later change the library by renaming a source file or adding or removing files, you'd have to make the same changes to your copy or copies of the library.

Fortunately, there's a way to exclude files from a library without changing the library itself. When you include a library on the compiler command line, the compiler lets you list one or more files to exclude, using "-x" specifiers. A "-x" specifier must be placed on the compiler command line immediately after the library to which it applies. Each "-x" is followed by the name of a source module to exclude, using the name as it appears in the library - that is, using the portable name format, without an extension and with "/" as the path separator. Simply use the exact text of the source value as it appears in the library file.

Note that you cannot exclude an entire sub-library from a library. If a library includes a sub-library, you must exclude the files from the sub-library individually. To do so, treat the sub-library name as a path prefix, and place a "/" after the sub-library name, then add the filename as it appears in the sub-library. For example, suppose that another library, desk.tl, includes the calc.tl library as a sub-library:

# desk accessory library
name: Desk Accessory Library
library: pen
library: pencil
library: calc

Now, suppose you compile a program including the desk.tl library, but you want to exclude the trig.t module included in the calc.tl library. To do this, you'd write a "-x" option like this:

t3make desk.tl -x calc/trig

Default Modules

TADS 3 includes a default source module called _main.t that contains some low-level support code that most programs need. Because most programs will not have any reason to customize this module, the compiler automatically includes the module; this saves you a little work, because you don't have to add the module to your t3make command line explicitly.

However, it is possible that you'll want to use a different version of the code in _main.t, in which case you will need to remove the default version from the build. To do this, use the compiler's -nodef option. This option tells the compiler not to include any default modules in the build; only the modules you explicitly list on the command line (or in a project file) are included in the build in this case.

Default Search Paths

Include files: The compiler automatically adds the default system header directory to the end of the include file search path. In effect, the compiler adds a -I option, after all user-specified -I options, specifying the system header directory. The location of the system header directory varies by system; on DOS and Windows, this is the directory containing the compiler. Refer to your system-specific release notes for details on other systems.

Source files: The compiler automatically adds the default system library source directory to the end of the source file search path. In effect, the compiler adds a -Fs option, after all user-specified -Fs options, specifying the system library source directory. The location of this directory varies by system; on DOS and Windows, this is the compiler install directory. Refer to your system-specific release notes for details on other systems.

Library files: On some systems, the compiler searches for library (.tl) files in additional locations when it can't find a .tl file in one of the regular source file locations. The particulars vary by system, so check your system-specific release notes for details. For example, on MS-DOS, Windows, and Unix systems, the compiler looks for an environment variable called TADSLIB, and searches in each directory listed there (on DOS and Windows, multiple directories can be listed in the TADSLIB variable by separating them with semicolons; on Unix, paths are separated with colons).

Dependency Tracking

The t3make utility will compile a source file if it finds any of the following conditions:

In addition, t3make will re-link the image (.t3) file under any of these conditions:

The dependency tracking mechanism isn't perfect, and it can be fooled under certain circumstances:

You can force a full recompilation with the t3make -r option; you can use this option if you encounter any situations where you suspect that t3make is missing some dependency and therefore failing to notice a file that requires recompilation.

Notes for TADS 2 Users

TADS 3 doesn't support the "pre-compiled header" feature of TADS 2. The point of TADS 2's pre-compiled headers is to reduce compilation time by letting you bundle up a part of your game that's basically done - a part that you're not actively working on - and compile it in advance, so that you don't have to compile it again each time through the edit-compile-test cycle.

TADS 3 doesn't have pre-compiled headers because it has full support for separate compilation, which is much better. With separate compilation, the compiler automatically figures out which parts of your program have changed since the last build, and recompiles only the changed parts. You get the same sort of time savings as with pre-compiled headers, but with virtually no work on your part, and no risk that you'll forget to recompile a file you've changed.

To take full advantage of TADS 3's separate compilation, don't use #include to assemble your program as you did in TADS 2. In TADS 2, you had to have a central file that #included all of your other source files. With TADS 3, don't do this. Instead, just add all of the source files to your project (the ".t3m" file, if you're using the command-line compiler).