Ubuntu Manpage: makepp_statements -- Various statements in a makefile

NAME

       makepp_statements -- Various statements in a makefile

DESCRIPTION

A: autoload, B: build_cache,
build_check, D: "define", E: export, G: global, I: ifdef,
ifeq,
ifmakeperl,
ifndef,
ifneq,
ifnsys,
ifntrue,
ifperl,
ifsys,
iftrue,
-include,
include,
_include, L: load_makefile, M: make,
makeperl,
makesub, N: "no_implicit_load", P: perl,
"perl_begin",
prebuild, R: register_command_parser,
register_input_suffix,
register_parser,
repository,
runtime, S: signature,
"sub", V: vpath

A statement is any line beginning with a word which does not have a ":" in it. (A colon
implies that the line is a rule.) For example, these are statements:

include extra_rules.mk
load_makefile subdir

Makepp has a number of builtin statements which you may occasionally need to use.

Note that wherever you see an underscore, you may also use a dash, because makepp converts
dashes to underscores in statement names.

Conditionals
Conditionals are special statements, which control what lines of the Makeppfile are
actually seen. The simplest form (where "ifxxx" stands for any of the conditional
statements documented below) is:

ifxxx ...
lines seen if the statement evaluates as true
endif

or:

ifxxx ...
lines seen if the statement evaluates as true
else
lines seen if the statement evaluates as false
endif

There is also the possibility to do complex combinations like this:

ifxxx ...
and ifxxx ...
and ifxxx ...
or ifxxx ...
and ifxxx ...
lines seen if the combined statements evaluate as true
else ifxxx ...
or ifxxx ...
and ifxxx ...
lines seen if the first combination evaluates as false
and these combined statements evaluate as true
else
lines seen if the statements above evaluate as false
endif

As is suggested by the indentation, "and" has higher precedence than "or". In other words
an "or" elects between two groups of "and"`s. There may be any number of "and ifxxx"`s,
"or ifxxx"`s and "else ifxxx"`s.

The "ifxxx" conditional statements are unique in that they may occur in the middle of rule
actions, as in the above example, without disrupting the rule.

ifeq string1, string2
ifneq string1, string2
ifeq ($(STR1),$(STR2))
makefile lines if true
else
makefile lines if false
endif

If the two strings match exactly (except for leading or trailing whitespace), then the
first set of lines is used; otherwise the second is used. The else clause is
optional.

There are two other acceptable syntaxes for the "ifeq" and "ifneq" statements:

ifeq string1, string2
ifeq string1 string2

which are equivalent. Of course you can quote the strings as needed.

"ifeq" and its friends "ifneq", "ifdef", "ifndef", "ifperl", "ifmakeperl", "ifsys" and
"iftrue" are primarily useful when you have to build a program under several different
conditions. For example,

BUILD_TYPE := debug # "debug" or "production"

ifeq ($(BUILD_TYPE), debug)
CFLAGS := -g
else
CFLAGS := -O2
endif

program : *.o
$(CC) $(CFLAGS) $(inputs) -o $(output) $(LIBS)
ifeq ($(BUILD_TYPE), production)
strip $(output)
endif

%.o : %.c
$(CC) $(CFLAGS) -c $(input) -o $(output)

If this is a production build, all files are compiled with the "-O2" option instead of
the "-g" option. Furthermore, the program "strip" is run on the resulting binary (in
case you happened to link with some libraries that were compiled in debug mode).

Sometimes it is easier to use the "$(if)" function or "$(perl)" function function
instead of a "ifeq" statement.

If you just want to see whether a symbol is blank or not, you only need to supply a
single argument, like this:

ifneq $(EXE_SUFFIX)
# what to do if $(EXE_SUFFIX) is not blank
endif

ifdef VARIABLE ...
ifndef VARIABLE ...
These statements work analogously to the "ifeq" and "ifneq" statements, except that
they test whether any of the variables is defined or not any is (i.e. none is
defined). A variable is defined if:

• It was given a value with an assignment earlier in the makefile. See
makepp_variables for details.

• It was given a value as a Perl variable in a "perl_begin" block.

• The variable is present in the environment.

• The variable is present on the command line, e.g., to invoke your makefile, you
typed

makepp CFLAGS=-O2

For example,

ifndef CFLAGS
CFLAGS := -g
endif

In this case, "CFLAGS" is set to "-g" only if it wasn't already defined. Note that
this statement could just as easily have been written using the "?=" assignment, like
this:

CFLAGS ?= -g

ifperl perlcode
ifmakeperl perlcode
These statements work analogously to the "ifeq" and "ifneq" statements, except that
the tests are in Perl. The first variant is plain Perl code, while the second variant
first passes the statement through Make-style variable expansion.

VERSION := 3.0
# VERSION is automatically also a Perl variable:
ifperl $VERSION <= 2
CPPFLAGS := -DNEW
endif
# quotes necessary for CFLAGS, because Perl sees only the value:
ifmakeperl my $$x = '$(CFLAGS)'; $$x =~ /-g/
CFLAGS := -g -O2
endif

ifsys wildcard ...
ifnsys wildcard ...
Tests if the current system makepp is running on matches any of the wildcards or not
any (i.e. none).

ifsys i[3-6]86
and ifsys Linux SunOS
... # An Intel platform with Linux or Solaris
else ifnsys sparc power*
... # Nor Sparc or PowerPC
endif

There are up to six different strings you can match against. The actual strings are
not standardized. Three of them reflect what the Perl instance was built for (not
necessarily the same as where it is running), and the others come from the system and
all vary wildly in form. You can find all of what the current platform matches by
typing the following command at the Shell:

perl -MConfig -e'print "$^O @Config{qw(archname myarchname)} "'; uname -mps

iftrue expression
ifntrue expression
Tests if the expression evaluates to some value other than zero or the empty string.

Other Multiline Statements
Conditionals may control a whole multiline statement, but they cannot be inside such a
statement.

define
{export|global|override}* define
define VARIABLE [assignment-operator]
variable value line 1
variable value line 2
endef

Defines $(VARIABLE)'s value to be all the lines between the "define" statement and the
"endef" statement. See multiline variables. The keywords "export" and "global" may
not be given at the same time.

perl_begin
This is the same as "perl", but using GNU make style statement syntax. This statement
introduces a block of code which is interpreted verbatim by perl. It can be useful
for defining functions, but you can do this more concisely with the "sub" statement.
A block of Perl code in your makefile can be useful to perform actions that are easier
in Perl than with makepp functions and rules.

The remainder of the line following the "perl_begin" statement is ignored. All text
up until a line that begins at the left margin with "perl_end" is sent verbatim to the
perl interpreter. There can be no spaces before "perl_end".

One example that I use this for is to make directories that might not necessarily
exist. It's common in makefiles to put all the .o files in a subdirectory (e.g., a
directory with a name i386, or sparc, or something that depends on the machine type).
But what if the directory does not exist yet? You can make each .o file depend on the
subdirectory, and put a rule in to build the subdirectory. But it's a lot easier just
to do this:

OBJDIR := $(ARCH) # Where we put .o files.
perl_begin
-d $OBJDIR or mkdir $OBJDIR; # Make sure the directory exists.
perl_end

This way, every time the makefile is run, the subdirectory will be created if it does
not exist.

Some operations are better expressed in terms of regular expressions than makepp's
text functions. For example,

perl_begin
if ($ARCH =~ /^i[56]86/) { # You could do this with: ifsys i[56]86
$CFLAGS = '-O6 -malign-double'; # On intel machines > 486, there
# is a substantial speed penalty
# for doubles that aren't quadword
# aligned.
} else {
$CFLAGS = '-O6';
}
perl_end

%.o: %.c
$(CC) $(CFLAGS) -c $(input) -o $(output)

Any make variable can be accessed directly as a Perl scalar. In this case, we've set
the value of "CFLAGS" differently based on a regular expression match on the
architecture flags.

As a final example, some pieces of information are easier to access directly from Perl
than from makepp. For example, you can access all of the configuration information
that perl knows about your system, including how to build shared libraries, etc.
(Type "perldoc Config" if you want to see what configuration information Perl has
available.)

perl_begin

use Config;

$ARCH = $Config{'archname'}; # Use perl's knowledge of the architecture.
$CC = $Config{'cc'}; # Use the same C compiler as Perl did.
$SHARED_OBJ_CFLAGS = $Config{'cccdlflags'};
# Flags needed to compile objects which will
# go into a shared library.
$SHARED_OBJ_LDFLAGS = $Config{'ccdlflags'} . " " . $Config{'lddlflags'};
# Linker flags to make a shared library.
$SHARED_CC_LINK = $Config{'ld'}; # Command to produce shared libraries.

$SHARED_EXTENSION = $Config{'dlext'}; # Extension of shared libraries.
perl_end

%.o: %.c
$(CC) $(CFLAGS) $(SHARED_OBJ_CFLAGS) -c $(input) -o $(output)

libmylib.$(DLEXT): *.o
$(SHARED_CC_LINK) $(inputs) -o $(output) $(SHARED_OBJ_LDFLAGS)

Note how we define a bunch of variables in the Perl block, and then we use them
afterwards in the rest of the makefile. You can use the full power of the perl
interpreter to set your variables in arbitrarily complicated ways. You can run shell
commands from your Perl code, access a database, or whatever you want.

perl perlcode
makeperl perlcode
This is the same as "perl_begin", but using Perl-style braces. The first variant is
plain Perl code, while the second variant first passes the statement through Make-
style variable expansion. Note that the difficulty of parsing Perl's braces has lead
to the following simple heuristic:

• If a double opening brace is found on the same or next line, a double closing
brace will terminate the block. It must be at the beginning of a line, but may be
preceded by whitespace.

• Else, if the closing brace is at the very end of the "perl" line this is a one
liner.

• Otherwise the closing brace must be at the very beginning of a following line,
i.e. no leading whitespace.

For an efficient way to call Perl scripts, see "run". Unlike the "$(perl)" function,
the return value of this block is ignored.

perl { print "passed this point in the makefile\n" }

perl
{
print "and this one too\n";
}

ifdef NOISY
perl {{
print "as well as this one\n"
}}
endif

You can use the Perl debugger for your embedded code, by running makepp itself in the
debugger, where ... are the arguments, if any, you normally pass:

perl -d -S mpp ...

It is hard to set breakpoints in Perl code that has not been loaded. You can work
around this by putting this line into your embedded Perl, just before where you want
to break:

$DB::single = 1;

Then you can type "c" at the debugger's prompt, to continue till that point.

sub
makesub
This statement provides a way to define a Perl subroutine inside your makefile. The
first variant is plain Perl code, while the second variant first passes the statement
through Make-style variable expansion. The syntax is identical to that of the Perl
sub statement, except that prototypes are meaningless.

For the three possibilities of putting the braces of the body, see the explanation at
the "perl" statement.

A Perl subroutine is invoked whenever a statement is seen, or when an expression like
"$(name words)" is seen. For example, suppose that for some reason you need to load
the contents of a file into a make variable. (You could do this by saying
"$(shell cat filename)" but it's possible to do it without ever invoking the shell.)
This can be done by placing the following into your makefile:

sub f_file_contents {
my ($file) = @_; # Name the argument.
open my $fh, $file or die "$file: $!\n";
local $/ = undef; # Slurp file in one read.
<$fh>;
}

ifdef NEWSUB
makesub f_VAR2
{{
$(VAR) * 2;
}}
endif

makesub f_VAR1 { $(VAR) + 1 }

Now, with this function defined, you can write

X = $(file_contents filename) # equivalent to builtin $(&cat filename)

and the variable "$(X)" will fetch the contents of the given file every time it gets
expanded. Use ":=" to do this exactly once, or ";=" to do this at most once.

See makepp_extending for more details and examples.

Simple Statements
autoload filename ...
Specifies one or more makefiles to load should an attempt to find a rule for a file in
this directory otherwise fail. This is useful when the makefile has rules whose
definitions depend (possibly indirectly) on a file in another directory that depends
(possibly indirectly) on other files in this directory (built by rules that do not
depend on the file in the other directory).

For example, your Makeppfile might look like this:

rules-to-build-files-that-otherdir/x-depends-on

more_rules.makeppfile: otherdir/x
action-to-build-more_rules.makeppfile

autoload more_rules.makeppfile

Note that we cannot reliably replace "autoload" with "include" here, because if
something other than the rule for more_rules.makeppfile tries to build otherdir/x
first, then more_rules.makeppfile will probably fail because otherdir/x won't exist
yet, because there is already an attempt to build it underway when Makeppfile is
implicitly loaded on its behalf.

WARNING: Be very careful about doing things in an autoloaded makefile that change the
behavior of rules in the directory's other makefile(s), as this will cause that
behavior to depend on whether or not some previously built target caused makefiles to
be autoloaded.

build_cache /path/to/build/cache
[global] build_cache /path/to/build/cache
Specifies a path to a build cache. See makepp_build_cache for details. The build
cache must already exist; see "How to manage a build cache" in makepp_build_cache for
how to make it in the first place. A "build_cache" statement in a makefile overrides
the "--build-cache" command line option for rules in the makefile, but it may be
overridden by the ":build_cache" rule modifier on a per-rule basis.

The keyword "global" may precede this statement with the same effect as the command
line option, i.e. the build cache applies in every makefile. This should best be
given in a RootMakeppfile to be certain it is seen early enough.

Specify "none" instead of a path to a directory if you want to disable the build cache
for all rules in this makefile.

build_check build_check_method
[global] build_check build_check_method
Specifies the default build check method for all rules in this makefile. See
makepp_build_check for details. The "build_check" statement overrides the
"--build-check-method" command line option for all rules in the makefile, but may be
overridden by the ":build_check" modifier on a per-rule basis.

The keyword "global" may precede this statement with the same effect as the command
line option, i.e. the build check method applies in every makefile which does not
specify its own. This should best be given in a RootMakeppfile to be certain it is
seen early enough.

Specify "build_check default" instead of a name if you want to return to the default.
With the keyword "global" this means the "exact_match" method, else this reverts the
current makefile to not having its own specific method.

export VAR ...
export assignment
export PATH := $(PWD):$(PATH)

Marks the given variables for export to subprocesses. See setting variables.

global VAR ...
global assignment
global MYPROJECT.INFO = info to be seen in all makefiles

Marks the given variables as global to all makefiles. See setting variables.

include makefile
This inserts the contents of another makefile into the current makefile. It can be
useful if you have boilerplate files with a number of rules or variables, and each
directory only needs to make a few modifications. The "include" statement also used
to be commonly used in traditional makes in conjunction with automatic include file
scanners, but this is no longer necessary with makepp.

"include" first considers the current directory, then the parent of the current
directory, then its parent, etc. It stops considering directories when it reaches the
root of the file system or when the file system device ID changes. (This means that
it will not find files located in other NFS mounts. This is to prevent problems with
network file systems or automounters and dead servers.) If it does not find a file of
the given name by the time its search is stopped, then it looks in the makepp data
directory (/usr/local/share/makepp if you installed makepp in /usr/local) for one of
the include files that comes with makepp.

If you want to include a template file in every makefile in a whole directory
hierarchy, you can place your makefile template at the top directory. The makefiles
do not have to know exactly where they are in the hierarchy; each makefile can contain
a line like this:

include standard_definitions.mk

instead of something more complicated, like this:

include ../../../standard_definitions.mk # Is this the right number of ..?

You can specify as many files as you want, and variables are allowed:

include file1 file2 file3 $(other_include_files)

If you're working on a build that needs to work with both GNU make and makepp,
sometimes it's convenient to have exactly identical makefiles but a different include
file. For example, all of your makefiles may contain a line like this:

include $(TOPDIR)/standard_rules.mk

and you want standard_rules.mk to be different for GNU make and makepp. To facilitate
this, the "include" statement first looks for a file with the suffix of .makepp before
looking for the file you asked for. In this case, it would first look for a file
called standard_rules.mk.makepp, and if that exists, it would load it instead of
standard_rules.mk. This way, when you run the makefile with GNU make, it loads
standard_rules.mk, but with makepp, it loads standard_rules.mk.makepp.

Because many legacy files put the rule to generate an include file after the include
statement, makepp will defer decisions about inexistant or stale includes till the end
of makefile loading. That is, unless it is invoked with "--rm-stale". For as many
times as the situation has improved by then (because a rule appeared) makepp will
reload the makefile, which may again make more such rules appear. This is obviously
an inefficient way to load makefiles, so try to avoid that. Worse, if your makefile
loading has side-effects (like appending to a global variable or a line like
"do_it_now := $(shell cat a >>b)" or its makepp equivalent "&cat a -o>>b") they will
happen as many times as the makefile needs to be loaded, so, again, try to avoid that!

_include makefile
-include makefile
A minor variant on "include", the "_include" statement includes the file if it exists
but doesn't generate a fatal error if it does not. The "_include" statement used to
be important for include file scanning with GNU make, but is less useful for makepp.

load_makefile /some/directory/somewhere/Makefile
load_makefile subdir
load_makefile VAR1=value1 VAR2=value2 subdir
This statement causes makepp to cd to the directory containing the makefile and load
its rules into makepp's internal database. If you specify just a directory instead of
a makefile, "load_makefile" looks for "Makeppfile", "makefile", or "Makefile" in that
directory.

Any variables you specify with the syntax "VAR=value" (or "VAR="value1 value2"") are
passed to the loaded makefiles. They override any settings in those makefiles, just
as if you had typed them on the command line.

Using "load_makefile" is different from the command

include dir/makefile

in two ways. First, "load_makefile" does not transfer any variables from the top-
level makefile into the subordinate makefile; each makefile exists in its own
namespace. The subordinate makefile cannot influence the variables in the top-level
makefile in any way.

Second, each build command is tagged with the directory of the makefile that it came
from. When makepp executes a rule from a different makefile, it first cd's to the
directory containing that makefile before executing the command. Makefiles which are
seen with the "include" statement are actually treated as part of the makefile that
included them, and therefore their rules are not tagged with a different directory.

You usually do not have to load a makefile explicitly, unless it has an unusual name,
or it has targets which are not contained in the same directory as the makefile
itself, or you have disabled implicit makefile loading. By default, if makepp is
trying to build a file and doesn't have a rule to build it, or if it is evaluating a
wildcarded filename in a directory, it will automatically attempt to load a makefile
from that directory. See "Tips for multiple directories" in makepp_cookbook for info
on building with multiple directories.

You cannot use "load_makefile" to load several makefiles that apply to the same
directory. Use "include" for several pieces of the makefile that apply to the same
directory, and "load_makefile" for makefiles that apply to different directories.

no_implicit_load
This statement turns off implicit loading of makefiles from a set of directories.
This can be useful if you want to load makefiles automatically from most directories,
but there are some directories which for various reasons you do not want makepp to
attempt to update. (E.g., maybe the directory has a makefile for some other version
of make which makepp does not understand.) For example,

no_implicit_load dir1 dir2/*

The above statement will turn off implicit loading for makefiles in "dir1" and all of
its subdirectories. It will also turn of implicit makefile loading for all
subdirectories of "dir2" (and all of their subdirectories), but not for "dir2" itself.

You may use wildcards in the statement. Non-directory files that match the wildcard
are ignored. You can also use functions to further specify the directories that you
are interested in, e.g.,

no_implicit_load $(filter-out dir1 dir2, *)

will turn off implicit loading for all subdirectories except dir1 and dir2 and their
subdirectories.

prebuild target
make target
The arguments (which undergo Make-style variable expansion) are built immediately.
This is useful when the list of targets that the Makefile can build depends on a
generated file in another directory.

Currently, it will quietly fail to build targets if there is a dependency loop among
the prebuilt targets and the Makefiles that must be loaded to build them, but that
ought to be treated as an error.

register_command_parser command_word parser
register_parser command_word parser
When lexically analyzing rule actions, use parser for command_word, which may be the
full path or just the basename. The basename is usually enough because the lexer
tries both.

The parser may either be a classname with or without the leading
"Mpp::CommandParser::". Such a class must have a member function called "factory"
that returns an object of that class. If the classname contains colons, it must be
quoted, so as not make this line look like a rule.

Or, because that class is usually not yet loaded, instead the factory function may
reside in the Makefile namespace. These functions have a prefix of "p_" which must
not be given. This is the case of the builtin parsers.

The effect is comparable to the ":parser" rule option. But for multi-command rules
this is the better way.

register_input_suffix command_word suffix ...
Add "suffix" ... to the list of input file suffixes recognized when an action
beginning with "command_word" is parsed. The parser would normally pick this up via
Mpp::CommandParser::input_filename_regexp, but it might instead ignore this entirely.

Parsers don't normally pick up all the arguments that aren't recognized as options,
because they might be arguments of unrecognized options. (For example, i386v is not
an input file of the command "gcc -b i386v foo.c".) Instead, they pick up only
positional arguments that look like input filenames.

It is not unusual to use standard tools with site-specific nonstandard suffixes in
order to signify that those files require special handling, such as different command
options and/or postprocessing steps. For example:

register_input_suffix cpp .vpp
%.v: %.vpp
cpp $< > $@

repository directory
repository destdir=srcdir
Specifies one or more repository directories. The first repository specified has
precedence over the others if the same file exists in multiple repositories and there
is no build command for it. See makepp_repositories for more details about
repositories.

If you specify just a directory after "repository", its contents are linked into the
current directory. You can link its contents into any arbitrary place in the file
system by specifying the location before an equals sign, e.g,

repository subdir1/subdir2=/users/joe/joes_nifty_library

You should put the repository statement near the top of your makefile, before any
rules that may need to use it.

runtime program,library
Store "library" as a runtime dependency of "program". Both "program" and "library"
may contain multiple words, in which case each word in "library" is stored as a
runtime dependency of each word in "program". When "program" is added automatically
as the executable dependency of a command by the "Mpp::CommandParser" base class, its
runtime dependencies (if any) are added as well. In order for this to happen,
"program" must be specified in the rule with a directory component, and without any
shell meta characters. The purpose of this statement is to capture dependencies on
libraries and other executables that are often loaded by the program, without having
to specify them explicitly as dependencies of each rule that invokes "program", or to
scan "program" to determine those dependencies (which could be prohibitively
difficult.)

Runtime dependencies are traversed recursively, so if "a" has a runtime dependency on
"b" and "b" has a runtime dependency on "c", then any rule that uses "./a" will have
implicit dependencies on both "b" and "c" (unless it uses a special
"Mpp::CommandParser" class that overrides this behavior).

Note that missing dependencies won't necessarily be added after you add this statement
to a makefile, unless the rule is re-scanned. Use the "--force-rescan" command line
option to ensure that this happens.

signature name
[global] [override] signature name
signature md5
signature C
signature c_compilation_md5
signature xml
signature xml-space
signature default

Sets the signature method for all rules following the "signature" statement, for which
no command parser chooses a method. You can override this for individual rules with
the ":signature" rule modifier.

If you add the keyword "override", then this method will override even the the choice
made by command parsers, but not those specified with the ":signature" rule modifier.
If you add the keyword "global", the effect applies to all rules yet to be read,
unless their makefile also has its own "signature" statement. This is equivalent to
the "--signature" command line option if given before any rule is read, e.g. in a
RootMakeppfile to be certain it is seen early enough. Likewise the keywords "global
override" for this statement are equivalent to the "--override-signature" command line
option.

Specify "signature default" instead of a name if you want to return to the default.
With the keyword "global" this means the simple modification time and file size
method. Else this reverts the current makefile to not having its own specific method,
using a global method if one was set.

For more information about signature methods, see makepp_signatures.

vpath pattern directory ...
Fetch all files matching pattern from each given directory. Pattern may contain at
most one "%" wildcard. This uses the transparent repository mechanism (unlike gmake
which rewrites filenames), but it does not recurse into subdirectories.

Commands
All builtin and self defined commands (see builtin commands and extending makepp), as well
as external cleanly programmed Perl scripts can be used like statements. In this case
they differ from rule actions in that they run in the same process as makepp and any input
or output files are not noted as dependencies or as having been built by makepp.

As with all statements, they are considered as such, if they are indented less than the
actions of the previous rule, if any.

This can be used for messages to be output while reading the makefile:

&echo The value of $$(VAR) is $(VAR)

Or instead of making many rules each depend on a directory creation rule, you can simply
create it on the fly. Note that commands which create files are processed again every
time the makefile is read., That's why we protect this one with a test -- though in this
special case that would not be necessary, as this command would do no harm when repeated:

ifperl !-d 'include'
&mkdir -p include # Create only if not present
endif

AUTHOR

       Gary Holt (holt-makepp@gholt.net)