NAME¶
makepp_extending -- How to extend makepp using Perl
DESCRIPTION¶
Makepp internally is flexible enough so that by writing a little bit of Perl
code, you can add functions or do a number of other operations.
General notes on writing Perl code to work with makepp¶
Each makefile lives in its own package. Thus definitions in one makefile do not
affect definitions in another makefile. A common set of functions including
all the standard textual manipulation functions is imported into the package
when it is created.
Makefile variables are stored as Perl scalars in that package. (There are
exceptions to this: automatic variables and the default value of variables
like CC are actually implemented as functions with no arguments. Target
specific vars, command line vars and environment vars are not seen this way.)
Thus any Perl code you write has access to all makefile variables. Global
variables are stored in the "Mpp::global" package. See Makefile
variables for the details.
Each of the statements (ifperl / ifmakeperl, perl / makeperl, sub / makesub),
the functions (perl / makeperl, map / makemap) and the rule action (perl /
makeperl) for writing Perl code directly in the makefile come in two flavours.
The first is absolutely normal Perl, meaning you have to use the
"f_" prefix as explained in the next section, if you want to call
makepp functions. The second variant first passes the statement through
Make-style variable expansion, meaning you have to double the "$"s
you want Perl to see.
End handling is special because makepp's huge (depending on your build system)
data structures would take several seconds to garbage collect with a normal
exit. So we do a brute force exit. In the main process you can still have
"END" blocks but if you have any global file handles they may not
get flushed. But you should be using the modern lexical filehandles, which get
closed properly when going out of scope.
In Perl code run directly as a rule action or via a command you define, it is
the opposite. "END" blocks will not be run, but global filehandles
get flushed for you. The "DESTROY" of global objects will never be
run.
Adding new textual functions¶
You can add a new function to makepp's repertoire by simply defining a Perl
subroutine of the same name but with a prefix of "f_". For example:
sub f_myfunc {
my $argument = &arg; # Name the argument.
my( undef, $mkfile, $mkfile_line ) = @_; # Name the arguments.
... do something here
return $return_value;
}
XYZ := $(myfunc my func arguments)
If your function takes no arguments, there is nothing to do. If your function
takes one argument, as in the example above, use the simple accessor &arg
to obtain it. If you expect more arguments, you need the more complex accessor
"args" described below.
These accessors processes the same three parameters that should be passed to any
"f_" function, namely the function arguments, the makefile object
and a line descriptor for messages. Therefore you can use the efficient
&arg form in the first case.
The &arg accessor takes care of the following for you: If the arguments were
already expanded (e.g. to find the name of the function in
"$(my$(function) arg)" the arg is passed as a string and just
returned. If the argument still needs expansion, this is the usual case, it is
instead a reference to a string. The &arg accessor expands it for you, for
which it needs the makefile object as its 2nd parameter.
If you expect more arguments, possibly in variable number, the job is performed
by "args". This accessor takes the same 3 parameters as arg, plus
additional parameters:
- max: number of args (default 2): give ~0 (maxint) for endless
- min: number of args (default 0 if max is ~0, else same as max)
- only_comma: don't eat space around commas, usual for non-filename
At most max, but at least min commas present before expansion are used to split
the arguments. Some examples from makepp's builtin functions:
my( $prefix, $text ) = args $_[0], $_[1], $_[2], 2, 2, 1; # addprefix
for my $cond ( args $_[0], undef, $_[2], ~0 ) ... # and, or
my @args= args $_[0], $_[1], $_[2], ~0, 1, 1; # call
my( $filters, $words ) = args $_[0], $_[1], $_[2]; # filter
The function should return a scalar string (not an array) which is then inserted
into the text at that point.
If your function encounters an error, it should die using the usual Perl die
statement. This will be trapped by makepp and an error message displaying the
file name and the line number of the expression causing the error will be
printed out.
There are essentially no limits on what the function can do; you can access the
file, run shell commands, etc.
At present, expressions appearing in dependencies and in the rule actions are
expanded once while expressions appearing in targets are expanded twice, so be
careful if your function has side effects and is present in an expression for
a target.
Note that the environment (in particular, the cwd) in which the function
evaluates will not necessarily match the environment in which the rules from
the Makefile in which the function was evaluated are executed. If this is a
problem for you, then your function probably ought to look something like
this:
sub f_foo {
...
chdir $makefile->{CWD};
... etc.
}
Putting functions into a Perl module¶
If you put functions into an include file, you will have one copy per Makeppfile
which uses it. To avoid that, you can write them as a normal Perl module with
an "Exporter" interface, and use that. This will load faster and
save memory:
perl { use mymodule }
perl {
use my::module; # put : on a new line so this is not parsed as a rule
}
If you need any of the functions normally available in a Makefile (like the
"f_" functions, "arg" or "args"), you must put
this line into your module:
use Mpp::Subs;
The drawback is that the module would be in a different package than a function
directly appearing in a makefile. So you need to pass in everything as
parameters, or construct names with Perl's "caller" function.
Calling external Perl scripts¶
If you call an external Perl script via "system", or as a rule action,
makepp will fork a new process (unless it's the last rule action) and fire off
a brand new perl interpreter. There's nothing wrong with that, except that
there's a more efficient way:
- &command arguments...
- This can be a rule action. It will call a function command with a
"c_" prefix, and pass it the remaining (optionally quoted makepp
style -- not exactly the same as Shell) arguments. If such a function
cannot be found, this passes all strings to "run".
sub c_mycmd { my @args = @_; ... }
$(phony callcmd):
&mycmd 'arg with space' arg2 "arg3" # calls c_mycmd
%.out: %.in
&myscript -o $(output) $(input) # calls external myscript
You can write your commands within the framework of the builtins, allowing
you to use the same standard options as they have, and the I/O handling
they give.
The block operator "Mpp::Cmds::frame" is followed by a single
letter option list of the builtins (maximally "qw(f i I o O r
s)"). Even if you specify your own option overriding one of these,
you still give the single letter of the standard option.
Each own option is specified as "[qw(n name), \$ref, arg,
sub]". The first two elements are short and long name,
followed by the variable reference and optionally by a boolean for whether
to take an argument. Without an arg, the variable is incremented each time
the option is given, else the option value is stored in it.
sub c_my_ocmd { # Typical output case
local @ARGV = @_;
Mpp::Cmds::frame {
... print something here with @ARGV, with options already automatically removed
} 'f', qw(o O);
}
sub c_my_icmd { # Typical input case with 2 options
local @ARGV = @_;
my( $short, $long );
Mpp::Cmds::frame {
... do something here with <>
} qw(i I r s), # s specifies only --separator, not -s
[qw(s short), \$short], # No option arg -> $short == 1
[qw(l long), \$long, 1, sub { warn "got arg $long"}];
}
Here comes a simple command which upcases only the first character of each
input record (equivalent to "&sed '$$_ =
"\u\L$$_"'"):
sub c_uc {
local @ARGV = @_;
Mpp::Cmds::frame {
print "\u\L$_" while <>;
} 'f', qw(i I o O r s);
}
Within the block handled by frame, you can have nested blocks for performing
critical operations, like opening other files.
Mpp::Cmds::perform { ... } 'message';
This will output message with "--verbose" (which every command
accepts) iff the command is successfully run. But if the block evaluates
as false, it dies with negated message.
- run script arguments...
- This is a normal Perl function you can use in any Perl context within your
makefile. It is similar to the multi-argument form of system, but it runs
the Perl script within the current process. For makepp statements, the
perl function or your own functions that is the process running makepp.
But for a rule that is the subprocess performing it. The script gets
parsed as many times as it gets called, but you can put the real work into
a lib, as pod2html does. This lib can then get used in the top level, so
that it's already present:
perl { use mylib } # gets forked to all rules which needn't reparse it
%.out: %.in
makeperl { run qw'myscript -o $(output) $(input)' }
If the script calls "exit", closes standard file descriptors or
relies on the system to clean up after it (open files, memory...), this
can be a problem with "run". If you call "run" within
statements or the perl function, makepp can get disturbed or the cleanup
only happens at the end of makepp.
If you have one the aforementioned problems, run the script externally, i.e.
as from the command line instead. Within a rule cleanup is less of a
problem, especially not as the last action of a rule, since the rule
subprocess will exit afterwards anyway, except on Windows.
Writing your own signature methods¶
Sometimes you want makepp to compute a signature method using a different
technique. For example, suppose you have a binary that depends on a shared
library. Ordinarily, if you change the shared library, you don't have to
relink executables that depend on it because the linking is done at run time.
(However, it is possible that relinking the executable might be necessary,
which is why I did not make this the default.) What you want makepp to do is
to have the same signature for the shared library even if it changes.
This can be accomplished in several ways. The easiest way is to create your own
new signature method (let's call it "shared_object"). You would use
this signature method only on rules that link binaries, like this:
myprogram : *.o lib1/lib1.so lib2/lib2.so
: signature shared_object
$(CC) $(inputs) -o $(output)
Now we have to create the signature method.
All signature methods must be their own class, and the class must contain a few
special items (see Mpp/Signature.pm in the distribution for details). The
class's name must be prefixed with "Mpp::Signature::", so in this
case our class should be called "Mpp::Signature::shared_object". We
have to create a file called
shared_object.pm and put it into a
Mpp::Signature directory somewhere in the Perl include path; the
easiest place might be in the
Mpp/Signature directory in the makepp
installation (e.g.,
/usr/local/share/makepp/Mpp/Signature or wherever
you installed it).
For precise details about what has to go in this class, you should look
carefully through the file
Mpp/Signature.pm and probably also
Mpp/Signature/exact_match.pm in the makepp distribution. But in our
case, all we want to do is to make a very small change to an existing
signature mechanism; if the file is a shared library, we want to have a
constant signature, whereas if the file is anything else, we want to rely on
makepp's normal signature mechanism. The best way to do this is to inherit
from "Mpp::Signature::c_compilation_md5", which is the signature
method that is usually chosen when makepp recognizes a link command.
So the file
Mpp/Signature/shared_object.pm might contain the following:
use strict;
package Mpp::Signature::shared_object;
use Mpp::Signature::c_compilation_md5;
our @ISA = qw(Mpp::Signature::c_compilation_md5); # Indicate inheritance.
our $shared_object = bless \@ISA; # A piece of magic that helps makepp find
# the subroutines for this method. All
# signature methods must have one of these.
# The value is not used, just any object.
# Now here's the method that gets called when we need the signature of
# any target or dependency for which this signature method is active:
sub signature {
my ($self, # This will be the same as $shared_object.
$finfo) = @_; # A special structure that contains everything
# makepp knows about this file. See
# Mpp/File.pm for details.
if ($finfo->{NAME} =~ /\.s[oa]$/) { # Does the file name end in .so or .sa?
return $finfo->file_exists ? 'exists' : '';
# Always return the same signature if the file
# exists. In this case, the signature is the
# string "exists".
}
Mpp::Signature::c_compilation_md5::signature;
# If the file didn't end in .so or .sa,
# delegate to makepp's usual signature method.
}
This file is provided as an example in the makepp distribution, with some
additional comments.
Incidentally, why don't we make this the default? Well, there are times when
changing a shared library will require a relinking of your program. If you
ever change either the symbols that a shared library defines, or the symbols
that it depends on other libraries for, a relink may sometimes be necessary.
Suppose, for example, that the shared library invokes some subroutines that your
program provides. E.g., suppose you change the shared library so it now calls
an external subroutine "xyz()". Unless you use the "-E" or
"--export-dynamic" option to the linker (for GNU binutils; other
linkers have different option names), the symbol "xyz()" may not be
accessible to the run-time linker even if it exists in your program.
Even worse, suppose you defined "xyz()" in another library (call it
libxyz), like this:
my_program: main.o lib1/lib1.so xyz/libxyz.a
Since "libxyz" is a
.a file and not a
.so file, then
"xyz()" may not be pulled in correctly from
libxyz.a unless
you relink your binary.
Mpp::Signature methods also control not only the string that is used to
determine if a file has changed, but the algorithm that is used to compare the
strings. For example, the signature method "target_newer" in the
makepp distribution merely requires that the targets be newer than the
dependencies, whereas the signature method "exact_match" (and
everything that depends on it, such as "md5" and
"c_compilation_md5") requires that the file have the same signature
as on the last build.
Here are some other kinds of signature methods that might be useful, to help you
realize the possibilities. If general purpose enough, some of these may
eventually be incorporated into makepp:
- •
- A signature method for shared libraries that returns a checksum of all the
exported symbols, and also all the symbols that it needs from other
libraries. This solves the problem with the example above, and guarantees
a correct link under all circumstances. An experimental attempt has been
made to do this in the makepp distribution (see
Mpp/Signature/shared_object.pm), but it will only work with GNU
binutils and ELF libraries at the moment.
- •
- A signature method that ignores a date stamp written into a file. E.g., if
you generate a .c file automatically using some program that
insists on putting a string in like this:
static char * date_stamp = "Generated automatically on 01 Apr 2004 by nobody";
you could write a signature method that specifically ignores changes in date
stamps. Thus if the date stamp is the only thing that has changed, makepp
will not rebuild.
- •
- A signature method that computes the signatures the normal way, but
ignores the architecture dependence when deciding whether to rebuild. This
could be useful for truly architecture-independent files; currently if you
build on one architecture, makepp will insist on rebuilding even
architecture-independent files when you switch to a different
architecture.
- •
- A signature method that knows how to ignore comments in latex files, as
the "c_compilation_md5" method knows how to ignore comments in C
files.
- •
- A signature method for automatic documentation extraction that checksums
only to the comments that a documentation extractor needs and ignores
other changes to the source file.
Unfinished¶
This document is not finished yet. It should cover how to write your own
scanners for include files and things like that.