This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License along with this program; see the file COPYING. If not, write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.

Introduction

Cons is a system for constructing, primarily, software, but is quite different from previous software construction systems. Cons was designed from the ground up to deal easily with the construction of software spread over multiple source directories. Cons makes it easy to create build scripts that are simple, understandable and maintainable. Cons ensures that complex software is easily and accurately reproducible.

Cons uses a number of techniques to accomplish all of this. Construction scripts are just Perl scripts, making them both easy to comprehend and very flexible. Global scoping of variables is replaced with an import/export mechanism for sharing information between scripts, significantly improving the readability and maintainability of each script. Construction environments are introduced: these are Perl objects that capture the information required for controlling the build process. Multiple environments are used when different semantics are required for generating products in the build tree. Cons implements automatic dependency analysis and uses this to globally sequence the entire build. Variant builds are easily produced from a single source tree. Intelligent build subsetting is possible, when working on localized changes. Overrides can be setup to easily override build instructions without modifying any scripts. MD5 cryptographic signatures are associated with derived files, and are used to accurately determine whether a given file needs to be rebuilt.

While offering all of the above, and more, Cons remains simple and easy to use. This will, hopefully, become clear as you read the remainder of this document.

Why Cons? Why not Make?

Cons is a make replacement. In the following paragraphs, we look at a few of the undesirable characteristics of make--and typical build environments based on make--that motivated the development of Cons.

Build complexity

Traditional make-based systems of any size tend to become quite complex. The original make utility and its derivatives have contributed to this tendency in a number of ways. Make is not good at dealing with systems that are spread over multiple directories. Various work-arounds are used to overcome this difficulty; the usual choice is for make to invoke itself recursively for each sub-directory of a build. This leads to complicated code, in which it is often unclear how a variable is set, or what effect the setting of a variable will have on the build as a whole. The make scripting language has gradually been extended to provide more possibilities, but these have largely served to clutter an already overextended language. Often, builds are done in multiple passes in order to provide appropriate products from one directory to another directory. This represents a further increase in build complexity.

Build reproducibility

The bane of all makes has always been the correct handling of dependencies. Most often, an attempt is made to do a reasonable job of dependencies within a single directory, but no serious attempt is made to do the job between directories. Even when dependencies are working correctly, make's reliance on a simple time stamp comparison to determine whether a file is out of date with respect to its dependents is not, in general, adequate for determining when a file should be rederived. If an external library, for example, is rebuilt and then ``snapped'' into place, the timestamps on its newly created files may well be earlier than the last local build, since it was built before it became visible.

Variant builds

Make provides only limited facilities for handling variant builds. With the proliferation of hardware platforms and the need for debuggable vs. optimized code, the ability to easily create these variants is essential. More importantly, if variants are created, it is important to either be able to separate the variants or to be able to reproduce the original or variant at will. With make it is very difficult to separate the builds into multiple build directories, separate from the source. And if this technique isn't used, it's also virtually impossible to guarantee at any given time which variant is present in the tree, without resorting to a complete rebuild.

Repositories

Make provides only limited support for building software from code that exists in a central repository directory structure. The VPATH feature of GNU make (and some other make implementations) is intended to provide this, but doesn't work as expected: it changes the path of target file to the VPATH name too early in its analysis, and therefore searches for all dependencies in the VPATH directory. To ensure correct development builds, it is important to be able to create a file in a local build directory and have any files in a code repository (a VPATH directory, in make terms) that depend on the local file get rebuilt properly. This isn't possible with VPATH, without coding a lot of complex repository knowledge directly into the makefiles.

Keeping it simple

A few of the difficulties with make have been cited above. In this and subsequent sections, we shall introduce Cons and show how these issues are addressed.

Perl scripts

Cons is Perl-based. That is, Cons scripts--Conscript and Construct files, the equivalent to Makefile or makefile--are all written in Perl. This provides an immediate benefit: the language for writing scripts is a familiar one. Even if you don't happen to be a Perl programmer, it helps to know that Perl is basically just a simple declarative language, with a well-defined flow of control, and familiar semantics. It has variables that behave basically the way you would expect them to, subroutines, flow of control, and so on. There is no special syntax introduced for Cons. The use of Perl as a scripting language simplifies the task of expressing the appropriate solution to the often complex requirements of a build.

Hello, World!

To ground the following discussion, here's how you could build the Hello, World! C application with Cons:

  $env = new cons();
  Program $env 'hello', 'hello.c';

If you install this script in a directory, naming the script Construct, and create the hello.c source file in the same directory, then you can type cons hello to build the application:

  % cons hello
  cc -c hello.c -o hello.o
  cc -o hello hello.o

Construction environments

A key simplification of Cons is the idea of a construction environment. A construction environment is an object characterized by a set of key/value pairs and a set of methods. In order to tell Cons how to build something, you invoke the appropriate method via an appropriate construction environment. Consider the following example:

  $env = new cons(
        CC      =>      'gcc',
        LIBS    =>      'libworld.a'
  );

  Program $env 'hello', 'hello.c';

In this case, rather than using the default construction environment, as is, we have overridden the value of CC so that the GNU C Compiler equivalent is used, instead. Since this version of Hello, World! requires a library, libworld.a, we have specified that any program linked in this environment should be linked with that library. If the library exists already, well and good, but if not, then we'll also have to include the statement:

  Library $env 'libworld', 'world.c';

Now if you type cons hello, the library will be built before the program is linked, and, of course, gcc will be used to compile both modules:

  % cons hello
  gcc -c hello.c -o hello.o
  gcc -c world.c -o world.o
  ar r libworld.a world.o
  ar: creating libworld.a
  ranlib libworld.a
  gcc -o hello hello.o libworld.a

Automatic and complete dependency analysis

With Cons, dependencies are handled automatically. Continuing the previous example, note that when we modify world.c, world.o is recompiled, libworld.a recreated, and hello relinked:

  % vi world.c
    [EDIT]
  % cons hello
  gcc -c world.c -o world.o
  ar r libworld.a world.o
  ar: creating libworld.a
  ranlib libworld.a
  gcc -o hello hello.o libworld.a

This is a relatively simple example: Cons ``knows'' world.o depends upon world.c, because the dependency is explicitly set up by the Library method. It also knows that libworld.a depends upon world.o and that hello depends upon libworld.a, all for similar reasons.

Now it turns out that hello.c also includes the interface definition file, world.h:

  % emacs world.h
    [EDIT]
  % cons hello
  gcc -c hello.c -o hello.o
  gcc -o hello hello.o libworld.a

How does Cons know that hello.c includes world.h, and that hello.o must therefore be recompiled? For now, suffice it to say that when considering whether or not hello.o is up-to-date, Cons invokes a scanner for its dependency, hello.c. This scanner enumerates the files included by hello.c to come up with a list of further dependencies, beyond those made explicit by the Cons script. This process is recursive: any files included by included files will also be scanned.

Isn't this expensive? The answer is--it depends. If you do a full build of a large system, the scanning time is insignificant. If you do a rebuild of a large system, then Cons will spend a fair amount of time thinking about it before it decides that nothing has to be done (although not necessarily more time than make!). The good news is that Cons makes it very easy to intelligently subset your build, when you are working on localized changes.

Automatic global build sequencing

Because Cons does full and accurate dependency analysis, and does this globally, for the entire build, Cons is able to use this information to take full control of the sequencing of the build. This sequencing is evident in the above examples, and is equivalent to what you would expect for make, given a full set of dependencies. With Cons, this extends trivially to larger, multi-directory builds. As a result, all of the complexity involved in making sure that a build is organized correctly--including multi-pass hierarchical builds--is eliminated. We'll discuss this further in the next sections.

Building large trees--still just as simple

A hierarchy of build scripts

A larger build, in Cons, is organized by creating a hierarchy of build scripts. At the top of the tree is a script called Construct. The rest of the scripts, by convention, are each called Conscript. These scripts are connected together, very simply, by the Build, Export, and Import commands.

The Build command

The Build command takes a list of Conscript file names, and arranges for them to be included in the build. For example:

  Build qw(
        drivers/display/Conscript
        drivers/mouse/Conscript
        parser/Conscript
        utilities/Conscript
  );

This is a simple two-level hierarchy of build scripts: all the subsidiary Conscript files are mentioned in the top-level Construct file. Notice that not all directories in the tree necessarily have build scripts associated with them.

This could also be written as a multi-level script. For example, the Construct file might contain this command:

  Build qw(
        parser/Conscript
        drivers/Conscript
        utilities/Conscript
  );

and the Conscript file in the drivers directory might contain this:

  Build qw(
        display/Conscript
        mouse/Conscript
  );

Experience has shown that the former model is a little easier to understand, since the whole construction tree is laid out in front of you, at the top-level. Hybrid schemes are also possible. A separately maintained component that needs to be incorporated into a build tree, for example, might hook into the build tree in one place, but define its own construction hierarchy.

By default, Cons does not change its working directory to the directory containing a subsidiary Conscript file it is including. This behavior can be enabled for a build by specifying, in the top-level Construct file:

  Conscript_chdir 1;

When enabled, Cons will change to the subsidiary Conscript file's containing directory while reading in that file, and then change back to the top-level directory once the file has been processed.

It is expected that this behavior will become the default in some future version of Cons. To prepare for this transition, builds that expect Cons to remain at the top of the build while it reads in a subsidiary Conscript file should explicitly disable this feature as follows:

  Conscript_chdir 0;

Relative, top-relative, and absolute file names

You may have noticed that the file names specified to the Build command are relative to the location of the script it is invoked from. This is generally true for other filename arguments to other commands, too, although we might as well mention here that if you begin a file name with a hash mark, ``#'', then that file is interpreted relative to the top-level directory (where the Construct file resides). And, not surprisingly, if you begin it with ``/'', then it is considered to be an absolute pathname. This is true even on systems which use a back slash rather than a forward slash to name absolute paths.

Using modules in build scripts

You may pull modules into each Conscript file using the normal Perl use or require statements:

  use English;
  require My::Module;

Each use or require only affects the one Conscript file in which it appears. To use a module in multiple Conscript files, you must put a use or require statement in each one that needs the module.

Scope of variables

The top-level Construct file and all Conscript files begin life in a common, separate Perl package. Cons controls the symbol table for the package so that, the symbol table for each script is empty, except for the Construct file, which gets some of the command line arguments. All of the variables that are set or used, therefore, are set by the script itself--not by some external script.

Variables can be explicitly imported by a script from its parent script. To import a variable, it must have been exported by the parent and initialized (otherwise an error will occur).

The Export command

The Export command is used as in the following example:

  $env = new cons();
  $INCLUDE = "#export/include";
  $LIB = "#export/lib";
  Export qw( env INCLUDE LIB );
  Build qw( util/Conscript );

The values of the simple variables mentioned in the Export list will be squirreled away by any subsequent Build commands. The Export command will only export Perl scalar variables, that is, variables whose name begins with $. Other variables, objects, etc. can be exported by reference--but all scripts will refer to the same object, and this object should be considered to be read-only by the subsidiary scripts and by the original exporting script. It's acceptable, however, to assign a new value to the exported scalar variable--that won't change the underlying variable referenced. This sequence, for example, is OK:

  $env = new cons();
  Export qw( env INCLUDE LIB );
  Build qw( util/Conscript );
  $env = new cons(CFLAGS => '-O');
  Build qw( other/Conscript );

It doesn't matter whether the variable is set before or after the Export command. The important thing is the value of the variable at the time the Build command is executed. This is what gets squirreled away. Any subsequent Export commands, by the way, invalidate the first: you must mention all the variables you wish to export on each Export command.

The Import command

Variables exported by the Export command can be imported into subsidiary scripts by the Import command. The subsidiary script always imports variables directly from the superior script. Consider this example:

  Import qw( env INCLUDE );

This is only legal if the parent script exported both $env and $INCLUDE. It also must have given each of these variables values. It is OK for the subsidiary script to only import a subset of the exported variables (in this example, $LIB, which was exported by the previous example, is not imported).

All the imported variables are automatically re-exported, so the sequence:

  Import qw ( env INCLUDE );
  Build qw ( beneath-me/Conscript );

will supply both $env and $INCLUDE to the subsidiary file. If only $env is to be exported, then the following will suffice:

  Import qw ( env INCLUDE );
  Export qw ( env );
  Build qw ( beneath-me/Conscript );

Needless to say, the variables may be modified locally before invoking Build on the subsidiary script.

Build script evaluation order

The only constraint on the ordering of build scripts is that superior scripts are evaluated before their inferior scripts. The top-level Construct file, for instance, is evaluated first, followed by any inferior scripts. This is all you really need to know about the evaluation order, since order is generally irrelevant. Consider the following Build command:

  Build qw(
        drivers/display/Conscript
        drivers/mouse/Conscript
        parser/Conscript
        utilities/Conscript
  );

We've chosen to put the script names in alphabetical order, simply because that's the most convenient for maintenance purposes. Changing the order will make no difference to the build.

A Model for sharing files

Some simple conventions

In any complex software system, a method for sharing build products needs to be established. We propose a simple set of conventions which are trivial to implement with Cons, but very effective.

The basic rule is to require that all build products which need to be shared between directories are shared via an intermediate directory. We have typically called this export, and, in a C environment, provided conventional sub-directories of this directory, such as include, lib, bin, etc.

These directories are defined by the top-level Construct file. A simple Construct file for a Hello, World! application, organized using multiple directories, might look like this:

  # Construct file for Hello, World!

  # Where to put all our shared products.
  $EXPORT = '#export';

  Export qw( CONS INCLUDE LIB BIN );

  # Standard directories for sharing products.
  $INCLUDE = "$EXPORT/include";
  $LIB = "$EXPORT/lib";
  $BIN = "$EXPORT/bin";

  # A standard construction environment.
  $CONS = new cons (
        CPPPATH => $INCLUDE,    # Include path for C Compilations
        LIBPATH => $LIB,        # Library path for linking programs
        LIBS => '-lworld',      # List of standard libraries
  );

  Build qw(
        hello/Conscript
        world/Conscript
  );

The world directory's Conscript file looks like this:

  # Conscript file for directory world
  Import qw( CONS INCLUDE LIB );

  # Install the products of this directory
  Install $CONS $LIB, 'libworld.a';
  Install $CONS $INCLUDE, 'world.h';

  # Internal products
  Library $CONS 'libworld.a', 'world.c';

and the hello directory's Conscript file looks like this:

  # Conscript file for directory hello
  Import qw( CONS BIN );

  # Exported products
  Install $CONS $BIN, 'hello';

  # Internal products
  Program $CONS 'hello', 'hello.c';

To construct a Hello, World! program with this directory structure, go to the top-level directory, and invoke cons with the appropriate arguments. In the following example, we tell Cons to build the directory export. To build a directory, Cons recursively builds all known products within that directory (only if they need rebuilding, of course). If any of those products depend upon other products in other directories, then those will be built, too.

  % cons export
  Install world/world.h as export/include/world.h
  cc -Iexport/include -c hello/hello.c -o hello/hello.o
  cc -Iexport/include -c world/world.c -o world/world.o
  ar r world/libworld.a world/world.o
  ar: creating world/libworld.a
  ranlib world/libworld.a
  Install world/libworld.a as export/lib/libworld.a
  cc -o hello/hello hello/hello.o -Lexport/lib -lworld
  Install hello/hello as export/bin/hello

Clean, understandable, location-independent scripts

You'll note that the two Conscript files are very clean and to-the-point. They simply specify products of the directory and how to build those products. The build instructions are minimal: they specify which construction environment to use, the name of the product, and the name of the inputs. Note also that the scripts are location-independent: if you wish to reorganize your source tree, you are free to do so: you only have to change the Construct file (in this example), to specify the new locations of the Conscript files. The use of an export tree makes this goal easy.

Note, too, how Cons takes care of little details for you. All the export directories, for example, were made automatically. And the installed files were really hard-linked into the respective export directories, to save space and time. This attention to detail saves considerable work, and makes it even easier to produce simple, maintainable scripts.

Separating source and build trees

It's often desirable to keep any derived files from the build completely separate from the source files. This makes it much easier to keep track of just what is a source file, and also makes it simpler to handle variant builds, especially if you want the variant builds to co-exist.

Separating build and source directories using the Link command

Cons provides a simple mechanism that handles all of these requirements. The Link command is invoked as in this example:

  Link 'build' => 'src';

The specified directories are ``linked'' to the specified source directory. Let's suppose that you setup a source directory, src, with the sub-directories world and hello below it, as in the previous example. You could then substitute for the original build lines the following:

  Build qw(
        build/world/Conscript
        build/hello/Conscript
  );

Notice that you treat the Conscript file as if it existed in the build directory. Now if you type the same command as before, you will get the following results:

  % cons export
  Install build/world/world.h as export/include/world.h
  cc -Iexport/include -c build/hello/hello.c -o build/hello/hello.o
  cc -Iexport/include -c build/world/world.c -o build/world/world.o
  ar r build/world/libworld.a build/world/world.o
  ar: creating build/world/libworld.a
  ranlib build/world/libworld.a
  Install build/world/libworld.a as export/lib/libworld.a
  cc -o build/hello/hello build/hello/hello.o -Lexport/lib -lworld
  Install build/hello/hello as export/bin/hello

Again, Cons has taken care of the details for you. In particular, you will notice that all the builds are done using source files and object files from the build directory. For example, build/world/world.o is compiled from build/world/world.c, and export/include/world.h is installed from build/world/world.h. This is accomplished on most systems by the simple expedient of ``hard'' linking the required files from each source directory into the appropriate build directory.

The links are maintained correctly by Cons, no matter what you do to the source directory. If you modify a source file, your editor may do this ``in place'' or it may rename it first and create a new file. In the latter case, any hard link will be lost. Cons will detect this condition the next time the source file is needed, and will relink it appropriately.

You'll also notice, by the way, that no changes were required to the underlying Conscript files. And we can go further, as we shall see in the next section.

Variant builds

Hello, World! for baNaNa and peAcH OS's

Variant builds require just another simple extension. Let's take as an example a requirement to allow builds for both the baNaNa and peAcH operating systems. In this case, we are using a distributed file system, such as NFS to access the particular system, and only one or the other of the systems has to be compiled for any given invocation of cons. Here's one way we could set up the Construct file for our Hello, World! application:

  # Construct file for Hello, World!

  die qq(OS must be specified) unless $OS = $ARG{OS};
  die qq(OS must be "peach" or "banana")
        if $OS ne "peach" && $OS ne "banana";

  # Where to put all our shared products.
  $EXPORT = "#export/$OS";

  Export qw( CONS INCLUDE LIB BIN );

  # Standard directories for sharing products.
  $INCLUDE = "$EXPORT/include";
  $LIB = "$EXPORT/lib";
  $BIN = "$EXPORT/bin";

  # A standard construction environment.
  $CONS = new cons (
        CPPPATH => $INCLUDE,    # Include path for C Compilations
        LIBPATH => $LIB,        # Library path for linking programs
        LIBS => '-lworld',      # List of standard libraries
  );

  # $BUILD is where we will derive everything.
  $BUILD = "#build/$OS";

  # Tell cons where the source files for $BUILD are.
  Link $BUILD => 'src';

  Build (
        "$BUILD/hello/Conscript",
        "$BUILD/world/Conscript",
  );

Now if we login to a peAcH system, we can build our Hello, World! application for that platform:

  % cons export OS=peach
  Install build/peach/world/world.h as export/peach/include/world.h
  cc -Iexport/peach/include -c build/peach/hello/hello.c -o build/peach/hello/hello.o
  cc -Iexport/peach/include -c build/peach/world/world.c -o build/peach/world/world.o
  ar r build/peach/world/libworld.a build/peach/world/world.o
  ar: creating build/peach/world/libworld.a
  ranlib build/peach/world/libworld.a
  Install build/peach/world/libworld.a as export/peach/lib/libworld.a
  cc -o build/peach/hello/hello build/peach/hello/hello.o -Lexport/peach/lib -lworld
  Install build/peach/hello/hello as export/peach/bin/hello

Variations on a theme

Other variations of this model are possible. For example, you might decide that you want to separate out your include files into platform dependent and platform independent files. In this case, you'd have to define an alternative to $INCLUDE for platform-dependent files. Most Conscript files, generating purely platform-independent include files, would not have to change.

You might also want to be able to compile your whole system with debugging or profiling, for example, enabled. You could do this with appropriate command line options, such as DEBUG=on. This would then be translated into the appropriate platform-specific requirements to enable debugging (this might include turning off optimization, for example). You could optionally vary the name space for these different types of systems, but, as we'll see in the next section, it's not essential to do this, since Cons is pretty smart about rebuilding things when you change options.

Signatures

MD5 cryptographic signatures

Whenever Cons creates a derived file, it stores a signature for that file. The signature is stored in a separate file, one per directory. After the previous example was compiled, the .consign file in the build/peach/world directory looked like this:

  world.o:834179303 23844c0b102ecdc0b4548d1cd1cbd8c6
  libworld.a:834179304 9bf6587fa06ec49d864811a105222c00

The first number is a timestamp--for a UNIX systems, this is typically the number of seconds since January 1st, 1970. The second value is an MD5 checksum. The Message Digest Algorithm is an algorithm that, given an input string, computes a strong cryptographic signature for that string. The MD5 checksum stored in the .consign file is, in effect, a digest of all the dependency information for the specified file. So, for example, for the world.o file, this includes at least the world.c file, and also any header files that Cons knows about that are included, directly or indirectly by world.c. Not only that, but the actual command line that was used to generate world.o is also fed into the computation of the signature. Similarly, libworld.a gets a signature which ``includes'' all the signatures of its constituents (and hence, transitively, the signatures of their constituents), as well as the command line that created the file.

The signature of a non-derived file is computed, by default, by taking the current modification time of the file and the file's entry name (unless there happens to be a current .consign entry for that file, in which case that signature is used).

Notice that there is no need for a derived file to depend upon any particular Construct or Conscript file--if changes to these files affect the file in question, then this will be automatically reflected in its signature, since relevant parts of the command line are included in the signature. Unrelated changes will have no effect.

When Cons considers whether to derive a particular file, then, it first computes the expected signature of the file. It then compares the file's last modification time with the time recorded in the .consign entry, if one exists. If these times match, then the signature stored in the .consign file is considered to be accurate. If the file's previous signature does not match the new, expected signature, then the file must be rederived.

Notice that a file will be rederived whenever anything about a dependent file changes. In particular, notice that any change to the modification time of a dependent (forward or backwards in time) will force recompilation of the derived file.

The use of these signatures is an extremely simple, efficient, and effective method of improving--dramatically--the reproducibility of a system.

We'll demonstrate this with a simple example:

  # Simple "Hello, World!" Construct file
  $CFLAGS = '-g' if $ARG{DEBUG} eq 'on';
  $CONS = new cons(CFLAGS => $CFLAGS);
  Program $CONS 'hello', 'hello.c';

Notice how Cons recompiles at the appropriate times:

  % cons hello
  cc -c hello.c -o hello.o
  cc -o hello hello.o
  % cons hello
  cons: "hello" is up-to-date.
  % cons DEBUG=on hello
  cc -g -c hello.c -o hello.o
  cc -o hello hello.o
  % cons DEBUG=on hello
  cons: "hello" is up-to-date.
  % cons hello
  cc -c hello.c -o hello.o
  cc -o hello hello.o

Code Repositories

Many software development organizations will have one or more central repository directory trees containing the current source code for one or more projects, as well as the derived object files, libraries, and executables. In order to reduce unnecessary recompilation, it is useful to use files from the repository to build development software--assuming, of course, that no newer dependency file exists in the local build tree.

Repository

Cons provides a mechanism to specify a list of code repositories that will be searched, in-order, for source files and derived files not found in the local build directory tree.

The following lines in a Construct file will instruct Cons to look first under the /usr/experiment/repository directory and then under the /usr/product/repository directory:

  Repository qw (
        /usr/experiment/repository
        /usr/product/repository
  );

The repository directories specified may contain source files, derived files (objects, libraries and executables), or both. If there is no local file (source or derived) under the directory in which Cons is executed, then the first copy of a same-named file found under a repository directory will be used to build any local derived files.

Cons maintains one global list of repositories directories. Cons will eliminate the current directory, and any non-existent directories, from the list.

Finding the Construct file in a Repository

Cons will also search for Construct and Conscript files in the repository tree or trees. This leads to a chicken-and-egg situation, though: how do you look in a repository tree for a Construct file if the Construct file tells you where the repository is? To get around this, repositories may be specified via -R options on the command line:

  % cons -R /usr/experiment/repository -R /usr/product/repository .

Any repository directories specified in the Construct or Conscript files will be appended to the repository directories specified by command-line -R options.

Repository source files

If the source code (include the Conscript file) for the library version of the Hello, World! C application is in a repository (with no derived files), Cons will use the repository source files to create the local object files and executable file:

  % cons -R /usr/src_only/repository hello
  gcc -c /usr/src_only/repository/hello.c -o hello.o
  gcc -c /usr/src_only/repository/world.c -o world.o
  ar r libworld.a world.o
  ar: creating libworld.a
  ranlib libworld.a
  gcc -o hello hello.o libworld.a

Creating a local source file will cause Cons to rebuild the appropriate derived file or files:

  % pico world.c
    [EDIT]
  % cons -R /usr/src_only/repository hello
  gcc -c world.c -o world.o
  ar r libworld.a world.o
  ar: creating libworld.a
  ranlib libworld.a
  gcc -o hello hello.o libworld.a

And removing the local source file will cause Cons to revert back to building the derived files from the repository source:

  % rm world.c
  % cons -R /usr/src_only/repository hello
  gcc -c /usr/src_only/repository/world.c -o world.o
  ar r libworld.a world.o
  ar: creating libworld.a
  ranlib libworld.a
  gcc -o hello hello.o libworld.a

Repository derived files

If a repository tree contains derived files (usually object files, libraries, or executables), Cons will perform its normal signature calculation to decide whether the repository file is up-to-date or a derived file must be built locally. This means that, in order to ensure correct signature calculation, a repository tree must also contain the .consign files that were created by Cons when generating the derived files.

This would usually be accomplished by building the software in the repository (or, alternatively, in a build directory, and then copying the result to the repository):

  % cd /usr/all/repository
  % cons hello
  gcc -c hello.c -o hello.o
  gcc -c world.c -o world.o
  ar r libworld.a world.o
  ar: creating libworld.a
  ranlib libworld.a
  gcc -o hello hello.o libworld.a

(This is safe even if the Construct file lists the /usr/all/repository directory in a Repository command because Cons will remove the current directory from the repository list.)

Now if we want to build a copy of the application with our own hello.c file, we only need to create the one necessary source file, and use the -R option to have Cons use other files from the repository:

  % mkdir $HOME/build1
  % cd $HOME/build1
  % ed hello.c
    [EDIT]
  % cons -R /usr/all/repository hello
  gcc -c hello.c -o hello.o
  gcc -o hello hello.o /usr/all/repository/libworld.a

Notice that Cons has not bothered to recreate a local libworld.a library (or recompile the world.o module), but instead uses the already-compiled version from the repository.

Because the MD5 signatures that Cons puts in the .consign file contain timestamps for the derived files, the signature timestamps must match the file timestamps for a signature to be considered valid.

Some software systems may alter the timestamps on repository files (by copying them, e.g.), in which case Cons will, by default, assume the repository signatures are invalid and rebuild files unnecessarily. This behavior may be altered by specifying:

  Repository_Sig_Times_OK 0;

This tells Cons to ignore timestamps when deciding whether a signature is valid. (Note that avoiding this sanity check means there must be proper control over the repository tree to ensure that the derived files cannot be modified without updating the .consign signature.)

Local copies of files

If the repository tree contains the complete results of a build, and we try to build from the repository without any files in our local tree, something moderately surprising happens:

  % mkdir $HOME/build2
  % cd $HOME/build2
  % cons -R /usr/all/repository hello
  cons: "hello" is up-to-date.

Why does Cons say that the hello program is up-to-date when there is no hello program in the local build directory? Because the repository (not the local directory) contains the up-to-date hello program, and Cons correctly determines that nothing needs to be done to rebuild this up-to-date copy of the file.

There are, however, many times in which it is appropriate to ensure that a local copy of a file always exists. A packaging or testing script, for example, may assume that certain generated files exist locally. Instead of making these subsidiary scripts aware of the repository directory, the Local command may be added to a Construct or Conscript file to specify that a certain file or files must appear in the local build directory:

  Local qw(
        hello
  );

Then, if we re-run the same command, Cons will make a local copy of the program from the repository copy (telling you that it is doing so):

  % cons -R /usr/all/repository hello
  Local copy of hello from /usr/all/repository/hello
  cons: "hello" is up-to-date.

Notice that, because the act of making the local copy is not considered a ``build'' of the hello file, Cons still reports that it is up-to-date.

Creating local copies is most useful for files that are being installed into an intermediate directory (for sharing with other directories) via the Install command. Accompanying the Install command for a file with a companion Local command is so common that Cons provides a Install_Local command as a convenient way to do both:

  Install_Local $env, '#export', 'hello';

is exactly equivalent to:

  Install $env '#export', 'hello';
  Local '#export/hello';

Both the Local and Install_Local commands update the local .consign file with the appropriate file signatures, so that future builds are performed correctly.

Repository dependency analysis

Due to its built-in scanning, Cons will search the specified repository trees for included .h files. Unless the compiler also knows about the repository trees, though, it will be unable to find .h files that only exist in a repository. If, for example, the hello.c file includes the hello.h file in its current directory:

  % cons -R /usr/all/repository hello
  gcc -c /usr/all/repository/hello.c -o hello.o
  /usr/all/repository/hello.c:1: hello.h: No such file or directory

Solving this problem forces some requirements onto the way construction environments are defined and onto the way the C #include preprocessor directive is used to include files.

In order to inform the compiler about the repository trees, Cons will add appropriate -I flags to the compilation commands. This means that the CPPPATH variable in the construct environment must explicitly specify all subdirectories which are to be searched for included files, including the current directory. Consequently, we can fix the above example by changing the environment creation in the Construct file as follows:

  $env = new cons(
        CC      => 'gcc',
        CPPPATH => '.',
        LIBS    => 'libworld.a',
  );

Due to the definition of the CPPPATH variable, this yields, when we re-execute the command:

  % cons -R /usr/all/repository hello
  gcc -c -I. -I/usr/all/repository /usr/all/repository/hello.c -o hello.o
  gcc -o hello hello.o /usr/all/repository/libworld.a

The order of the -I flags replicates, for the C preprocessor, the same repository-directory search path that Cons uses for its own dependency analysis. If there are multiple repositories and multiple CPPPATH directories, Cons will append the repository directories to the beginning of each CPPPATH directory, rapidly multiplying the number of -I flags. As an extreme example, a Construct file containing:

  Repository qw(
        /u1
        /u2
  );

  $env = new cons(
        CPPPATH => 'a:b:c',
  );

Would yield a compilation command of:

  cc -Ia -I/u1/a -I/u2/a -Ib -I/u1/b -I/u2/b -Ic -I/u1/c -I/u2/c -c hello.c -o hello.o

Because Cons relies on the compiler's -I flags to communicate the order in which repository directories must be searched, Cons' handling of repository directories is fundamentally incompatible with using double-quotes on the #include directives in your C source code:

  #include "file.h"     /* DON'T USE DOUBLE-QUOTES LIKE THIS */

This is because most C preprocessors, when faced with such a directive, will always first search the directory containing the source file. This undermines the elaborate -I options that Cons constructs to make the preprocessor conform to its preferred search path.

Consequently, when using repository trees in Cons, always use angle-brackets for included files:

  #include <file.h>     /* USE ANGLE-BRACKETS INSTEAD */

Repository_List

Cons provides a Repository_List command to return a list of all repository directories in their current search order. This can be used for debugging, or to do more complex Perl stuff:

  @list = Repository_List;
  print join(' ', @list), "\n";

Repository interaction with other Cons features

Cons' handling of repository trees interacts correctly with other Cons features--which is to say, it generally does what you would expect.

Most notably, repository trees interact correctly, and rather powerfully, with the 'Link' command. A repository tree may contain one or more subdirectories for version builds established via Link to a source subdirectory. Cons will search for derived files in the appropriate build subdirectories under the repository tree.

Default targets

Until now, we've demonstrated invoking Cons with an explicit target to build:

  % cons hello

Normally, Cons does not build anything unless a target is specified, but specifying '.' (the current directory) will build everything:

  % cons                # does not build anything

  % cons .              # builds everything under the top-level directory

Adding the Default method to any Construct or Conscript file will add the specified targets to a list of default targets. Cons will build these defaults if there are no targets specified on the command line. So adding the following line to the top-level Construct file will mimic Make's typical behavior of building everything by default:

  Default '.';

The following would add the hello and goodbye commands (in the same directory as the Construct or Conscript file) to the default list:

  Default qw(
        hello
        goodbye
  );

The Default method may be used more than once to add targets to the default list.

Selective builds

Cons provides two methods for reducing the size of given build. The first is by specifying targets on the command line, and the second is a method for pruning the build tree. We'll consider target specification first.

Selective targeting

Like make, Cons allows the specification of ``targets'' on the command line. Cons targets may be either files or directories. When a directory is specified, this is simply a short-hand notation for every derivable product--that Cons knows about--in the specified directory and below. For example:

  % cons build/hello/hello.o

means build hello.o and everything that hello.o might need. This is from a previous version of the Hello, World! program in which hello.o depended upon export/include/world.h. If that file is not up-to-date (because someone modified src/world/world.h), then it will be rebuilt, even though it is in a directory remote from build/hello.

In this example:

  % cons build

Everything in the build directory is built, if necessary. Again, this may cause more files to be built. In particular, both export/include/world.h and export/lib/libworld.a are required by the build/hello directory, and so they will be built if they are out-of-date.

If we do, instead:

  % cons export

then only the files that should be installed in the export directory will be rebuilt, if necessary, and then installed there. Note that cons build might build files that cons export doesn't build, and vice-versa.

No ``special'' targets

With Cons, make-style ``special'' targets are not required. The simplest analog with Cons is to use special export directories, instead. Let's suppose, for example, that you have a whole series of unit tests that are associated with your code. The tests live in the source directory near the code. Normally, however, you don't want to build these tests. One solution is to provide all the build instructions for creating the tests, and then to install the tests into a separate part of the tree. If we install the tests in a top-level directory called tests, then:

  % cons tests

will build all the tests.

  % cons export

will build the production version of the system (but not the tests), and:

  % cons build

should probably be avoided (since it will compile tests unecessarily).

If you want to build just a single test, then you could explicitly name the test (in either the tests directory or the build directory). You could also aggregate the tests into a convenient hierarchy within the tests directory. This hierarchy need not necessarily match the source hierarchy, in much the same manner that the include hierarchy probably doesn't match the source hierarchy (the include hierarchy is unlikely to be more than two levels deep, for C programs).

If you want to build absolutely everything in the tree (subject to whatever options you select), you can use:

  % cons .

This is not particularly efficient, since it will redundantly walk all the trees, including the source tree. The source tree, of course, may have buildable objects in it--nothing stops you from doing this, even if you normally build in a separate build tree.

Build Pruning

In conjunction with target selection, build pruning can be used to reduce the scope of the build. In the previous peAcH and baNaNa example, we have already seen how script-driven build pruning can be used to make only half of the potential build available for any given invocation of cons. Cons also provides, as a convenience, a command line convention that allows you to specify which Conscript files actually get ``built''--that is, incorporated into the build tree. For example:

  % cons build +world

The + argument introduces a Perl regular expression. This must, of course, be quoted at the shell level if there are any shell meta-characters within the expression. The expression is matched against each Conscript file which has been mentioned in a Build statement, and only those scripts with matching names are actually incorporated into the build tree. Multiple such arguments are allowed, in which case a match against any of them is sufficient to cause a script to be included.

In the example, above, the hello program will not be built, since Cons will have no knowledge of the script hello/Conscript. The libworld.a archive will be built, however, if need be.

There are a couple of uses for build pruning via the command line. Perhaps the most useful is the ability to make local changes, and then, with sufficient knowledge of the consequences of those changes, restrict the size of the build tree in order to speed up the rebuild time. A second use for build pruning is to actively prevent the recompilation of certain files that you know will recompile due to, for example, a modified header file. You may know that either the changes to the header file are immaterial, or that the changes may be safely ignored for most of the tree, for testing purposes.With Cons, the view is that it is pragmatic to admit this type of behavior, with the understanding that on the next full build everything that needs to be rebuilt will be. There is no equivalent to a ``make touch'' command, to mark files as permanently up-to-date. So any risk that is incurred by build pruning is mitigated. For release quality work, obviously, we recommend that you do not use build pruning (it's perfectly OK to use during integration, however, for checking compilation, etc. Just be sure to do an unconstrained build before committing the integration).

Temporary overrides

Cons provides a very simple mechanism for overriding aspects of a build. The essence is that you write an override file containing one or more Override commands, and you specify this on the command line, when you run cons:

  % cons -o over export

will build the export directory, with all derived files subject to the overrides present in the over file. If you leave out the -o option, then everything necessary to remove all overrides will be rebuilt.

Overriding environment variables

The override file can contain two types of overrides. The first is incoming environment variables. These are normally accessible by the Construct file from the %ENV hash variable. These can trivially be overridden in the override file by setting the appropriate elements of %ENV (these could also be overridden in the user's environment, of course).

The Override command

The second type of override is accomplished with the Override command, which looks like this:

  Override <regexp>, <var1> => <value1>, <var2> => <value2>, ...;

The regular expression regexp is matched against every derived file that is a candidate for the build. If the derived file matches, then the variable/value pairs are used to override the values in the construction environment associated with the derived file.

Let's suppose that we have a construction environment like this:

  $CONS = new cons(
        COPT => '',
        CDBG => '-g',
        CFLAGS => '%COPT %CDBG',
  );

Then if we have an override file over containing this command:

  Override '\.o$', COPT => '-O', CDBG => '';

then any cons invocation with -o over that creates .o files via this environment will cause them to be compiled with -O and no -g. The override could, of course, be restricted to a single directory by the appropriate selection of a regular expression.

Here's the original version of the Hello, World! program, built with this environment. Note that Cons rebuilds the appropriate pieces when the override is applied or removed:

  % cons hello
  cc -g -c hello.c -o hello.o
  cc -o hello hello.o
  % cons -o over hello
  cc -O -c hello.c -o hello.o
  cc -o hello hello.o
  % cons -o over hello
  cons: "hello" is up-to-date.
  % cons hello
  cc -g -c hello.c -o hello.o
  cc -o hello hello.o

It's important that the Override command only be used for temporary, on-the-fly overrides necessary for development because the overrides are not platform independent and because they rely too much on intimate knowledge of the workings of the scripts. For temporary use, however, they are exactly what you want.

Note that it is still useful to provide, say, the ability to create a fully optimized version of a system for production use--from the Construct and Conscript files. This way you can tailor the optimized system to the platform. Where optimizer trade-offs need to be made (particular files may not be compiled with full optimization, for example), then these can be recorded for posterity (and reproducibility) directly in the scripts.

More on construction environments

Default construction variables

We have mentioned, and used, the concept of a construction environment, many times in the preceding pages. Now it's time to make this a little more concrete. With the following statement:

  $env = new cons();

a reference to a new, default construction environment is created. This contains a number of construction variables and some methods. At the present writing, the default list of construction variables is defined as follows:

  CC            => 'cc',
  CFLAGS        => '',
  CCCOM         => '%CC %CFLAGS %_IFLAGS -c %< -o %>',
  INCDIRPREFIX  => '-I',
  CXX           => '%CC',
  CXXFLAGS      => '%CFLAGS',
  CXXCOM        => '%CXX %CXXFLAGS %_IFLAGS -c %< -o %>',
  LINK          => '%CXX',
  LINKCOM       => '%LINK %LDFLAGS -o %> %< %_LDIRS %LIBS',
  LINKMODULECOM => '%LD -r -o %> %<',
  LIBDIRPREFIX  => '-L',
  AR            => 'ar',
  ARFLAGS       => 'r',
  ARCOM         => "%AR %ARFLAGS %> %<\n%RANLIB %>",
  RANLIB        => 'ranlib',
  AS            => 'as',
  ASFLAGS       => '',
  ASCOM         => '%AS %ASFLAGS %< -o %>',
  LD            => 'ld',
  LDFLAGS       => '',
  PREFLIB       => 'lib',
  SUFLIB        => '.a',
  SUFLIBS       => '.so:.a',
  SUFOBJ        => '.o',
  ENV           => { 'PATH' => '/bin:/usr/bin' },

On Win32 systems (Windows NT), the following construction variables are overridden in the default:

  CC            => 'cl',
  CFLAGS        => '/nologo',
  CCCOM         => '%CC %CFLAGS %_IFLAGS /c %< /Fo%>',
  CXXCOM        => '%CXX %CXXFLAGS %_IFLAGS /c %< /Fo%>',
  INCDIRPREFIX  => '/I',
  LINK          => 'link',
  LINKCOM       => '%LINK %LDFLAGS /out:%> %< %_LDIRS %LIBS',
  LINKMODULECOM => '%LD /r /o %> %<',
  LIBDIRPREFIX  => '/LIBPATH:',
  AR            => 'lib',
  ARFLAGS       => '/nologo ',
  ARCOM         => "%AR %ARFLAGS /out:%> %<",
  RANLIB        => '',
  LD            => 'link',
  LDFLAGS       => '/nologo ',
  PREFLIB       => '',
  SUFEXE        => '.exe',
  SUFLIB        => '.lib',
  SUFLIBS       => '.dll:.lib',
  SUFOBJ        => '.obj',

These variables are used by the various methods associated with the environment, in particular any method that ultimately invokes an external command will substitute these variables into the final command, as appropriate. For example, the Objects method takes a number of source files and arranges to derive, if necessary, the corresponding object files. For example:

  Objects $env 'foo.c', 'bar.c';

This will arrange to produce, if necessary, foo.o and bar.o. The command invoked is simply %CCCOM, which expands through substitution, to the appropriate external command required to build each object. We will explore the substitution rules further under the Command method, below.

The construction variables are also used for other purposes. For example, CPPPATH is used to specify a colon-separated path of include directories. These are intended to be passed to the C preprocessor and are also used by the C-file scanning machinery to determine the dependencies involved in a C Compilation. Variables beginning with underscore, are created by various methods, and should normally be considered ``internal'' variables. For example, when a method is called which calls for the creation of an object from a C source, the variable _IFLAGS is created: this corresponds to the -I switches required by the C compiler to represent the directories specified by CPPPATH.

Note that, for any particular environment, the value of a variable is set once, and then never reset (to change a variable, you must create a new environment. Methods are provided for copying existing environments for this purpose). Some internal variables, such as _IFLAGS are created on demand, but once set, they remain fixed for the life of the environment.

The CFLAGS, LDFLAGS, and ARFLAGS variables all supply a place for passing options to the compiler, loader, and archiver, respectively. Less obviously, the INCDIRPREFIX variable specifies the option string to be appended to the beginning of each include directory so that the compiler knows where to find .h files. Similarly, the LIBDIRPREFIX variable specifies the option string to be appended to the beginning of each directory that the linker should search for libraries.

Another variable, ENV, is used to determine the system environment during the execution of an external command. By default, the only environment variable that is set is PATH, which is the execution path for a UNIX command. For the utmost reproducibility, you should really arrange to set your own execution path, in your top-level Construct file (or perhaps by importing an appropriate construction package with the Perl use command). The default variables are intended to get you off the ground.

Interpolating construction variables

Construction environment variables may be interpolated in the source and target file names by prefixing the construction variable name with %.

  $env = new cons(
        DESTDIR =>      'programs',
        SRCDIR  =>      'src',
  );
  Program $env '%DESTDIR/hello', '%SRCDIR/hello.c';

Expansion of construction variables is recursive--that is, the file name(s) will be re-expanded until no more substitutions can be made. If a construction variable is not defined in the environment, then the null string will be substituted.

Default construction methods

The list of default construction methods includes the following:

The `new` constructor

The new method is a Perl object constructor. That is, it is not invoked via a reference to an existing construction environment reference, but, rather statically, using the name of the Perl package where the constructor is defined. The method is invoked like this:

  $env = new cons(<overrides>);

The environment you get back is blessed into the package cons, which means that it will have associated with it the default methods described below. Individual construction variables can be overridden by providing name/value pairs in an override list. Note that to override any command environment variable (i.e. anything under ENV), you will have to override all of them. You can get around this difficulty by using the copy method on an existing construction environment.

The `clone` method

The clone method creates a clone of an existing construction environment, and can be called as in the following example:

  $env2 = $env1->clone(<overrides>);

You can provide overrides in the usual manner to create a different environment from the original. If you just want a new name for the same environment (which may be helpful when exporting environments to existing components), you can just use simple assignment.

The `copy` method

The copy method extracts the externally defined construction variables from an environment and returns them as a list of name/value pairs. Overrides can also be provided, in which case, the overridden values will be returned, as appropriate. The returned list can be assigned to a hash, as shown in the prototype, below, but it can also be manipulated in other ways:

  %env = $env1->copy(<overrides>);

The value of ENV, which is itself a hash, is also copied to a new hash, so this may be changed without fear of affecting the original environment. So, for example, if you really want to override just the PATH variable in the default environment, you could do the following:

  %cons = new cons()->copy();
  $cons{ENV}{PATH} = "<your path here>";
  $cons = new cons(%cons);

This will leave anything else that might be in the default execution environment undisturbed.

The `Install` method

The Install method arranges for the specified files to be installed in the specified directory. The installation is optimized: the file is not copied if it can be linked. If this is not the desired behavior, you will need to use a different method to install the file. It is called as follows:

  Install $env <directory>, <names>;

Note that, while the files to be installed may be arbitrarily named, only the last component of each name is used for the installed target name. So, for example, if you arrange to install foo/bar in baz, this will create a bar file in the baz directory (not foo/bar).

The `InstallAs` method

The InstallAs method arranges for the specified source file(s) to be installed as the specified target file(s). Multiple files should be specified as a file list. The installation is optimized: the file is not copied if it can be linked. If this is not the desired behavior, you will need to use a different method to install the file. It is called as follows:

InstallAs works in two ways:

Single file install:

  InstallAs $env TgtFile, SrcFile;

Multiple file install:

  InstallAs $env ['tgt1', 'tgt2'], ['src1', 'src2'];

Or, even as:

  @srcs = qw(src1 src2 src3);
  @tgts = qw(tgt1 tgt2 tgt3);
  InstallAs $env [@tgts], [@srcs];

Both the target and the sources lists should be of the same length.

The `Precious` method

The Precious method asks cons not to delete the specified file or list of files before building them again. It is invoked as:

  Precious <files>;

This is especially useful for allowing incremental updates to libraries or debug information files which are updated rather than rebuilt anew each time. Cons will still delete the files when the -r flag is specified.

The `Command` method

The Command method is a catchall method which can be used to arrange for any external command to be called to update the target. For this command, a target file and list of inputs is provided. In addition a construction command line, or lines, is provided as a string (this string may have multiple commands embedded within it, separated by new lines). Command is called as follows:

  Command $env <target>, <inputs>, <construction command>;

The target is made dependent upon the list of input files specified, and the inputs must be built successfully or Cons will not attempt to build the target.

Within the construction command, any variable from the construction environment may be introduced by prefixing the name of the construction variable with %. This is recursive: the command is expanded until no more substitutions can be made. If a construction variable is not defined in the environment, then the null string will be substituted. A doubled %% will be replaced by a single % in the construction command.

There are several pseudo variables which will also be expanded:

%>

The target file name (in a multi-target command, this is always the first target mentioned).

%0

Same as %>.

%1, %2, ..., %9

These refer to the first through ninth input file, respectively.

%<

The full set of inputs. If any of these have been used anywhere else in the current command line (via %1, %2, etc.), then those will be deleted from the list provided by %<. Consider the following command found in a Conscript file in the test directory:

  Command $env 'tgt', qw(foo bar baz), qq(
        echo %< -i %1 > %>
        echo %< -i %2 >> %>
        echo %< -i %3 >> %>
  );

If tgt needed to be updated, then this would result in the execution of the following commands, assuming that no remapping has been established for the test directory:

  echo test/bar test/baz -i test/foo > test/tgt
  echo test/foo test/baz -i test/bar >> test/tgt
  echo test/foo test/bar -i test/baz >> test/tgt

Any of the above pseudo variables may be followed immediately by one of the following suffixes to select a portion of the expanded path name:

  :a    the absolute path to the file name
  :b    the directory plus the file name stripped of any suffix
  :d    the directory
  :f    the file name
  :s    the file name suffix
  :F    the file name stripped of any suffix

Continuing with the above example, %<:f would expand to foo bar baz, and %:d> would expand to test.

It is possible to programmatically rewrite part of the command by enclosing part of it between %[ and %]. This will call the construction variable named as the first word enclosed in the brackets as a Perl code reference; the results of this call will be used to replace the contents of the brackets in the command line. For example, given an existing input file named tgt.in:

  @keywords = qw(foo bar baz);
  $env = new cons(X_COMMA => sub { join(",", @_) });
  Command $env 'tgt', 'tgt.in', qq(
        echo '# Keywords: %[X_COMMA @keywords %]' > %>
        cat %< >> %>
  );

This will execute:

  echo '# Keywords: foo,bar,baz' > tgt
  cat tgt.in >> tgt

After substitution occurs, strings of white space are converted into single blanks, and leading and trailing white space is eliminated. It is therefore not possible to introduce variable length white space in strings passed into a command, without resorting to some sort of shell quoting.

If a multi-line command string is provided, the commands are executed sequentially. If any of the commands fails, then none of the rest are executed, and the target is not marked as updated, i.e. a new signature is not stored for the target.

Normally, if all the commands succeed, and return a zero status (or whatever platform-specific indication of success is required), then a new signature is stored for the target. If a command erroneously reports success even after a failure, then Cons will assume that the target file created by that command is accurate and up-to-date.

The first word of each command string, after expansion, is assumed to be an executable command looked up on the PATH environment variable (which is, in turn, specified by the ENV construction variable). If this command is found on the path, then the target will depend upon it: the command will therefore be automatically built, as necessary. It's possible to write multi-part commands to some shells, separated by semi-colons. Only the first command word will be depended upon, however, so if you write your command strings this way, you must either explicitly set up a dependency (with the Depends method), or be sure that the command you are using is a system command which is expected to be available. If it isn't available, you will, of course, get an error.

If any command (even one within a multi-line command) begins with [perl], the remainder of that command line will be evaluated by the running Perl instead of being forked by the shell. If an error occurs in parsing the Perl or if the Perl expression returns 0 or undef, the command will be considered to have failed. For example, here is a simple command which creates a file foo directly from Perl:

  $env = new cons();
  Command $env 'foo',
    qq([perl] open(FOO,'>foo');print FOO "hi\\n"; close(FOO); 1);

Note that when the command is executed, you are in the same package as when the Construct or Conscript file was read, so you can call Perl functions you've defined in the same Construct or Conscript file in which the Command appears:

  $env = new cons();
  sub create_file {
        my $file = shift;
        open(FILE, ">$file");
        print FILE "hi\n";
        close(FILE);
        return 1;
  }
  Command $env 'foo', "[perl] &create_file('%>')";

The Perl string will be used to generate the signature for the derived file, so if you change the string, the file will be rebuilt. The contents of any subroutines you call, however, are not part of the signature, so if you modify a called subroutine such as create_file above, the target will not be rebuilt. Caveat user.

Cons normally prints a command before executing it. This behavior is suppressed if the first character of the command is @. Note that you may need to separate the @ from the command name or escape it to prevent @cmd from looking like an array to Perl quote operators that perform interpolation:

  # The first command line is incorrect,
  # because "@cp" looks like an array
  # to the Perl qq// function.
  # Use the second form instead.
  Command $env 'foo', 'foo.in', qq(
        @cp %< tempfile
        @ cp tempfile %>
  );

If there are shell meta characters anywhere in the expanded command line, such as <, >, quotes, or semi-colon, then the command will actually be executed by invoking a shell. This means that a command such as:

  cd foo

alone will typically fail, since there is no command cd on the path. But the command string:

  cd $<:d; tar cf $>:f $<:f

when expanded will still contain the shell meta character semi-colon, and a shell will be invoked to interpret the command. Since cd is interpreted by this sub-shell, the command will execute as expected.

To specify a command with multiple targets, you can specify a reference to a list of targets. In Perl, a list reference can be created by enclosing a list in square brackets. Hence the following command:

  Command $env ['foo.h', 'foo.c'], 'foo.template', q(
        gen %1
  );

could be used in a case where the command gen creates two files, both foo.h and foo.c.

The `Objects` method

The Objects method arranges to create the object files that correspond to the specified source files. It is invoked as shown below:

  @files = Objects $env <source or object files>;

Under Unix, source files ending in .s and .c are currently supported, and will be compiled into a name of the same file ending in .o. By default, all files are created by invoking the external command which results from expanding the CCCOM construction variable, with %< and %> set to the source and object files, respectively (see the Command method for expansion details). The variable CPPPATH is also used when scanning source files for dependencies. This is a colon separated list of pathnames, and is also used to create the construction variable _IFLAGS, which will contain the appropriate list of -I options for the compilation. Any relative pathnames in CPPPATH is interpreted relative to the directory in which the associated construction environment was created (absolute and top-relative names may also be used). This variable is used by CCCOM. The behavior of this command can be modified by changing any of the variables which are interpolated into CCCOM, such as CC, CFLAGS, and, indirectly, CPPPATH. It's also possible to replace the value of CCCOM, itself. As a convenience, this file returns the list of object filenames.

The `Program` method

The Program method arranges to link the specified program with the specified object files. It is invoked in the following manner:

  Program $env <program name>, <source or object files>;

The program name will have the value of the SUFEXE construction variable appended (by default, .exe on Win32 systems, nothing on Unix systems) if the suffix is not already present.

Source files may be specified in place of objects files--the Objects method will be invoked to arrange the conversion of all the files into object files, and hence all the observations about the Objects method, above, apply to this method also.

The actual linking of the program will be handled by an external command which results from expanding the LINKCOM construction variable, with %< set to the object files to be linked (in the order presented), and %> set to the target (see the Command method for expansion details). The user may set additional variables in the construction environment, including LINK, to define which program to use for linking, LIBPATH, a colon-separated list of library search paths, for use with library specifications of the form -llib, and LIBS, specifying the list of libraries to link against (in either -llib form or just as pathnames. Relative pathnames in both LIBPATH and LIBS are interpreted relative to the directory in which the associated construction environment is created (absolute and top-relative names may also be used). Cons automatically sets up dependencies on any libraries mentioned in LIBS: those libraries will be built before the command is linked.

The `Library` method

The Library method arranges to create the specified library from the specified object files. It is invoked as follows:

  Library $env <library name>, <source or object files>;

The library name will have the value of the SUFLIB construction variable appended (by default, .lib on Win32 systems, .a on Unix systems) if the suffix is not already present.

The actual creation of the library will be handled by an external command which results from expanding the ARCOM construction variable, with %< set to the library members (in the order presented), and %> to the library to be created (see the Command method for expansion details). The user may set variables in the construction environment which will affect the operation of the command. These include AR, the archive program to use, ARFLAGS, which can be used to modify the flags given to the program specified by AR, and RANLIB, the name of a archive index generation program, if needed (if the particular need does not require the latter functionality, then ARCOM must be redefined to not reference RANLIB).

The Library method allows the same library to be specified in multiple method invocations. All of the contributing objects from all the invocations (which may be from different directories) are combined and generated by a single archive command. Note, however, that if you prune a build so that only part of a library is specified, then only that part of the library will be generated (the rest will disappear!).

The `Module` method

The Module method is a combination of the Program and Command methods. Rather than generating an executable program directly, this command allows you to specify your own command to actually generate a module. The method is invoked as follows:

  Module $env <module name>, <source or object files>, <construction command>;

This command is useful in instances where you wish to create, for example, dynamically loaded modules, or statically linked code libraries.

The `Depends` method

The Depends method allows you to specify additional dependencies for a target. It is invoked as follows:

  Depends $env <target>, <dependencies>;

This may be occasionally useful, especially in cases where no scanner exists (or is writable) for particular types of files. Normally, dependencies are calculated automatically from a combination of the explicit dependencies set up by the method invocation or by scanning source files.

A set of identical dependencies for multiple targets may be specified using a reference to a list of targets. In Perl, a list reference can be created by enclosing a list in square brackets. Hence the following command:

  Depends $env ['foo', 'bar'], 'input_file_1', 'input_file_2';

specifies that both the foo and bar files depend on the listed input files.

The `Ignore` method

The Ignore method allows you to ignore explicitly dependencies that Cons infers on its own. It is invoked as follows:

  Ignore <patterns>;

This can be used to avoid recompilations due to changes in system header files or utilities that are known to not affect the generated targets.

If, for example, a program is built in an NFS-mounted directory on multiple systems that have different copies of stdio.h, the differences will affect the signatures of all derived targets built from source files that #include <stdio.h>. This will cause all those targets to be rebuilt when changing systems. If this is not desirable behavior, then the following line will remove the dependencies on the stdio.h file:

  Ignore '^/usr/include/stdio\.h$';

Note that the arguments to the Ignore method are regular expressions, so special characters must be escaped and you may wish to anchor the beginning or end of the expression with ^ or $ characters.

The `Salt` method

The Salt method adds a constant value to the signature calculation for every derived file. It is invoked as follows:

  Salt $string;

Changing the Salt value will force a complete rebuild of every derived file. This can be used to force rebuilds in certain desired circumstances. For example,

  Salt `uname -s`;

Would force a complete rebuild of every derived file whenever the operating system on which the build is performed (as reported by uname -s) changes.

The `UseCache` method

The UseCache method instructs Cons to maintain a cache of derived files, to be shared among separate build trees of the same project.

  UseCache("cache/<buildname>") || warn("cache directory not found");

The `SourcePath` method

The SourcePath mathod returns the real source path name of a file, as opposted to the path name within a build directory. It is invoked as follows:

  $path = SourcePath <buildpath>;

The `ConsPath` method

The ConsPath method returns true if the supplied path is a derivable file, and returns undef (false) otherwise. It is invoked as follows:

  $result = ConsPath <path>;

The `SplitPath` method

The SplitPath method looks up multiple path names in a string separated by the default path separator for the operating system (':' on UNIX systems, ';' on Windows NT), and returns the fully-qualified names. It is invoked as follows:

  @paths = SplitPath <pathlist>;

The SplitPath method will convert names prefixed '#' to the appropriate top-level build name (without the '#') and will convert relative names to top-level names.

The `DirPath` method

The DirPath method returns the build path name(s) of a directory or list of directories. It is invoked as follows:

  $cwd = DirPath <paths>;

The most common use for the DirPath method is:

  $cwd = DirPath '.';

to fetch the path to the current directory of a subsidiary Conscript file.

The `FilePath` method

The FilePath method returns the build path name(s) of a file or list of files. It is invoked as follows:

  $file = FilePath <path>;

The `Help` method

The Help method specifies help text that will be displayed when the user invokes cons -h. This can be used to provide documentation of specific targets, values, build options, etc. for the build tree. It is invoked as follows:

  Help <helptext>;

The Help method may only be called once, and should typically be specified in the top-level Construct file.

Extending Cons

Overriding construction variables

There are several ways of extending Cons, which vary in degree of difficulty. The simplest method is to define your own construction environment, based on the default environment, but modified to reflect your particular needs. This will often suffice for C-based applications. You can use the new constructor, and the clone and copy methods to create hybrid environments. These changes can be entirely transparent to the underlying Conscript files.

Adding new methods

For slightly more demanding changes, you may wish to add new methods to the cons package. Here's an example of a very simple extension, InstallScript, which installs a tcl script in a requested location, but edits the script first to reflect a platform-dependent path that needs to be installed in the script:

  # cons::InstallScript - Create a platform dependent version of a shell
  # script by replacing string ``#!your-path-here'' with platform specific
  # path $BIN_DIR.

  sub cons::InstallScript {
        my ($env, $dst, $src) = @_;
        Command $env $dst, $src, qq(
                sed s+your-path-here+$BIN_DIR+ %< > %>
                chmod oug+x %>
        );
  }

Notice that this method is defined directly in the cons package (by prefixing the name with cons::). A change made in this manner will be globally visible to all environments, and could be called as in the following example:

  InstallScript $env "$BIN/foo", "foo.tcl";

For a small improvement in generality, the BINDIR variable could be passed in as an argument or taken from the construction environment--as %BINDIR.

Overriding methods

Instead of adding the method to the cons name space, you could define a new package which inherits existing methods from the cons package and overrides or adds others. This can be done using Perl's inheritance mechanisms.

The following example defines a new package cons::switch which overrides the standard Library method. The overridden method builds linked library modules, rather than library archives. A new constructor is provided. Environments created with this constructor will have the new library method; others won't.

  package cons::switch;
  BEGIN {@ISA = 'cons'}

  sub new {
        shift;
        bless new cons(@_);
  }

  sub Library {
        my($env) = shift;
        my($lib) = shift;
        my(@objs) = Objects $env @_;
        Command $env $lib, @objs, q(
                %LD -r %LDFLAGS %< -o %>
        );
  }

This functionality could be invoked as in the following example:

  $env = new cons::switch(@overrides);
  ...
  Library $env 'lib.o', 'foo.c', 'bar.c';

Invoking Cons

The cons command is usually invoked from the root of the build tree. A Construct file must exist in that directory. If the -f argument is used, then an alternate Construct file may be used (and, possibly, an alternate root, since cons will cd to Construct file's containing directory).

If cons is invoked from a child of the root of the build tree with the -t argument, it will walk up the directory hierarchy looking for a Construct file. (An alternate name may still be specified with -f.) The targets supplied on the command line will be modified to be relative to the discovered Construct file. For example, from a directory containing a top-level Construct file, the following invocation:

  % cd libfoo/subdir
  % cons -t target

is exactly equivalent to:

  % cons libfoo/subdir/target

If there are any Default targets specified in the directory hierarchy's Construct or Conscript files, only the default targets at or below the directory from which cons -t was invoked will be built.

The command is invoked as follows:

  cons <arguments> -- <construct-args>

where arguments can be any of the following, in any order:

target: Build the specified target. If target is a directory, then recursively build everything within that directory.
+pattern: Limit the Conscript files considered to just those that match pattern, which is a Perl regular expression. Multiple + arguments are accepted.
name=<val>: Sets name to value val in the ARG hash passed to the top-level Construct file.
-cc: Show command that would have been executed, when retrieving from cache. No indication that the file has been retrieved is given; this is useful for generating build logs that can be compared with real build logs.
-cd: Disable all caching. Do not retrieve from cache nor flush to cache.
-cr: Build dependencies in random order. This is useful when building multiple similar trees with caching enabled.
-cs: Synchronize existing build targets that are found to be up-to-date with cache. This is useful if caching has been disabled with -cc or just recently enabled with UseCache.
-d: Enable dependency debugging.
-f <file>: Use the specified file instead of Construct (but first change to containing directory of file).
-h: Show a help message local to the current build if one such is defined, and exit.
-k: Keep going as far as possible after errors.
-o <file>: Read override file file.
-p: Show construction products in specified trees. No build is attempted.
-pa: Show construction products and associated actions. No build is attempted.
-pw: Show products and where they are defined. No build is attempted.
-q: Don't be verbose about Installing and Removing targets.
-r: Remove construction products associated with <targets>. No build is attempted.
-R <repos>: Search for files in repos. Multiple -R repos directories are searched in the order specified.
-t: Traverse up the directory hierarchy looking for a Construct file, if none exists in the current directory. Targets will be modified to be relative to the Construct file.
-v: Show cons version and continue processing.
-V: Show cons version and exit.
-wf <file>: Write all filenames considered into file.
-x: Show a help message similar to this one, and exit.

And construct-args can be any arguments that you wish to process in the Construct file. Note that there should be a -- separating the arguments to cons and the arguments that you wish to process in the Construct file.

Processing of construct-args can be done by any standard package like Getopt or its variants, or any user defined package. cons will pass in the construct-args as @ARGV and will not attempt to interpret anything after the --.

  % cons -R /usr/local/repository -d os=solaris +driver -- -c test -f DEBUG

would pass the following to cons

  -R /usr/local/repository -d os=solaris +driver

and the following, to the top level Construct file as @ARGV

  -c test -f DEBUG

Note that cons -r . is equivalent to a full recursive make clean, but requires no support in the Construct file or any Conscript files. This is most useful if you are compiling files into source directories (if you separate the build and export directories, then you can just remove the directories).

The options -p, -pa, and -pw are extremely useful for use as an aid in reading scripts or debugging them. If you want to know what script installs export/include/foo.h, for example, just type:

  % cons -pw export/include/foo.h

Using and writing dependency scanners

QuickScan allows simple target-independent scanners to be set up for source files. Only one QuickScan scanner may be associated with any given source file and environment.

QuickScan is invoked as follows:

  QuickScan CONSENV CODEREF, FILENAME [, PATH]

The subroutine referenced by CODEREF is expected to return a list of filenames included directly by FILE. These filenames will, in turn, be scanned. The optional PATH argument supplies a lookup path for finding FILENAME and/or files returned by the user-supplied subroutine. The PATH may be a reference to an array of lookup-directory names, or a string of names separated by the system's separator character (':' on UNIX systems, ';' on Windows NT).

The subroutine is called once for each line in the file, with $_ set to the current line. If the subroutine needs to look at additional lines, or, for that matter, the entire file, then it may read them itself, from the filehandle SCAN. It may also terminate the loop, if it knows that no further include information is available, by closing the filehandle.

Whether or not a lookup path is provided, QuickScan first tries to lookup the file relative to the current directory (for the top-level file supplied directly to QuickScan), or from the directory containing the file which referenced the file. This is not very general, but seems good enough--especially if you have the luxury of writing your own utilities and can control the use of the search path in a standard way. Finally, the search path is, currently, colon separated. This may not make the NT camp happy.

Here's a real example, taken from a Construct file here:

  sub cons::SMFgen {
      my($env, @tables) = @_;
      foreach $t (@tables) {
          $env->QuickScan(sub { /\b\S*?\.smf\b/g }, "$t.smf",
                          $env->{SMF_INCLUDE_PATH});
          $env->Command(
              ["$t.smdb.cc","$t.smdb.h","$t.snmp.cc","$t.ami.cc", "$t.http.cc"],
              "$t.smf",
              q(
                smfgen %( %SMF_INCLUDE_OPT %) %<
              )
          );
      }
  }

[NOTE that the form $env->QuickScan ... and $env->Command ... should not be necessary, but, for some reason, is required for this particular invocation. This appears to be a bug in Perl or a misunderstanding on my part; this invocation style does not always appear to be necessary.]

This finds all names of the form <name>.smf in the file. It will return the names even if they're found within comments, but that's OK (the mechanism is forgiving of extra files; they're just ignored on the assumption that the missing file will be noticed when the program, in this example, smfgen, is actually invoked).

A scanner is only invoked for a given source file if it is needed by some target in the tree. It is only ever invoked once for a given source file.

Here is another way to build the same scanner. This one uses an explicit code reference, and also (unecessarily, in this case) reads the whole file itself:

  sub myscan {
      my(@includes);
      do {
          push(@includes, /\b\S*?\.smf\b/g);
      } while <SCAN>;
      @includes
  }

Note that the order of the loop is reversed, with the loop test at the end. This is because the first line is already read for you. This scanner can be attached to a source file by:

    QuickScan $env \myscan, "$_.smf";

SUPPORT AND SUGGESTIONS

Cons is maintained by the user community. To subscribe, send mail to cons-discuss-request@gnu.org with body subscribe.

Please report any suggestions through the cons-discuss@gnu.org mailing list.

BUGS

Sure to be some. Please report any bugs through the bug-cons@gnu.org mailing list.

INFORMATION ABOUT CONS

Information about CONS can be obtained from the official cons web site http://www.dsmit.com/cons/ or its mirrors listed there.

The cons maintainers can be contacted by email at cons-maintainers@gnu.org

AUTHORS

Originally by Bob Sidebotham. Then significantly enriched by the members of the Cons community cons-discuss@gnu.org.

The Cons community would like to thank Ulrich Pfeifer for the original pod documentation derived from the cons.html file. Cons documentation is now a part of the program itself.