summaryrefslogtreecommitdiff
path: root/doc/design/engine.xml
diff options
context:
space:
mode:
authorLuca Falavigna <dktrkranz@debian.org>2010-01-02 20:56:35 +0100
committerLuca Falavigna <dktrkranz@debian.org>2010-01-02 20:56:35 +0100
commit64c458487151933ee0ba093cf4ac69e177d9be37 (patch)
treef6e3755704f53406eea85532e4ffe5d5ef50b7f0 /doc/design/engine.xml
parent2aec9cc58398cac1376509a7d75edb83b41f984e (diff)
parent72c578fd4b0b4a5a43e18594339ac4ff26c376dc (diff)
Merge commit 'upstream/1.2.0.d20091224'
Diffstat (limited to 'doc/design/engine.xml')
-rw-r--r--doc/design/engine.xml1964
1 files changed, 1964 insertions, 0 deletions
diff --git a/doc/design/engine.xml b/doc/design/engine.xml
new file mode 100644
index 0000000..1a1e335
--- /dev/null
+++ b/doc/design/engine.xml
@@ -0,0 +1,1964 @@
+<!--
+
+ Copyright (c) 2001, 2002, 2003 Steven Knight
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice shall be included
+ in all copies or substantial portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY
+ KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE
+ WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+-->
+
+<section id="sect-principles">
+ <title>General Principles</title>
+
+ <section>
+ <title>Keyword arguments</title>
+
+ <para>
+
+ All methods and functions in this API will support the use of keyword
+ arguments in calls, for the sake of explicitness and readability.
+ For brevity in the hands of experts, most methods and functions
+ will also support positional arguments for their most-commonly-used
+ arguments. As an explicit example, the following two lines will each
+ arrange for an executable program named <filename>foo</filename> (or
+ <filename>foo.exe</filename> on a Win32 system) to be compiled from
+ the <filename>foo.c</filename> source file:
+
+ </para>
+
+ <programlisting>
+ env.Program(target = 'foo', source = 'foo.c')
+
+ env.Program('foo', 'foo.c')
+ </programlisting>
+
+ </section>
+
+ <section>
+ <title>Internal object representation</title>
+
+ <para>
+
+ All methods and functions use internal (Python) objects that
+ represent the external objects (files, for example) for which they
+ perform dependency analysis.
+
+ </para>
+
+ <para>
+
+ All methods and functions in this API that accept an external object
+ as an argument will accept <emphasis>either</emphasis> a string
+ description or an object reference. For example, the two following
+ two-line examples are equivalent:
+
+ </para>
+
+ <programlisting>
+ env.Object(target = 'foo.o', source = 'foo.c')
+ env.Program(target = 'foo', 'foo.o') # builds foo from foo.o
+
+ foo_obj = env.Object(target = 'foo.o', source = 'foo.c')
+ env.Program(target = 'foo', foo_obj) # builds foo from foo.o
+ </programlisting>
+
+ </section>
+
+</section>
+
+
+
+<section id="sect-envs">
+ <title>&ConsEnvs</title>
+
+ <para>
+
+ A &consenv; is the basic means by which a software system interacts
+ with the &SCons; Python API to control a build process.
+
+ </para>
+
+ <para>
+
+ A &consenv; is an object with associated methods for generating target
+ files of various types (&Builder; objects), other associated object
+ methods for automatically determining dependencies from the contents
+ of various types of source files (&Scanner; objects), and a dictionary
+ of values used by these methods.
+
+ </para>
+
+ <para>
+
+ Passing no arguments to the &Environment; instantiation creates a
+ &consenv; with default values for the current platform:
+
+ </para>
+
+ <programlisting>
+ env = Environment()
+ </programlisting>
+
+ <section>
+ <title>&Consvars;</title>
+
+ <para>
+
+ A &consenv; has an associated dictionary of &consvars; that control how
+ the build is performed. By default, the &Environment; method creates
+ a &consenv; with values that make most software build "out of the box"
+ on the host system. These default values will be generated at the
+ time &SCons; is installed using functionality similar to that provided
+ by GNU &Autoconf;.
+ <footnote>
+ <para>
+ It would be nice if we could avoid re-inventing the wheel here by
+ using some other Python-based tool &Autoconf replacement--like what
+ was supposed to come out of the Software Carpentry configuration
+ tool contest. It will probably be most efficient to roll our own
+ logic initially and convert if something better does come along.
+ </para>
+ </footnote>
+ At a minimum, there will be pre-configured sets of default values
+ that will provide reasonable defaults for UNIX and Windows NT.
+
+ </para>
+
+ <para>
+
+ The default &consenv; values may be overridden when a new &consenv; is
+ created by specifying keyword arguments:
+
+ </para>
+
+ <programlisting>
+ env = Environment(CC = 'gcc',
+ CCFLAGS = '-g',
+ CPPPATH = ['.', 'src', '/usr/include'],
+ LIBPATH = ['/usr/lib', '.'])
+ </programlisting>
+
+ </section>
+
+ <section>
+ <title>Fetching &consvars;</title>
+
+ <para>
+
+ A copy of the dictionary of &consvars; can be returned using
+ the &Dictionary; method:
+
+ </para>
+
+ <programlisting>
+ env = Environment()
+ dict = env.Dictionary()
+ </programlisting>
+
+ <para>
+
+ If any arguments are supplied, then just the corresponding value(s)
+ are returned:
+
+ </para>
+
+ <programlisting>
+ ccflags = env.Dictionary('CCFLAGS')
+ cc, ld = env.Dictionary('CC', 'LD')
+ </programlisting>
+
+ </section>
+
+ <section>
+ <title>Copying a &consenv;</title>
+
+ <para>
+
+ A method exists to return a copy of an existing environment, with
+ any overridden values specified as keyword arguments to the method:
+
+ </para>
+
+ <programlisting>
+ env = Environment()
+ debug = env.Copy(CCFLAGS = '-g')
+ </programlisting>
+
+ </section>
+
+ <section>
+ <title>Multiple &consenvs;</title>
+
+ <para>
+
+ Different external objects often require different build
+ characteristics. Multiple &consenvs; may be defined, each with
+ different values:
+
+ </para>
+
+ <programlisting>
+ env = Environment(CCFLAGS = '')
+ debug = Environment(CCFLAGS = '-g')
+ env.Make(target = 'hello', source = 'hello.c')
+ debug.Make(target = 'hello-debug', source = 'hello.c')
+ </programlisting>
+
+ <para>
+
+ Dictionaries of values from multiple &consenvs; may be passed to the
+ &Environment; instantiation or the &Copy; method, in which case the
+ last-specified dictionary value wins:
+
+ </para>
+
+ <programlisting>
+ env1 = Environment(CCFLAGS = '-O', LDFLAGS = '-d')
+ env2 = Environment(CCFLAGS = '-g')
+ new = Environment(env1.Dictionary(), env2.Dictionary())
+ </programlisting>
+
+ <para>
+
+ The <varname>new</varname> environment in the above example retains
+ <literal>LDFLAGS = '-d'</literal> from the <varname>env1</varname>
+ environment, and <literal>CCFLAGS = '-g'</literal> from the
+ <varname>env2</varname> environment.
+
+ </para>
+
+ <!--
+
+ hardware details
+ current directory
+ OS environment variables
+ compilers and options,
+ aliases for commands,
+ versions of tools
+
+ environment overrides a la Cons
+
+ compilation options
+
+ cross compilation via selection of tool+options
+
+ paths for header files (specify alternate path)
+
+ accomodate smart compilers that can tell you
+ "I know how to turn .c or .ccp into .o",
+ "I know how to turn .f into .o"
+
+ -->
+
+ </section>
+
+ <section>
+ <title>Variable substitution</title>
+
+ <para>
+
+ Within a construction command, any variable from the &consenv; may
+ be interpolated by prefixing the name of the construction with
+ <symbol>$</symbol>:
+
+ </para>
+
+ <programlisting>
+ MyBuilder = Builder(command = "$XX $XXFLAGS -c $_INPUTS -o $target")
+
+ env.Command(targets = 'bar.out', sources = 'bar.in',
+ command = "sed '1d' < $source > $target")
+ </programlisting>
+
+ <para>
+
+ Variable substitution is recursive: the command line is expanded
+ until no more substitutions can be made.
+
+ </para>
+
+ <para>
+
+ Variable names following the <symbol>$</symbol> may be enclosed in
+ braces. This can be used to concatenate an interpolated value with an
+ alphanumeric character:
+
+ </para>
+
+ <programlisting>
+ VerboseBuilder = Builder(command = "$XX -${XXFLAGS}v > $target")
+ </programlisting>
+
+ <para>
+
+ The variable within braces may contain a pair of parentheses
+ after a Python function name to be evaluated (for example,
+ <literal>${map()}</literal>). &SCons; will interpolate the return
+ value from the function (presumably a string):
+
+ </para>
+
+ <programlisting>
+ env = Environment(FUNC = myfunc)
+ env.Command(target = 'foo.out', source = 'foo.in',
+ command = "${FUNC($<)}")
+ </programlisting>
+
+ <para>
+
+ If a referenced variable is not defined in the &consenv;,
+ the null string is interpolated.
+
+ </para>
+
+ <para>
+
+ The following special variables can also be used:
+
+ </para>
+
+ <variablelist>
+
+ <varlistentry>
+ <term><literal>$targets</literal></term>
+ <listitem>
+ <para>
+
+ All target file names. If multiple targets are specified in an
+ array, <literal>$targets</literal> expands to the entire list of
+ targets, separated by a single space.
+
+ </para>
+
+ <para>
+
+ Individual targets from a list may be extracted by enclosing
+ the <literal>targets</literal> keyword in braces and using the
+ appropriate Python array index or slice:
+
+ </para>
+
+ <programlisting>
+ ${targets[0]} # expands to the first target
+
+ ${targets[1:]} # expands to all but the first target
+
+ ${targets[1:-1]} # expands to all but the first and last targets
+ </programlisting>
+
+ </listitem>
+ </varlistentry>
+
+ <varlistentry>
+ <term><literal>$target</literal></term>
+ <listitem>
+ <para>
+
+ A synonym for <literal>${targets[0]}</literal>, the first target
+ specified.
+
+ </para>
+ </listitem>
+ </varlistentry>
+
+ <varlistentry>
+ <term><literal>$sources</literal></term>
+ <listitem>
+ <para>
+
+ All input file names. Any input file names that
+ are used anywhere else on the current command
+ line (via <literal>${sources[0]}</literal>,
+ <literal>${sources{[1]}</literal>, etc.) are removed from the
+ expanded list.
+
+ </para>
+ </listitem>
+ </varlistentry>
+
+ </variablelist>
+
+ <para>
+
+ Any of the above special variables may be enclosed in braces and
+ followed immediately by one of the following attributes to select just
+ a portion of the expanded path name:
+
+ </para>
+
+ <variablelist>
+
+ <varlistentry>
+ <term><literal>.base</literal></term>
+ <listitem>
+ <para>
+
+ Basename: the directory plus the file name, minus any file suffix.
+
+ </para>
+ </listitem>
+ </varlistentry>
+
+ <varlistentry>
+ <term><literal>.dir</literal></term>
+ <listitem>
+ <para>
+
+ The directory in which the file lives. This is a relative path,
+ where appropriate.
+
+ </para>
+ </listitem>
+ </varlistentry>
+
+ <varlistentry>
+ <term><literal>.file</literal></term>
+ <listitem>
+ <para>
+
+ The file name, minus any directory portion.
+
+ </para>
+ </listitem>
+ </varlistentry>
+
+ <varlistentry>
+ <term><literal>.suffix</literal></term>
+ <listitem>
+ <para>
+
+ The file name suffix (that is, the right-most dot in the file name,
+ and all characters to the right of that).
+
+ </para>
+ </listitem>
+ </varlistentry>
+
+ <varlistentry>
+ <term><literal>.filebase</literal></term>
+ <listitem>
+ <para>
+
+ The file name (no directory portion), minus any file suffix.
+
+ </para>
+ </listitem>
+ </varlistentry>
+
+ <varlistentry>
+ <term><literal>.abspath</literal></term>
+ <listitem>
+ <para>
+
+ The absolute path to the file.
+
+ </para>
+ </listitem>
+ </varlistentry>
+
+ </variablelist>
+
+ </section>
+
+</section>
+
+
+
+<section id="sect-builders">
+ <title>&Builder; Objects</title>
+
+ <para>
+
+ By default, &SCons; supplies (and uses) a number of pre-defined
+ &Builder; objects:
+
+ </para>
+
+ <informaltable>
+ <tgroup cols="2">
+ <tbody>
+
+ <row>
+ <entry>&Object;</entry>
+ <entry>compile or assemble an object file</entry>
+ </row>
+
+ <row>
+ <entry>&Library;</entry>
+ <entry>archive files into a library</entry>
+ </row>
+
+ <row>
+ <entry>&SharedLibrary;</entry>
+ <entry>archive files into a shared library</entry>
+ </row>
+
+ <row>
+ <entry>&Program;</entry>
+ <entry>link objects and/or libraries into an executable</entry>
+ </row>
+
+ <row>
+ <entry>&MakeBuilder;</entry>
+ <entry>build according to file suffixes; see below</entry>
+ </row>
+
+ </tbody>
+ </tgroup>
+ </informaltable>
+
+<!--
+&Library; and &SharedLibrary; have nearly identical
+semantics, just different
+tools and &consenvs (paths, etc.) that they use.
+In other words, you can construct a shared library
+using just the &Library; &Builder; object
+with a different environment.
+I think that's a better way to do it.
+Feedback?
+-->
+
+ <para>
+
+ A &consenv; can be explicitly initialized with associated &Builder;
+ objects that will be bound to the &consenv; object:
+
+ </para>
+
+ <programlisting>
+ env = Environment(BUILDERS = ['Object', 'Program'])
+ </programlisting>
+
+ <para>
+
+ &Builder; objects bound to a &consenv; can be called directly as
+ methods. When invoked, a &Builder; object returns a (list of) objects
+ that it will build:
+
+ </para>
+
+ <programlisting>
+ obj = env.Object(target ='hello.o', source = 'hello.c')
+ lib = env.Library(target ='libfoo.a',
+ source = ['aaa.c', 'bbb.c'])
+ slib = env.SharedLibrary(target ='libbar.so',
+ source = ['xxx.c', 'yyy.c'])
+ prog = env.Program(target ='hello',
+ source = ['hello.o', 'libfoo.a', 'libbar.so'])
+ </programlisting>
+
+ <section>
+ <title>Specifying multiple inputs</title>
+
+ <para>
+
+ Multiple input files that go into creating a target file may be passed
+ in as a single string, with the individual file names separated by
+ white space:
+
+ </para>
+
+ <programlisting>
+ env.Library(target = 'foo.a', source = 'aaa.c bbb.c ccc.c')
+ env.Object(target = 'yyy.o', source = 'yyy.c')
+ env.Program(target = 'bar', source = 'xxx.c yyy.o foo.a')
+ </programlisting>
+
+ <para>
+
+ Alternatively, multiple input files that go into creating a target
+ file may be passed in as an array. This allows input files to be
+ specified using their object representation:
+
+ </para>
+
+ <programlisting>
+ env.Library(target = 'foo.a', source = ['aaa.c', 'bbb.c', 'ccc.c'])
+ yyy_obj = env.Object(target = 'yyy.o', source = 'yyy.c')
+ env.Program(target = 'bar', source = ['xxx.c', yyy_obj, 'foo.a'])
+ </programlisting>
+
+ <para>
+
+ Individual string elements within an array of input files are
+ <emphasis>not</emphasis> further split into white-space separated
+ file names. This allows file names that contain white space to
+ be specified by putting the value into an array:
+
+ <programlisting>
+ env.Program(target = 'foo', source = ['an input file.c'])
+ </programlisting>
+
+ </para>
+
+ </section>
+
+ <section>
+ <title>Specifying multiple targets</title>
+
+ <para>
+
+ Conversely, the generated target may be a string listing multiple
+ files separated by white space:
+
+ </para>
+
+ <programlisting>
+ env.Object(target = 'grammar.o y.tab.h', source = 'grammar.y')
+ </programlisting>
+
+ <para>
+
+ An array of multiple target files can be used to mix string and object
+ representations, or to accomodate file names that contain white space:
+
+ </para>
+
+ <programlisting>
+ env.Program(target = ['my program'], source = 'input.c')
+ </programlisting>
+
+ </section>
+
+ <section>
+ <title>File prefixes and suffixes</title>
+
+ <para>
+
+ For portability, if the target file name does not already have an
+ appropriate file prefix or suffix, the &Builder; objects will
+ append one appropriate for the file type on the current system:
+
+ </para>
+
+ <programlisting>
+ # builds 'hello.o' on UNIX, 'hello.obj' on Windows NT:
+ obj = env.Object(target ='hello', source = 'hello.c')
+
+ # builds 'libfoo.a' on UNIX, 'foo.lib' on Windows NT:
+ lib = env.Library(target ='foo', source = ['aaa.c', 'bbb.c'])
+
+ # builds 'libbar.so' on UNIX, 'bar.dll' on Windows NT:
+ slib = env.SharedLibrary(target ='bar', source = ['xxx.c', 'yyy.c'])
+
+ # builds 'hello' on UNIX, 'hello.exe' on Windows NT:
+ prog = env.Program(target ='hello',
+ source = ['hello.o', 'libfoo.a', 'libbar.so'])
+ </programlisting>
+
+ </section>
+
+ <section>
+ <title>&Builder; object exceptions</title>
+
+ <para>
+
+ &Builder; objects raise the following exceptions on error:
+
+ <!--
+ LIST THESE ONCE WE FIGURE OUT WHAT THEY ARE FROM CODING THEM.
+ -->
+
+ </para>
+ </section>
+
+ <section>
+ <title>User-defined &Builder; objects</title>
+
+ <para>
+
+ Users can define additional &Builder; objects for specific external
+ object types unknown to &SCons;. A &Builder; object may build its
+ target by executing an external command:
+
+ </para>
+
+ <programlisting>
+ WebPage = Builder(command = 'htmlgen $HTMLGENFLAGS $sources > $target',
+ suffix = '.html',
+ src_suffix = '.in')
+ </programlisting>
+
+ <para>
+
+ Alternatively, a &Builder; object may also build its target by
+ executing a Python function:
+
+ </para>
+
+ <programlisting>
+ def update(dest):
+ # [code to update the object]
+ return 1
+
+ OtherBuilder1 = Builder(function = update,
+ src_suffix = ['.in', '.input'])
+ </programlisting>
+
+ <para>
+
+ An optional argument to pass to the function may be specified:
+
+ </para>
+
+ <programlisting>
+ def update_arg(dest, arg):
+ # [code to update the object]
+ return 1
+
+ OtherBuilder2 = Builder(function = update_arg,
+ function_arg = 'xyzzy',
+ src_suffix = ['.in', '.input'])
+ </programlisting>
+
+ <para>
+
+ Both an external command and an internal function may be specified,
+ in which case the function will be called to build the object first,
+ followed by the command line.
+
+ </para>
+
+ <!--
+ NEED AN EXAMPLE HERE.
+ -->
+
+ <para>
+
+ User-defined &Builder; objects can be used like the default &Builder;
+ objects to initialize &consenvs;.
+
+ </para>
+
+ <programlisting>
+ WebPage = Builder(command = 'htmlgen $HTMLGENFLAGS $sources > $target',
+ suffix = '.html',
+ src_suffix = '.in')
+ env = Environment(BUILDERS = ['WebPage'])
+ env.WebPage(target = 'foo.html', source = 'foo.in')
+ # Builds 'bar.html' on UNIX, 'bar.htm' on Windows NT:
+ env.WebPage(target = 'bar', source = 'bar.in')
+ </programlisting>
+
+ <para>
+
+ The command-line specification can interpolate variables from the
+ &consenv;; see "Variable substitution," above.
+
+ </para>
+
+ <para>
+
+ A &Builder; object may optionally be initialized with a list of:
+
+ </para>
+
+ <itemizedlist>
+ <listitem>
+ <para>
+
+ the prefix of the target file (e.g., 'lib' for libraries)
+
+ </para>
+ </listitem>
+
+ <listitem>
+ <para>
+
+ the suffix of the target file (e.g., '.a' for libraries)
+
+ </para>
+ </listitem>
+
+ <listitem>
+ <para>
+
+ the expected suffixes of the input files
+ (e.g., '.o' for object files)
+
+ </para>
+ </listitem>
+ </itemizedlist>
+
+ <para>
+
+ These arguments are used in automatic
+ dependency analysis and to generate output file names that don't
+ have suffixes supplied explicitly.
+
+ </para>
+ </section>
+
+ <section>
+ <title>Copying &Builder; Objects</title>
+
+ <para>
+
+ A &Copy; method exists to return a copy of an existing &Builder;
+ object, with any overridden values specified as keyword arguments to
+ the method:
+
+ </para>
+
+ <programlisting>
+ build = Builder(function = my_build)
+ build_out = build.Copy(suffix = '.out')
+ </programlisting>
+
+ <para>
+
+ Typically, &Builder; objects will be supplied by a tool-master or
+ administrator through a shared &consenv;.
+
+ </para>
+ </section>
+
+ <section>
+ <title>Special-purpose build rules</title>
+
+ <para>
+
+ A pre-defined &Command; builder exists to associate a target file with
+ a specific command or list of commands for building the file:
+
+ </para>
+
+ <programlisting>
+ env.Command(target = 'foo.out', source =
+ command = 'foo.in', "foo.process $sources > $target")
+
+ commands = [ "bar.process -o .tmpfile $sources",
+ "mv .tmpfile $target" ]
+ env.Command(target = 'bar.out', source = 'bar.in', command = commands)
+ </programlisting>
+
+ <para>
+ This is useful when it's too cumbersome to create a &Builder;
+ object just to build a single file in a special way.
+
+ </para>
+ </section>
+
+ <section>
+ <title>The &MakeBuilder; &Builder;</title>
+
+ <para>
+
+ A pre-defined &Builder; object named &MakeBuilder; exists to make
+ simple builds as easy as possible for users, at the expense of
+ sacrificing some build portability.
+
+ </para>
+
+ <para>
+
+ The following minimal example builds the 'hello' program from the
+ 'hello.c' source file:
+
+ </para>
+
+ <programlisting>
+ Environment().Make('hello', 'hello.c')
+ </programlisting>
+
+ <para>
+
+ Users of the &MakeBuilder; &Builder; object are not required to
+ understand intermediate steps involved in generating a file--for
+ example, the distinction between compiling source code into an object
+ file, and then linking object files into an executable. The details
+ of intermediate steps are handled by the invoked method. Users that
+ need to, however, can specify intermediate steps explicitly:
+
+ </para>
+
+ <programlisting>
+ env = Environment()
+ env.Make(target = 'hello.o', source = 'hello.c')
+ env.Make(target = 'hello', source = 'hello.o')
+ </programlisting>
+
+ <para>
+
+ The &MakeBuilder; method understands the file suffixes specified and
+ "does the right thing" to generate the target object and program
+ files, respectively. It does this by examining the specified output
+ suffixes for the &Builder; objects bound to the environment.
+
+ </para>
+
+ <para>
+
+ Because file name suffixes in the target and source file names
+ must be specified, the &MakeBuilder; method can't be used
+ portably across operating systems. In other words, for the
+ example above, the &MakeBuilder; builder will not generate
+ <filename>hello.exe</filename> on Windows NT.
+
+ </para>
+
+ </section>
+
+ <section>
+ <title>&Builder; maps</title>
+
+<!--
+Do we even need this anymore?
+Now that the individual builders
+have specified <literal>suffix</literal>
+and <literal>src_suffix</literal> values,
+all of the information we need to support
+the &MakeBuilder; builder is right there in the environment.
+I think this is a holdover from before I
+added the <literal>suffix</literal> arguments.
+If you want &MakeBuilder; to do something different,
+you set it up with another environment...
+-->
+
+ <para>
+
+ The <function>env.Make</function> method "does the right thing" to
+ build different file types because it uses a dictionary from the
+ &consenv; that maps file suffixes to the appropriate &Builder; object.
+ This &BUILDERMAP; can be initialized at instantiation:
+
+ </para>
+
+ <programlisting>
+ env = Environment(BUILDERMAP = {
+ '.o' : Object,
+ '.a' : Library,
+ '.html' : WebPage,
+ '' : Program,
+ })
+ </programlisting>
+
+ <para>
+
+ With the &BUILDERMAP; properly initialized, the
+ <function>env.Make</function> method can be used to build additional
+ file types:
+
+ </para>
+
+ <programlisting>
+ env.Make(target = 'index.html', source = 'index.input')
+ </programlisting>
+
+ <para>
+
+ &Builder; objects referenced in the &BUILDERMAP; do not need to be
+ listed separately in the &BUILDERS; variable. The &consenv; will
+ bind the union of the &Builder; objects listed in both variables.
+
+ </para>
+
+ <!--
+
+ YYY support scanners which detect files which haven't been generated yet
+
+ -->
+
+ </section>
+
+</section>
+
+
+
+<section id="sect-deps">
+ <title>Dependencies</title>
+
+ <section>
+ <title>Automatic dependencies</title>
+
+ <para>
+
+ By default, &SCons; assumes that a target file has <literal>automatic
+ dependencies</literal> on the:
+
+ </para>
+
+ <blockquote>
+ <simplelist>
+
+ <member>tool used to build the target file</member>
+
+ <member>contents of the input files</member>
+
+ <member>command line used to build the target file</member>
+
+ </simplelist>
+ </blockquote>
+
+ <para>
+
+ If any of these changes, the target file will be rebuilt.
+
+ </para>
+ </section>
+
+ <section>
+ <title>Implicit dependencies</title>
+
+ <para>
+
+ Additionally, &SCons; can scan the contents of files for
+ <literal>implicit dependencies</literal> on other files. For
+ example, &SCons; will scan the contents of a <filename>.c</filename>
+ file and determine that any object created from it is
+ dependent on any <filename>.h</filename> files specified via
+ <literal>#include</literal>. &SCons;, therefore, "does the right
+ thing" without needing to have these dependencies listed explicitly:
+
+ </para>
+
+ <programlisting>
+ % cat Construct
+ env = Environment()
+ env.Program('hello', 'hello.c')
+ % cat hello.c
+ #include "hello_string.h"
+ main()
+ {
+ printf("%s\n", STRING);
+ }
+ % cat > hello_string.h
+ #define STRING "Hello, world!\n"
+ % scons .
+ gcc -c hello.c -o hello.o
+ gcc -o hello hello.c
+ % ./hello
+ Hello, world!
+ % cat > hello_string.h
+ #define STRING "Hello, world, hello!\n"
+ % scons .
+ gcc -c hello.c -o hello.o
+ gcc -o hello hello.c
+ % ./hello
+ Hello, world, hello!
+ %
+ </programlisting>
+
+ </section>
+
+ <section>
+ <title>Ignoring dependencies</title>
+
+ <para>
+
+ Undesirable <literal>automatic dependencies</literal> or
+ <literal>implicit dependencies</literal> may be ignored:
+
+ </para>
+
+ <programlisting>
+ env.Program(target = 'bar', source = 'bar.c')
+ env.Ignore('bar', '/usr/bin/gcc', 'version.h')
+ </programlisting>
+
+ <para>
+
+ In the above example, the <filename>bar</filename> program will not
+ be rebuilt if the <filename>/usr/bin/gcc</filename> compiler or the
+ <filename>version.h</filename> file change.
+
+ </para>
+ </section>
+
+ <section>
+ <title>Explicit dependencies</title>
+
+ <para>
+
+ Dependencies that are unknown to &SCons; may be specified explicitly
+ in an &SCons; configuration file:
+
+ </para>
+
+ <programlisting>
+ env.Dependency(target = 'output1', dependency = 'input_1 input_2')
+ env.Dependency(target = 'output2', dependency = ['input_1', 'input_2'])
+ env.Dependency(target = 'output3', dependency = ['white space input'])
+
+ env.Dependency(target = 'output_a output_b', dependency = 'input_3')
+ env.Dependency(target = ['output_c', 'output_d'], dependency = 'input_4')
+ env.Dependency(target = ['white space output'], dependency = 'input_5')
+ </programlisting>
+
+ <para>
+
+ Just like the <literal>target</literal> keyword argument, the
+ <literal>dependency</literal> keyword argument may be specified as a
+ string of white-space separated file names, or as an array.
+
+ </para>
+
+ <para>
+
+ A dependency on an &SCons; configuration file itself may be specified
+ explicitly to force a rebuild whenever the configuration file changes:
+
+ </para>
+
+ <programlisting>
+ env.Dependency(target = 'archive.tar.gz', dependency = 'SConstruct')
+ </programlisting>
+
+ </section>
+
+</section>
+
+
+
+<section id="sect-scanners">
+ <title>&Scanner; Objects</title>
+
+ <para>
+
+ Analagous to the previously-described &Builder; objects, &SCons;
+ supplies (and uses) &Scanner; objects to search the contents of
+ a file for implicit dependency files:
+
+ </para>
+
+ <informaltable>
+ <tgroup cols="2">
+ <tbody>
+
+ <row>
+ <entry>CScan</entry>
+ <entry>scan .{c,C,cc,cxx,cpp} files for #include dependencies</entry>
+ </row>
+
+ </tbody>
+ </tgroup>
+ </informaltable>
+
+ <para>
+
+ A &consenv; can be explicitly initialized with
+ associated &Scanner; objects:
+
+ </para>
+
+ <programlisting>
+ env = Environment(SCANNERS = ['CScan', 'M4Scan'])
+ </programlisting>
+
+ <para>
+
+ &Scanner; objects bound to a &consenv; can be
+ associated directly with specified files:
+
+ </para>
+
+ <programlisting>
+ env.CScan('foo.c', 'bar.c')
+ env.M4Scan('input.m4')
+ </programlisting>
+
+ <section>
+ <title>User-defined &Scanner; objects</title>
+
+ <para>
+
+ A user may define a &Scanner; object to scan a type of file for
+ implicit dependencies:
+
+ </para>
+
+ <programlisting>
+ def scanner1(file_contents):
+ # search for dependencies
+ return dependency_list
+
+ FirstScan = Scanner(function = scanner1)
+ </programlisting>
+
+ <para>
+
+ The scanner function must return a list of dependencies that its finds
+ based on analyzing the file contents it is passed as an argument.
+
+ </para>
+
+ <para>
+
+ The scanner function, when invoked, will be passed the calling
+ environment. The scanner function can use &consenvs; from the passed
+ environment to affect how it performs its dependency scan--the
+ canonical example being to use some sort of search-path construction
+ variable to look for dependency files in other directories:
+
+ </para>
+
+ <programlisting>
+ def scanner2(file_contents, env):
+ path = env.{'SCANNERPATH'} # XXX
+ # search for dependencies using 'path'
+ return dependency_list
+
+ SecondScan = Scanner(function = scanner2)
+ </programlisting>
+
+ <para>
+
+ The user may specify an additional argument when the &Scanner; object
+ is created. When the scanner is invoked, the additional argument
+ will be passed to the scanner funciton, which can be used in any way
+ the scanner function sees fit:
+
+ </para>
+
+ <programlisting>
+ def scanner3(file_contents, env, arg):
+ # skip 'arg' lines, then search for dependencies
+ return dependency_list
+
+ Skip_3_Lines_Scan = Scanner(function = scanner2, argument = 3)
+ Skip_6_Lines_Scan = Scanner(function = scanner2, argument = 6)
+ </programlisting>
+
+ </section>
+
+ <section>
+ <title>Copying &Scanner; Objects</title>
+
+ <para>
+
+ A method exists to return a copy of an existing &Scanner; object,
+ with any overridden values specified as keyword arguments to the
+ method:
+
+ </para>
+
+ <programlisting>
+ scan = Scanner(function = my_scan)
+ scan_path = scan.Copy(path = '%SCANNERPATH')
+ </programlisting>
+
+ <para>
+
+ Typically, &Scanner; objects will be supplied by a tool-master or
+ administrator through a shared &consenv;.
+
+ </para>
+ </section>
+
+ <section>
+ <title>&Scanner; maps</title>
+
+<!--
+If the &BUILDERMAP; proves unnecessary,
+we could/should get rid of this one, too,
+by adding a parallel <literal>src_suffix</literal>
+argument to the &Scanner; factory...
+Comments?
+-->
+
+ <para>
+
+ Each &consenv; has a &SCANNERMAP;, a dictionary that associates
+ different file suffixes with a scanner object that can be used to
+ generate a list of dependencies from the contents of that file. This
+ &SCANNERMAP; can be initialized at instantiation:
+
+ </para>
+
+ <programlisting>
+ env = Environment(SCANNERMAP = {
+ '.c' : CScan,
+ '.cc' : CScan,
+ '.m4' : M4Scan,
+ })
+ </programlisting>
+
+ <para>
+
+ &Scanner; objects referenced in the &SCANNERMAP; do not need to
+ be listed separately in the &SCANNERS; variable. The &consenv;
+ will bind the union of the &Scanner; objects listed
+ in both variables.
+
+ </para>
+
+ </section>
+
+</section>
+
+
+
+<section id="sect-targets">
+ <title>Targets</title>
+
+ <para>
+
+ The methods in the build engine API described so far merely
+ establish associations that describe file dependencies, how a
+ file should be scanned, etc. Since the real point is to actually
+ <emphasis>build</emphasis> files, &SCons; also has methods that
+ actually direct the build engine to build, or otherwise manipulate,
+ target files.
+
+ </para>
+
+ <section>
+ <title>Building targets</title>
+ <para>
+
+ One or more targets may be built as follows:
+
+ </para>
+
+ <programlisting>
+ env.Build(target = ['foo', 'bar'])
+ </programlisting>
+
+ <para>
+
+ Note that specifying a directory (or other collective object) will
+ cause all subsidiary/dependent objects to be built as well:
+
+ </para>
+
+ <programlisting>
+ env.Build(target = '.')
+
+ env.Build(target = 'builddir')
+ </programlisting>
+
+ <para>
+
+ By default, &SCons; explicitly removes a target file before
+ invoking the underlying function or command(s) to build it.
+
+ </para>
+ </section>
+
+ <section>
+ <title>Removing targets</title>
+
+ <para>
+
+ A "cleanup" operation of removing generated (target) files is
+ performed as follows:
+
+ </para>
+
+ <programlisting>
+ env.Clean(target = ['foo', 'bar'])
+ </programlisting>
+
+ <para>
+
+ Like the &Build; method, the &Clean; method may be passed a
+ directory or other collective object, in which case the subsidiary
+ target objects under the directory will be removed:
+
+ </para>
+
+ <programlisting>
+ env.Clean(target = '.')
+
+ env.Clean(target = 'builddir')
+ </programlisting>
+
+ <para>
+
+ (The directories themselves are not removed.)
+
+ </para>
+ </section>
+
+ <section>
+ <title>Suppressing cleanup removal of build-targets</title>
+
+ <para>
+
+ By default, &SCons; explicitly removes all build-targets
+ when invoked to perform "cleanup". Files that should not be
+ removed during "cleanup" can be specified via the
+ &NoClean; method:
+
+ </para>
+
+ <programlisting>
+ env.Library(target = 'libfoo.a', source = ['aaa.c', 'bbb.c', 'ccc.c'])
+ env.NoClean('libfoo.a')
+ </programlisting>
+
+ <para>
+
+ The NoClean operation has precedence over the Clean operation.
+ A target that is specified as both Clean and NoClean, will not
+ be removed during a clean.
+
+ In the following example, target 'foo' will not be removed
+ during "cleanup":
+
+ <programlisting>
+ env.Clean(target = 'foo')
+ env.NoClean('foo')
+ </programlisting>
+
+
+ </para>
+
+ </section>
+
+ <section>
+ <title>Suppressing build-target removal</title>
+
+ <para>
+
+ As mentioned, by default, &SCons; explicitly removes a target
+ file before invoking the underlying function or command(s) to build
+ it. Files that should not be removed before rebuilding can be
+ specified via the &Precious; method:
+
+ </para>
+
+ <programlisting>
+ env.Library(target = 'libfoo.a', source = ['aaa.c', 'bbb.c', 'ccc.c'])
+ env.Precious('libfoo.a')
+ </programlisting>
+
+ </section>
+
+ <section>
+ <title>Default targets</title>
+
+ <para>
+
+ The user may specify default targets that will be built if there are no
+ targets supplied on the command line:
+
+ </para>
+
+ <programlisting>
+ env.Default('install', 'src')
+ </programlisting>
+
+ <para>
+
+ Multiple calls to the &Default; method (typically one per &SConscript;
+ file) append their arguments to the list of default targets.
+
+ </para>
+ </section>
+
+ <section>
+ <title>File installation</title>
+
+ <para>
+
+ Files may be installed in a destination directory:
+
+ </para>
+
+ <programlisting>
+ env.Install('/usr/bin', 'program1', 'program2')
+ </programlisting>
+
+ <para>
+
+ Files may be renamed on installation:
+
+ </para>
+
+ <programlisting>
+ env.InstallAs('/usr/bin/xyzzy', 'xyzzy.in')
+ </programlisting>
+
+ <para>
+
+ Multiple files may be renamed on installation by specifying
+ equal-length lists of target and source files:
+
+ </para>
+
+ <programlisting>
+ env.InstallAs(['/usr/bin/foo', '/usr/bin/bar'],
+ ['foo.in', 'bar.in'])
+ </programlisting>
+
+ </section>
+
+ <section>
+ <title>Target aliases</title>
+
+ <para>
+
+ In order to provide convenient "shortcut" target names that expand to
+ a specified list of targets, aliases may be established:
+
+ </para>
+
+ <programlisting>
+ env.Alias(alias = 'install',
+ targets = ['/sbin', '/usr/lib', '/usr/share/man'])
+ </programlisting>
+
+ <para>
+
+ In this example, specifying a target of <literal>install</literal>
+ will cause all the files in the associated directories to be built
+ (that is, installed).
+
+ </para>
+
+ <para>
+
+ An &Alias; may include one or more other &Aliases; in its list:
+
+ </para>
+
+ <programlisting>
+ env.Alias(alias = 'libraries', targets = ['lib'])
+ env.Alias(alias = 'programs', targets = ['libraries', 'src'])
+ </programlisting>
+
+ </section>
+
+</section>
+
+
+
+<section id="sect-custom">
+ <title>Customizing output</title>
+
+<!--
+Take this whole section with a grain of salt.
+I whipped it up without a great deal of thought
+to try to add a "competitive advantage"
+for the second round of the Software Carpentry contest.
+In particular, hard-coding the
+analysis points and the keywords that specify them
+feels inflexible,
+but I can't think of another way it would be
+done effectively.
+I dunno, maybe this is fine as it is...
+-->
+
+ <para>
+
+ The &SCons; API supports the ability to customize, redirect, or
+ suppress its printed output through user-defined functions.
+ &SCons; has several pre-defined points in its build process at
+ which it calls a function to (potentially) print output. User-defined
+ functions can be specified for these call-back points when &Build;
+ or &Clean;is invoked:
+
+ </para>
+
+ <programlisting>
+ env.Build(target = '.',
+ on_analysis = dump_dependency,
+ pre_update = my_print_command,
+ post_update = my_error_handler)
+ on_error = my_error_handler)
+ </programlisting>
+
+ <para>
+
+ The specific call-back points are:
+
+ </para>
+
+ <variablelist>
+
+ <varlistentry>
+ <term><literal>on_analysis</literal></term>
+ <listitem>
+ <para>
+
+ Called for every object, immediately after the object has been
+ analyzed to see if it's out-of-date. Typically used to print a
+ trace of considered objects for debugging of unexpected dependencies.
+
+ </para>
+ </listitem>
+ </varlistentry>
+
+ <varlistentry>
+ <term><literal>pre_update</literal></term>
+ <listitem>
+ <para>
+
+ Called for every object that has been determined to be out-of-date
+ before its update function or command is executed. Typically used
+ to print the command being called to update a target.
+
+ </para>
+ </listitem>
+ </varlistentry>
+
+ <varlistentry>
+ <term><literal>post_update</literal></term>
+ <listitem>
+ <para>
+
+ Called for every object after its update function or command has
+ been executed. Typically used to report that a top-level specified
+ target is up-to-date or was not remade.
+
+ </para>
+ </listitem>
+ </varlistentry>
+
+ <varlistentry>
+ <term><literal>on_error</literal></term>
+ <listitem>
+ <para>
+
+ Called for every error returned by an update function or command.
+ Typically used to report errors with some string that will be
+ identifiable to build-analysis tools.
+
+ </para>
+ </listitem>
+ </varlistentry>
+
+ </variablelist>
+
+ <para>
+
+ Functions for each of these call-back points all take the same
+ arguments:
+
+ </para>
+
+ <programlisting>
+ my_dump_dependency(target, level, status, update, dependencies)
+ </programlisting>
+
+ <para>
+
+ where the arguments are:
+
+ </para>
+
+ <variablelist>
+
+ <varlistentry>
+ <term><literal>target</literal></term>
+ <listitem>
+ <para>
+
+ The target object being considered.
+
+ </para>
+ </listitem>
+ </varlistentry>
+
+ <varlistentry>
+ <term><literal>level</literal></term>
+ <listitem>
+ <para>
+
+ Specifies how many levels the dependency analysis has
+ recursed in order to consider the <literal>target</literal>.
+ A value of <literal>0</literal> specifies a top-level
+ <literal>target</literal> (that is, one passed to the
+ &Build; or &Clean; method). Objects which a top-level
+ <literal>target</literal> is directly dependent upon have a
+ <literal>level</literal> of <1>, their direct dependencies have a
+ <literal>level</literal> of <2>, etc. Typically used to indent
+ output to reflect the recursive levels.
+
+ </para>
+ </listitem>
+ </varlistentry>
+
+ <varlistentry>
+ <term><literal>status</literal></term>
+ <listitem>
+ <para>
+
+ A string specifying the current status of the target
+ (<literal>"unknown"</literal>, <literal>"built"</literal>,
+ <literal>"error"</literal>, <literal>"analyzed"</literal>, etc.). A
+ complete list will be enumerated and described during implementation.
+
+ </para>
+ </listitem>
+ </varlistentry>
+
+ <varlistentry>
+ <term><literal>update</literal></term>
+ <listitem>
+ <para>
+
+ The command line or function name that will be (or has been) executed
+ to update the <literal>target</literal>.
+
+ </para>
+ </listitem>
+ </varlistentry>
+
+ <varlistentry>
+ <term><literal>dependencies</literal></term>
+ <listitem>
+ <para>
+
+ A list of direct dependencies of the target.
+
+ </para>
+ </listitem>
+ </varlistentry>
+
+ </variablelist>
+
+</section>
+
+
+
+<section id="separate">
+ <title>Separate source and build trees</title>
+
+<!--
+I've never liked Cons' use of the name <literal>Link</literal>
+for this functionality,
+mainly because the term is overloaded
+with linking object files into an executable.
+Yet I've never come up with anything better.
+Any suggestions?
+-->
+
+<!--
+Also, I made this an &Environment; method because
+it logically belongs in the API reference
+(the build engine needs to know about it),
+and I thought it was clean to have
+everything in the build-engine API
+be called through an &Environment; object.
+But <literal>&Link</literal> isn't really
+associated with a specific environment
+(the &Cons; classic implementation just
+leaves it as a bare function call),
+so maybe we should just follow that example
+and not call it through an environment...
+-->
+
+ <para>
+
+ &SCons; allows target files to be built completely separately from
+ the source files by "linking" a build directory to an underlying
+ source directory:
+
+ </para>
+
+ <programlisting>
+ env.Link('build', 'src')
+
+ SConscript('build/SConscript')
+ </programlisting>
+
+ <para>
+
+ &SCons; will copy (or hard link) necessary files (including the
+ &SConscript; file) into the build directory hierarchy. This allows the
+ source directory to remain uncluttered by derived files.
+
+ </para>
+
+</section>
+
+
+
+<section id="sect-variant">
+ <title>Variant builds</title>
+
+ <para>
+
+ The &Link; method may be used in conjunction with multiple
+ &consenvs; to support variant builds. The following
+ &SConstruct; and &SConscript; files would build separate debug and
+ production versions of the same program side-by-side:
+
+ </para>
+
+ <programlisting>
+ % cat SConstruct
+ env = Environment()
+ env.Link('build/debug', 'src')
+ env.Link('build/production', 'src')
+ flags = '-g'
+ SConscript('build/debug/SConscript', Export(env))
+ flags = '-O'
+ SConscript('build/production/SConscript', Export(env))
+ % cat src/SConscript
+ env = Environment(CCFLAGS = flags)
+ env.Program('hello', 'hello.c')
+ </programlisting>
+
+ <para>
+
+ The following example would build the appropriate program for the current
+ compilation platform, without having to clean any directories of object
+ or executable files for other architectures:
+
+ </para>
+
+ <programlisting>
+ % cat SConstruct
+ build_platform = os.path.join('build', sys.platform)
+ Link(build_platform, 'src')
+ SConscript(os.path.join(build_platform, 'SConscript'))
+ % cat src/SConscript
+ env = Environment
+ env.Program('hello', 'hello.c')
+ </programlisting>
+
+</section>
+
+
+
+<section id="sect-repositories">
+ <title>Code repositories</title>
+
+<!--
+Like &Link;, &Repository; and &Local; are part of the
+API reference, but not really tied to any specific environment.
+Is it better to be consistent about calling
+everything in the API through an environment,
+or to leave these independent so as
+not to complicate their calling interface?
+-->
+
+ <para>
+
+ &SCons; may use files from one or more shared code repositories in order
+ to build local copies of changed target files. A repository would
+ typically be a central directory tree, maintained by an integrator,
+ with known good libraries and executables.
+
+ </para>
+
+ <programlisting>
+ Repository('/home/source/1.1', '/home/source/1.0')
+ </programlisting>
+
+ <para>
+
+ Specified repositories will be searched in-order for any file
+ (configuration file, input file, target file) that does not exist
+ in the local directory tree. When building a local target file,
+ &SCons; will rewrite path names in the build command to use the
+ necessary repository files. This includes modifying lists of
+ <option>-I</option> or <option>-L</option> flags to specify an
+ appropriate set of include paths for dependency analysis.
+
+ </para>
+ <para>
+
+ &SCons; will modify the Python <varname>sys.path</varname> variable to
+ reflect the addition of repositories to the search path, so that any
+ imported modules or packages necessary for the build can be found in a
+ repository, as well.
+
+ </para>
+ <para>
+
+ If an up-to-date target file is found in a code repository, the file
+ will not be rebuilt or copied locally. Files that must exist locally
+ (for example, to run tests) may be specified:
+
+ </para>
+
+ <programlisting>
+ Local('program', 'libfoo.a')
+ </programlisting>
+
+ <para>
+
+ in which case &SCons; will copy or link an up-to-date copy of the
+ file from the appropriate repository.
+
+ </para>
+
+</section>
+
+
+
+<section id="sect-caching">
+ <title>Derived-file caching</title>
+
+<!--
+There should be extensions to this part of the API for
+auxiliary functions like cleaning the cache.
+-->
+
+ <para>
+
+ &SCons; can maintain a cache directory of target files which may be
+ shared among multiple builds. This reduces build times by allowing
+ developers working on a project together to share common target
+ files:
+
+ </para>
+
+ <programlisting>
+ Cache('/var/tmp/build.cache/i386')
+ </programlisting>
+
+ <para>
+
+ When a target file is generated, a copy is added to the cache.
+ When generating a target file, if &SCons; determines that a file
+ that has been built with the exact same dependencies already exists
+ in the specified cache, &SCons; will copy the cached file rather
+ than re-building the target.
+
+ </para>
+ <para>
+
+ Command-line options exist to modify the &SCons; caching behavior
+ for a specific build, including disabling caching, building
+ dependencies in random order, and displaying commands as if cached
+ files were built.
+
+ </para>
+
+</section>
+
+
+
+<section id="sect-jobs">
+ <title>Job management</title>
+
+<!--
+This has been completely superseded by
+the more sophisticated &Task; manager
+that Anthony Roach has contributed.
+I need to write that up...
+-->
+
+ <para>
+
+ A simple API exists to inform the Build Engine how many jobs may
+ be run simultaneously:
+
+ </para>
+
+ <programlisting>
+ Jobs(limit = 4)
+ </programlisting>
+
+</section>