Data::Rlist - A lightweight data language for Perl, Python and C++
use Data::Rlist;
File and string I/O for any Perl data $thing:
### Compile data as text.
WriteData $thing, $filename; # compile data into file
WriteData $thing, \$string; # compile data into buffer
$string_ref = WriteData $thing; # dto.
$string = OutlineData $thing; # compile printable text
$string = StringizeData $thing; # compile text in a compact form (no newlines)
$string = SqueezeData $thing; # compile text in a super-compact form (no whitespace)
### Parse data from text.
$thing = ReadData $filename; # parse data from file
$thing = ReadData \$string; # parse data from string buffer
ReadData, WriteData etc. are auto-exported functions. Alternately we use:
### Qualified functions to parse text.
$thing = Data::Rlist::read($filename);
$thing = Data::Rlist::read($string_ref);
$thing = Data::Rlist::read_string($string_or_string_ref);
### Qualified functions to compile data into text.
Data::Rlist::write($thing, $filename);
$string_ref = Data::Rlist::write_string($thing);
$string = Data::Rlist::write_string_value($thing);
### Print data to STDOUT.
PrintData $thing;
The object-oriented interface:
### For objects the '-output' attribute refers to a string buffer or is a filename.
### The '-data' attribute defines the value or reference to be compiled into text.
$object = new Data::Rlist(-data => $thing, -output => \$target)
$string_ref = $object->write; # compile into $target, return \$target
$string_ref = $object->write_string; # compile into new string ($target not touched)
$string = $object->write_string_value; # dto. but return string value
### Print data to STDOUT.
print $object->write_string_value;
print ${$object->write}; # returns \$target
### Set output file and write $thing to disk.
$object->set(-output => ".foorc");
$object->write; # write "./.foorc", return 1
$object->write(".barrc"); # write "./.barrc" (the filename overrides -output)
### The '-input' attribute defines the text to be compiled, either as
### string reference or filename.
$object->set(-input => \$input_string); # assign some text
$thing = $object->read; # parse $input_string into Perl data
$thing = $object->read($other_string); # parse $other_string (the argument overrides -input)
$object->set(-input => ".foorc"); # assign some input file
$foorc = $object->read; # parse ".foorc"
$barrc = $object->read(".barrc"); # parse some other file
$thing = $object->read(\$string); # parse some string buffer
$thing = $object->read_string($string_or_ref); # dto.
Create deep-copies of any Perl data. The metaphor ``keelhaul'' vividly connotes that $thing is stringified, then compiled back:
### Compile a value or ref $thing into text, then parse back into data.
$reloaded = KeelhaulData $thing;
$reloaded = Data::Rlist::keelhaul($thing);
$object = new Data::Rlist(-data => $thing);
$reloaded = $object->keelhaul;
Do deep-comparisons of any Perl data:
### Deep-compare $a and $b and get a description of all type/value differences.
@diffs = CompareData($a, $b);
For more information see compile, keelhaul, and deep_compare.
Random-Lists (Rlist) is a tag/value text format, which can ``stringify'' any data structure in 7-bit ASCII text. Its basic types are lists and scalars. For example,
( "hello", "world" )
{ "hello" = "world"; }
designates two lists of scalars, the first of which is sequential, the second associative. The syntax is similar, but not equal to Perl's. The format...
- allows the definition of hierachical and constant data,
- knows no user-defined types, no keywords, no variables,
- knows no arithmetic expressions,
- uses 7-bit-ASCII character encoding and escape sequences,
- uses C-style numbers and strings,
- has an extremely minimal syntax implementable in any programming language and system.
You can write any Perl data structure into files as legible text. Like with CSV the lexical overhead of Rlist is minimal: files are merely data.
You can load the text in Perl, C++ or Python programs. No information will be lost between different program languages, and floating-point numbers keep their precision. The C++ parser returns double, std::string, std::vector and std::map objects.
You can also compile CSV and XML text from Perl data (using functions of this package).
Since we have no user-defined types the data is structured out of simple scalars and lists. It is conceivable, however, to develop a simple type system and store type information along with the actual data. Otherwise the data structures are tacit consents between the users of the data.
The Rlist implemenations for Perl and C++ are fast and scalable so (e.g.) a file can have hundreds of megabytes of data, and is processable in constant time, with constant memory requirements. See also C++.
Rlist text uses the 7-bit-ASCII character set. The 95 printable character codes 32 to 126 occupy one character. Codes 0 to 31 and 127 to 255 require four characters each: the \ escape character followed by the octal code number. For example, the German Umlaut character ü (252) is translated into \374. An exception are the following codes:
ASCII ESCAPED AS
----- ----------
9 tab \t
10 linefeed \n
13 return \r
34 quote " \"
39 quote ' \'
92 backslash \ \\
Values are either scalars, array elements or the value of a pair.
Each value is constant.
By definition the default scalar value is the empty string "". So in Perl undef is compiled
into "".
Strings are characters placed in double-quotes (single-quotes are not allowed). An exception are strings consisting only of [a-zA-Z_0-9-/~:.@] characters; such strings ``look like'' identifiers (aka symbols) and for them the quotes are optional.
When a string is quoted it is also escaped, which means its characters are converted to the input character set of 7-bit ASCII.
A string can also be defined using a line-oriented form of quoting based on the UNIX shell here-document syntax and RFC 111. Multi-line quoted strings can be expressed with
<<DELIMITER
Following the sigil << an identifier specifies how to terminate the string scalar. The value of the scalar will be all lines following the current line down to the line starting with the delimiter. There must be no space between the << and the identifier.
Numbers adhere to the IEEE 754 syntax for integer- and floating-point numbers (i.e., the same lexical conventions as in C and C++ apply). String constants follow the C language lexicography.
EXAMPLES
Quoted strings:
"Hello, World!"
Unquoted strings (symbols, identifiers):
foobar cogito.ergo.sum Memento::mori
Here-document strings:
<<hamlet
"This above all: to thine own self be true". - (Act I, Scene III).
hamlet
Integegers and floats:
38 10e-6 -.7 3.141592653589793
For more information see is_symbol, is_number and escape.
We have two types of lists: sequential (aka array) and associative (aka map, hash, dictionary).
EXAMPLES
Arrays:
( 1, 2, ( 3, "Audiatur et altera pars!" ) )
Maps:
{
key = value;
standalone-key;
Pi = 3.14159;
"meta-syntactic names" = (foo, bar, "lorem ipsum", Acme, ___);
var = {
log = {
messages = <<LOG;
Nov 27 21:55:04 localhost kernel: TSC appears to be running slowly. Marking it as unstable
Nov 27 22:34:27 localhost kernel: Uniform CD-ROM driver Revision: 3.20
Nov 27 22:34:27 localhost kernel: Loading iSCSI transport class v2.0-724.<6>PNP: No PS/2 controller found. Probing ports directly.
Nov 27 22:34:27 localhost kernel: wifi0: Atheros 5212: mem=0x26000000, irq=11
LOG
};
};
}
Binary data can be represented as base64-encoded string, or here-document string. For example,
use MIME::Base64;
$str = encode_base64($binary_buf);
The returned string $str is broken into lines of no more than 76 characters each and it will end
with "\n" unless it is empty. When $str written to a file, the file contents look like:
{
random_string = <<___
w5BFJIB3UxX/NVQkpKkCxEulDJ0ZR3ku1dBw9iPu2UVNIr71Y0qsL4WxvR/rN8VgswNDygI0xelb
aK3FytOrFg6c1EgaOtEudmUdCfGamjsRNHE2s5RiY0ZiaC5E5XCm9H087dAjUHPtOiZEpZVt3wAc
KfoV97kETH3BU8/bFGOqscCIVLUwD9NIIBWtAw6m4evm42kNhDdQKA3dNXvhbI260pUzwXiLYg8q
MDO8rSdcpL4Lm+tYikKrgCih9UxpWbfus+yHWIoKo/6tW4KFoufGFf3zcgnurYSSG2KRLKkmyEa+
s19vvUNmjOH0j1Ph0ZTi2pFucIhok4krJi0B5yNbQStQaq23v7sTqNom/xdRgAITROUIoel5sQIn
CqxenNM/M4uiUBV9OhyP
___
;
}
Each line except the last in the here-doc-string has 75 characters, plus the newline. To compile here-document strings is enabled by default for multiline strings. The above file, let's call it random.rls, can be created with the following code:
use Data::Rlist;
use MIME::Base64;
$binary_data = join('', map { chr(int rand 256) } 1..300);
$sample = { random_string => encode_base64($binary_data) };
WriteData $sample, 'random.rls';
See also Encode, the MIME::Base64 manpage.
Rlist text can define embedded Perl programs, called nanonscripts. The embedded program text
has the form of a here-document with the special delimiter "perl". After the Rlist text has
been parsed you call evaluate_nanoscripts to run the embedded codes in the order of
definiton. The function arranges it that within the eval...
the $root variable refers to the root of the input (as unblessed array- or hash-reference);
the $this variable refers to the array or hash that stores the eval'd nanoscript;
the $where variable stores the name of the key or the index within $this.
The nanoscript can use this information to oriented itself within the parsed data, or even to modify the data in-place. The result the code returns is put in place of the nanoscript text. Alternatively you can also eval the embedded Perl codes on your programmatically; see the nanoscripts and result functions.
EXAMPLES
Single nanoscript:
hello = <<perl;
print "Hello, World!";
perl
List of nanoscripts:
( <<perl, <<perl, <<perl, <<perl )
print "Hello World!\n" # english
perl
print "Hallo Welt!\n" # german
perl
print "Bonjour le monde!\n" # french
perl
print "Olá mundo!\n" # spanish
perl
Modify the parsed data in place:
( <<perl )
$rlist = ReadData(\'{ foo = bar; }');
perl
The original data as returned by read was:
[ "\$rlist = ReadData('{ foo = bar; }');\n" ]
By calling evaluate_nanoscripts this data morphs into:
{ 'foo' => 'bar' }
Rlist supports multiple forms of comments: // or # single-line-comments, and /* */ multi-line-comments. You may use all three forms at will.
The core functions to cultivate package objects are new, dock, set and get. When a regular package function is called in object context some or all arguments can be omitted. They will be reused from object attributes. This is true for the following functions: read, write, read_string, write_string, read_csv, write_csv, read_conf, write_conf and keelhaul.
When not called in object context, however, the first argument has an indifferent meaning. read expects an input file or string, write the data to compile etc.
Create a Data::Rlist object from the hash ATTRIBUTES. For example,
$self = Data::Rlist->new(-input => 'this.dat',
-data => $thing,
-output => 'that.dat');
For this object the call $self->read() reads from this.dat, and $self->write() writes any Perl data $thing to that.dat.
REGULAR OBJECT ATTRIBUTES
-input => INPUT
-filter => FILTER
-filter_args => FILTER-ARGS
Defines what Rlist text to parse and how to preprocess an input file. INPUT is a filename or string reference. FILTER can be 1 to select the standard C preprocessor cpp.
These attributes are applied by read, read_string, read_conf and read_csv.
-data => DATA
-options => OPTIONS
-output => OUTPUT
Defines the Perl data to be compiled into text (DATA), how it shall be compiled (OPTIONS) and where to store the compiled text (OUTPUT). When OUTPUT is string reference the compiled text will be stored in that string. When OUTPUT is undef a new string is created. When OUTPUT is a string value it is a filename.
These attributes are applied by write, write_string, write_conf, write_csv and keelhaul.
-header => HEADER
Defines an array of text lines, each of which will by prefixed by a # and then written at the top of the output file.
-delimiter => DELIMITER
Defines the field delimiter for .csv-files. Applied by read_csv and read_conf.
-columns => STRINGS
Defines the column names for .csv-files to be written into the first line.
ATTRIBUTES THAT MASQUERADE PACKAGE GLOBALS
The attributes listed below raise new values for package globals for the time an object method runs.
-InputRecordSeparator => FLAG
Masquerades $/, which affects how lines are read and written to and from Rlist- and CSV-files. You may also set $/ by yourself. See perlport and perlvar.
-MaxDepth => INTEGER
-SafeCppMode => FLAG
-RoundScientific => FLAG
Masquerade $Data::Rlist::MaxDepth, $Data::Rlist::SafeCppMode and $Data::Rlist::RoundScientific.
When the "precision" compile option is defined, Data::Rlist::round is called during
compilation to round all numbers to a certain count of decimal places. When the
$Data::Rlist::RoundScientific flag is true, round formats the number in either normal or
exponential (scientific) notation, whichever is more appropriate for its magnitude. By setting
-RoundScientific this sort of formatting can be enabled per object.
-EchoStderr => FLAG
Print read errors and warnings message on STDERR (default: off).
-DefaultCsvDelimiter => REGEX
-DefaultConfDelimiter => REGEX
Masquerades $Data::Rlist::DefaultCsvDelimiter and $Data::Rlist::DefaultConfDelimiter. These
globals define the default regexes to use when the -options attribute does not specifiy the
"delimiter" regex. Applied by read_csv and read_conf.
-DefaultConfSeparator => STRING
Masquerades $Data::Rlist::DefaultConfSeparator, the default string to use when the -options
attribute does not specifiy the "separator" string. Applied by
write_conf.
Exclusively links some object SELF to the package. This means that some of SELF's attribute masqquerade few package globals for the time SUB run. SELF then locks the package, and $Data::Rlist::Locked is true. Here is an example for input preprocessed by cpp, which temporarily sets $Data::Rlist::SafeCppMode to 1:
$self = Data::Rlist->new(-SafeCppMode => 1, -filter => 1);
Reset or initialize object attributes, then return SELF. Each ATTRIBUTE is a name/value-pair. See new for a list of valid names. For example,
$obj->set(-input => \$str, -output => 'temp.rls', -options => 'squeezed');
Get some attribute NAME from object SELF. Unless NAME exists returns DEFAULT. The require method has no default value, hence it dies unless NAME exists. has returns true when NAME exists, false otherwise. For NAME the leading hyphen is optional. For example,
$self->get('foo'); # returns $self->{-foo} or undef
$self->get(-foo=>); # dto.
$self->get('foo', 42); # returns $self->{-foo} or 42
Parse data from INPUT, which specifies some Rlist-text. See also errors, write.
PARAMETERS
INPUT shall be either
- some Rlist object created by new,
- a string reference, in which case read and read_string parse Rlist text from it,
- a string scalar, in which case read assumes a file to parse.
See open_input for the FILTER and FILTER-ARGS parameters, which are used to preprocess an input file. When an input file cannot be open'd and flock'd this function dies. When INPUT is an object you specify FILTER and FILTER-ARGS to overload the -filter and -filter_args attributes.
RESULT
read returns the parsed data as array- or hash-reference, or undef if there was no data. The latter may also be the case when file consist only of comments/whitespace.
NOTES
This function may die. Dying is Perl's mechanism to raise exceptions, which eventually can be catched with eval. For example,
my $host = eval { use Sys::Hostname; hostname; } || 'some unknown machine';
This code fragment traps the die exception to that eval returns undef, or the result of calling hostname. The following example uses eval to trap exceptions thrown by read:
$object = new Data::Rlist(-input => $thingfile);
$thing = eval { $object->read };
unless (defined $thing) {
if ($object->errors) {
print STDERR "$thingfile has syntax errors"
} else {
print STDERR "$thingfile not found, is locked or empty"
}
} else {
# Can use $thing
.
.
}
Parse data from INPUT, which specifies some comma-separated-values (CSV) text. Both functions
- read data from strings or files,
- use an optional delimiter,
- ignore delimiters in quoted strings,
- ignore empty lines,
- ignore lines begun with #.
read_conf is a variant of read_csv dedicated to configuration files. Such files consist of lines of the form
key = value
That is, read_conf simply uses a default delimiter of '\s*=\s*', while read_csv uses
'\s*,\s*'. Hence read_csv can be used as well for configuration files. For example, a
delimiter of '\s+' splits the line at horizontal whitespace into multiple values (but, of
course, not from within quoted strings). For more information see ReadCSV, ReadConf,
write_csv and write_conf.
PARAMETERS
For INPUT see read. For FILTER, FILTER-ARGS see open_input. OPTIONS can be used to
set the "delimiter" regex. For read_csv the delimiter defaults to
'\s*,\s*', and for read_conf to '\s*=\s*'. These defaults are defined by the
$Data::Rlist::DefaultCsvDelimiter and $Data::Rlist::DefaultConfDelimiter.
RESULT
Both functions return a list of lists. Each embedded array defines the fields in a line, and may be of variable length.
EXAMPLES
Un/quoting of values happens implicitly. Given a file db.conf
# Comment
SERVER = hostname
DATABASE = database_name
LOGIN = "user,password"
the call
$opts = Data::Rlist::read_conf('db.conf');
returns (as $opts)
[ [ 'SERVER', 'hostname' ],
[ 'DATABASE', 'database_name' ],
[ 'LOGIN', 'user,password' ]
]
To convert such an array into a hash %conf, use
%conf = map { @$_ } @{ReadConf 'db.conf'};
The write_conf function can be used to update db.conf from $opts, so that
push @$opts, [ 'MAGIC VALUE' => 3.14_15 ];
Data::Rlist::write_conf('db.conf', { precision => 2 });
yields
SERVER = hostname
DATABASE = database_name
LOGIN = "user,password"
"MAGIC VALUE" = 3.1415
Calls read to parse Rlist language productions from the string or string-reference INPUT. INPUT may be an object-reference, in which case read_string attempts to parse the string-reference defined by the -input attribute.
Return the last result of calling read, which is either undef or some array- or hash-reference. When called as method (i.e., SELF is specified) returns the result that occured the last time SELF had called read.
Return undef or an array-ref of nanoscripts defined by the last call to read. When called as method returns the nanoscripts defined by the last time SELF had called read. The result has the form:
[ [ $hash_or_array_ref, $key_or_index ], # 1st nanoscript
[ $hash_or_array_ref, $key_or_index ], # 2nd nanoscript
.
.
.
]
This information defines the location of the embedded Perl script, which can then be evald. See als evaluate_nanoscripts.
Evaluates all nanoscripts defined by the last call to read. When called as method evaluates the nanoscripts defined by the last time SELF had called read. Returns the number of scripts or 0 if none were available. Each script is replaced by the result of eval'ing it. (See details here.)
You can eval embedded codes on your own, by processing the array returned by the nanoscripts function.
Returns the array of messages / syntax errors / warnings that occurred in the last call to read. Returns a list of strings. When called as method returns the messages that occured the last time SELF had called read.
Returns the number of times the last compile violated $Data::Rlist::MaxDepth. When called as method returns the information for the last time SELF had called compile. (Note that compile is not called directly, but through write.)
Returns true when the last call to parse yielded undef, because there was nothing to parse. When called as method returns the information for the last time SELF had called read.
Transliterates Perl data into Rlist text. write is auto-exported as WriteData.
PARAMETERS
DATA is either an object generated by new, or any Perl data including undef. When DATA is some Data::Rlist object the Perl data to be compiled is defined by its -data attribute. (When -data refers to another Rlist object, this other object is invoked.)
OUTPUT defines the place where to compile to (filename or some string-reference). When undef writes to some new string to which it returns a reference OUTPUT defaults to the -output attribute when DATA defines an object.
OPTIONS defines how to compile the text from DATA. The argument defaults to the -options
attribute when DATA is an object. When undef or "fast" uses compile_fast, when
"perl" uses compile_Perl, otherwise compile.
HEADER is a reference to an array of strings that shall be printed literally at the top of an output file. Defaults to the -header attribute when DATA is an object.
RESULT
When write creates a file it returns 0 for failure or 1 for success. Otherwise it returns a string reference.
EXAMPLES
$self = new Data::Rlist(-data => $thing, -output => $output);
$self->write; # Compile $thing into a file ($output is a filename)
# or string ($output is a string reference).
Data::Rlist::write($thing, $output); # dto., but using the functional interface.
print $self->write_string_value; # Print $thing to STDOUT.
print Data::Rlist::write_string_value($thing); # dto.
PrintData($thing); # dto.
Write DATA as comma-separated-values (CSV) to file or string OUTPUT. write_conf writes
configuration files where each line contains a tagname, a separator and a value. The main
difference between write_conf and write_csv are the default values for "separator" and
"auto_quote".
PARAMETERS
For DATA and OUTPUT see write. DATA defines the data to be compiled. But because of the limitations of CSV-files this may not be just any Perl data. It must be a reference to an array of array references. For example,
[ [ a, b, c ], # line 1
[ d, e, f, g ], # line 2
.
.
]
and for write_conf
[ [ tag, value ], # line 1
.
.
]
From OPTIONS is read the comma-separator ("separator"), how to quote
("auto_quote"), the linefeed ("eol_space") and the numeric precision ("precision"). The
defaults are:
FUNCTION SEPARATOR AUTO-QUOTING
-------- --------- ------------
write_csv() ',' no
write_conf() ' = ' yes
Optionally COLUMNS (as an array-referernce) specify the column names to be written as the first line. The optional HEADER array is written as #-comments before the actual data. When called as methods, DATA, OPTIONS, COLUMNS and HEADER defaults to the value of the -data, -options, -columns and -header attributes.
Note that write_csv uses the current value of $/ to separate lines. When called as method you may temporarily overload $/ using the -InputRecordSeparator attribute.
RESULT
When a file was created both function return 0 for failure, or 1 for success. Otherwise they return a string reference (the compiled text).
EXAMPLES
Functional interface:
use Data::Rlist; # imports WriteCSV
WriteCSV($thing, "foo.dat");
WriteCSV($thing, "foo.dat", { separator => '; ' }, [qw/GBKNR VBKNR EL LaD/]);
WriteCSV($thing, \$target_string);
$string_ref = WriteCSV($thing);
Object-oriented interface:
$object = new Data::Rlist(-data => $thing, -output => "foo.dat",
-options => { separator => '; ' },
-columns => [qw/GBKNR VBKNR EL LaD LaD_V/]);
$object->write_csv; # Write $thing as CSV to foo.dat
$object->write; # Write $thing as Rlist to foo.dat
$object->set(-output => \$target_string);
$object->write_csv; # Write $thing as CSV to $target_string
Please see read_csv for more examples.
Stringify any Perl data DATA and return a reference to the string. Works like write but
always compiles to a new string to which it returns a reference. Consequently, when called as
method this function does not use the -output and -options attributes, and the default for
OPTIONS is "string".
Stringify any Perl dat DATA and return the compiled text string value. OPTIONS default to
"default". For example,
print "\n\$thing dumped: ", Data::Rlist::write_string_value($thing);
$self = new Data::Rlist(-data => $thing);
print "\nsame \$thing dumped: ", $self->write_string_value;
Do a deep copy of DATA according to OPTIONS. DATA is any Perl data or some Data::Rlist object. keelhaul first compiles arbitary Perl data to Rlist text, then restores the data from exactly this text. By ``keelhauling data'' one can therefore
- adjust the accuracy of numbers,
- break circular-references and
- drop \*foo{THING}s.
This function is useful when DATA had been hatched by some other code, and you don't know whether it is hierachical, or if typeglob-refs nist inside. You may then simply keelhaul it to clean it from its (wild) past. Multiple data sets can so be brought to the same, common basis. For example, to bring all numbers in
$thing = { foo => [ [ .00057260 ], -1.6804e-4 ] };
to a certain accuracy, use
$deep_copy_of_thing = Data::Rlist::keelhaul($thing, { precision => 4 });
All number scalars in $thing are are rounded to 4 decimal places, so they're finally comparable as floating-point numbers. To $deep_copy_of_thing is assigned the hash-reference
{ foo => [ [ 0.0006 ], -0.0002 ] }
Likewise one can convert all floats to integers:
$make_integers = new Data::Rlist(-data => $thing, -options => { precision => 0 });
$thing_without_floats = $make_integers->keelhaul;
When keelhaul is called in an array context it also returns the text from which the copy had been built. For example,
$deep_copy = Data::Rlist::keelhaul($thing);
($deep_copy, $rlist_text) = Data::Rlist::keelhaul($thing);
$deep_copy = new Data::Rlist(-data => $thing)->keelhaul;
It is then guarantee that the following statement never throws:
die if deep_compare($deep_copy, ReadData(\$rlist_text));
NOTES
keelhaul won't throw die nor return an error, but be prepared for the following effects:
ARRAY, HASH, SCALAR and REF references were compiled, whether blessed or not. (Since compiling does not store type information, keelhaul will turn blessed references into barbars again.)
IO, GLOB and FORMAT references have been converted into strings.
Depending on the compile options, CODE references were invoked, deparsed back into their function bodies, or dropped.
Depending on the compile options floats have been rounded, or have been converted to integers.
undef'd array elements had been converted into the default scalar value "".
Anything deeper than $Data::Rlist::MaxDepth had been thrown away. However, this only would happen when $Data::Rlist::MaxDepth is not 0.
When the data contains objects, no special methods are triggered to ``freeze'' and ``thaw'' the objects.
See also compile, equal and deep_compare
Get the hash-ref $Data::Rlist::PredefinedOptions{PREDEF-NAME}. PREDEF-NAME defaults to
"default" (i.e., the options for writing files).
Completes OPTIONS with BASIC-OPTIONS: all pairs not already in OPTIONS are copied from
BASIC-OPTIONS. Both arguments define hashes or some predefined options name, and default to "default". This function returns a new hash of
compile options. (Even when OPTIONS defines a hash it is copied into a new
one.) For example,
$options = complete_options({ precision => 0 }, 'squeezed')
merges the predefined options for "squeezed" text (no whitespace at all,
no here-docs, numbers rounded) with a numeric precision of 0. This converts all floats to
integers. The following call completes $them by some other hash:
$options = complete_options($them, { delimiter => '\s+' })
That is, it copies "delimiter" unless such a key already exists into $them. Note that
$them itself isn't modified.
Open/close Rlist text file or string INPUT for parsing. Used internally by read and read_csv.
PREPROCESSING
The function can preprocess the INPUT file using FILTER. Use the special value 1 to select the default C preprocessor (gcc -E -Wp,-C). FILTER-ARGS is an optional string of additional command-line arguments to be appended to FILTER. For example,
my $foo = Data::Rlist::read("foo", 1, "-DEXTRA")
eventually does not parse foo, but the output of the command
gcc -E -Wp,-C -DEXTRA foo
Hence within foo C-preprocessor-statements become possible
{
#ifdef EXTRA
#include "extra.rlist"
#endif
123 = (1, 2, 3);
foobar = {
.
.
SAFE CPP MODE
This mode uses sed and a temporary file. It is enabled by setting $Data::Rlist::SafeCppMode to 1 (the default is 0). It protects single-line #-comments when FILTER begins with either gcc, g++ or cpp. open_input then additionally runs sed to convert all input lines beginning with whitespace plus the # character. Only the following cpp-commands are excluded, and only when they appear in column 1:
- #include and #pragma
- #define and #undef
- #if, #ifdef, #else and #endif.
For all other lines sed converts # into ##. This prevents the C preprocessor from evaluating them. Because of Perl's limited open function, which isn't able to dissolve long pipes, the invocation of sed requires a temporary file. The temporary file is created in the same directory as the input file. When you only use // and /* */ comments, however, ``Safe CPP Mode'' is not required.
Lexical scanner. Called by parse to split the current line into tokens. lex reads # or // single-line-comment and /* */ multi-line-comment as regular white-spaces. Otherwise it returns tokens according to the following table:
RESULT MEANING
------ -------
'{' '}' Punctuation
'(' ')' Punctuation
',' Operator
';' Punctuation
'=' Operator
'v' Constant value as number, string, list or hash
'??' Error
undef EOF
lex appends all here-doc-lines with a newline character. For example,
<<test1
a
b
test1
is effectively read as "a\nb\n", which is the same value as the equivalent here-doc in Perl has.
So, not all strings can be encoded as a here-doc. For example, it might not be quite obvious to
many programmers that "foo\nbar" cannot be expressed as here-doc.
Read the next line of text from the current input. Return 0 if at_eof, 1 otherwise.
Return true if current input file/string is exhausted, false otherwise.
Read Rlist language productions from current input. This is a fast, non-recursive parser driven by the parser map %Data::Rlist::Rules, and fed by lex. It is called internally by read.
Build Rlist text from any Perl data DATA. When FH is defined compile directly to this file and
return 1. Otherwise (FH is undef) build a string and return a reference to it. This is the
compilation function called when the OPTIONS argument passed to write
orwrite_string is not omitted, and is not "fast" or "perl". DATA is compiled as
follows:
Reference-types SCALAR, HASH, ARRAY and REF are compiled into text, whether blessed or not.
Reference-types CODE are compiled depending on the "code_refs" setting in
OPTIONS.
Reference-types GLOB (typeglob-refs), IO and FORMAT (file-
and directory handles) cannot be dissolved, and are compiled into the strings "?GLOB?",
"?IO?" and "?FORMAT?".
undef'd values in arrays are compiled into the default Rlist "".
Build Rlist text from any Perl data DATA. Do this as fast as actually possible with pure Perl.
Note that this is the default compilation function called when OPTIONS are omitted, or "fast" is
passed (see write and write_string). DATA is compiled as follows:
Reference-types SCALAR, HASH, ARRAY and REF are compiled into text, whether blessed or not.
CODE, GLOB, IO and FORMAT are compiled into the strings "?CODE?", "?IO?",
"?GLOB?" and "?FORMAT?".
undef'd values in arrays are compiled into the default Rlist "".
The main difference to compile is that compile_fast considers no compile options. Thus it cannot call code, implicitly round numbers, and cannot detect recursively-defined data. Also compile_fast returns a reference to the compiled string, which is a reference to a unique package variable. Subsequent calls to compile_fast reassign this variable. Because of this behaviors, compile_fast is very... fast!
Like compile_fast, but do not compile Rlist text - compile DATA into Perl syntax. It can
then be eval'd. This renders more compact, and more exact output as the Data::Dumper manpage. For
example, only strings are quoted. To enable this compilation function you must pass "perl" to
write and write_string, as the OPTIONS argument.
The utility functions in this section are generally useful when handling stringified data. These functions are either very fast, or smart, or both. For example, quote, unquote, escape and unescape internally use precompiled regexes and precomputed ASCII tables. For this employing these functions should be faster then using own variants.
Returns true when a scalar looks like a positive or negative integer constant. The function applies the compiled regex $Data::Rlist::REInteger.
Test for strings that look like numbers. is_number can be used to test whether a scalar looks like a integer/float constant (numeric literal). The function applies the compiled regex $Data::Rlist::REFloat. Note that it doesn't match
- leading or trailing whitespace,
- lexical conventions such as the "0b" (binary), "0" (octal), "0x" (hex) prefix to denote a
number-base other than decimal, and
- Perls' legible numbers, e.g. 3.14_15_92,
- the IEEE 754 notations of Infinite and NaN.
See also
$ perldoc -q "whether a scalar is a number"
Test for symbolic names. is_symbol can be used to test whether a scalar looks like a symbolic name. Such strings need not to be quoted. Rlist defines symbolic names as a superset of C identifier names:
[a-zA-Z_0-9] # C/C++ character set for identifiers
[a-zA-Z_0-9\-/\~:\.@] # Rlist character set for symbolic names
[a-zA-Z_][a-zA-Z_0-9]* # match C/C++ identifier
[a-zA-Z_\-/\~:@][a-zA-Z_0-9\-/\~:\.@]* # match Rlist symbolic name
For example, scoped/structured names such as std::foo, msg.warnings, --verbose,
calculation-info need not be quoted. Note that is_symbol does not catch leading or
trailing whitespace. Another restriction is that "." cannot be used as first character, since it
could also begin a number.
Returns true when the scalar is an integer, a number, a symbolic name or some string returned by quote.
The opposite of is_value. On such texts compile amd compile_fast would call quote.
Converts TEXT into 7-bit-ASCII. All characters not in the set of the 95 printable ASCII characters are escaped (see below). The following ASCII codes will be converted to escaped octal numbers, i.e. 3 digits prefixed by a slash:
0x00 to 0x1F
0x80 to 0xFF
" ' \
The difference between the two functions is that quote additionally places TEXT into
double-quotes. For example, quote(qq'``Früher Mittag\n''') returns "\"Fr\374her
Mittag\n\"", while escape returns \"Fr\374her Mittag\n\"
Return quote(TEXT) if is_random_text(TEXT); otherwise (TEXT defines a symbolic name or number) return TEXT.
Return unquote(TEXT) when the first character of TEXT is "; otherwise returns TEXT.
HERE-DOC-STRING shall be a here-document. The function checks whether each line begins with a common prefix, and if so, strips that off. If no prefix it takes the amount of leading whitespace found the first line and removes that much off each subsequent line.
Unless COLUMNS is defined returns the new here-doc-string. Otherwise, takes the string and
reformats it into a paragraph having no line more than COLUMNS characters long. FIRSTTAB will be
the indent for the first line, DEFAULTTAB the indent for every subsequent line. Unless passed,
FIRSTTAB and DEFAULTTAB default to the empty string "".
This function combines recipes 1.11 and 1.12 from the Perl Cookbook.
Divide the string INPUT into a list of strings. DELIMITER is a regular expression specifying where
to split (default: '\s+'). The function won't split at DELIMITERs inside quotes, or which are
backslashed. For example, to split INPUT at commas use '\s*,\s*'.
parse_quoted works like split_quoted but additionally removes all quotes and backslashes
from the splitted fields. Both functions effectively simplify the interface of
Text::ParseWords. In an array context they return a list of substrings, otherwise the count of
substrings. An empty array is returned in case of unbalanced double-quotes, e.g.
split_quoted('foo,"bar').
EXAMPLES
sub split_and_list($) {
print ($i++, " '$_'\n") foreach split_quoted(shift)
}
split_and_list(q("fee foo" bar))
0 '"fee foo"'
1 'bar'
split_and_list(q("fee foo"\ bar))
0 '"fee foo"\ bar'
The default DELIMITER '\s+' handles newlines. split_quoted("foo\nbar\n") returns
('foo', 'bar', '') and hence can be used to to split a large string of uncho(m)p'd input
lines into words:
split_and_list("foo \r\n bar\n")
0 'foo'
1 'bar'
2 ''
The DELIMITER matches everywhere outside of quoted constructs, so in case of the default '\s+'
you may want to remove heading/trailing whitespace. Consider
split_and_list("\nfoo")
split_and_list("\tfoo")
0 ''
1 'foo'
and
split_and_list(" foo ")
0 ''
1 'foo'
2 ''
parse_quoted additionally removes all quotes and backslashes from the splitted fields:
sub parse_and_list($) {
print ($i++, " '$_'\n") foreach parse_quoted(shift)
}
parse_and_list(q("fee foo" bar))
0 'fee foo'
1 'bar'
parse_and_list(q("fee foo"\ bar))
0 'fee foo bar'
MORE EXAMPLES
String 'field\ one "field\ two"':
('field\ one', '"field\ two"') # split_quoted
('field one', 'field two') # parse_quoted
String 'field\,one, field", two"' with a DELIMITER of '\s*,\s*':
('field\,one', 'field", two"') # split_quoted
('field,one', 'field, two') # parse_quoted
Split a large string $soup (mnemonic: slurped from a file) into lines, at LF or CR+LF:
@lines = split_quoted($soup, '\r*\n');
Then transform all @lines by correctly splitting each line into ``naked'' values:
@table = map { [ parse_quoted($_, '\s*,\s') ] } @lines
Here is some more complete code to parse a .csv-file with quoted fields, escaped commas:
open my $fh, "foo.csv" or die $!;
local $/; # enable localized slurp mode
my $content = <$fh>; # slurp whole file at once
close $fh;
my @lines = split_quoted($content, '\r*\n');
die q(unbalanced " in input) unless @lines;
my @table = map { [ map { parse_quoted($_, '\s*,\s') } ] } @lines
In core this is what read_csv does. A nice way to make sure what split_quoted and parse_quoted return is using deep_compare. For example, the following code shall never die:
croak if deep_compare([split_quoted("fee fie foo")], ['fee', 'fie', 'foo']);
croak if deep_compare( parse_quoted('"fee fie foo"'), 1);
The 2nd call to parse_quoted happens in scalar context, hence shall return 1 because there's one string to parse.
Compare and round floating-point numbers. NUM1 and NUM2 are string- or number scalars. equal returns true if NUM1 and NUM2 are equal to PRECISION number of decimal places (default: 6).
Normally round will return a number in fixed-point notation. When the package-global $Data::Rlist::RoundScientific is true, however, round formats the number in either normal or exponential (scientific) notation, whichever is more appropriate for its magnitude. This differs slightly from fixed-point notation in that insignificant zeroes to the right of the decimal point are not included. Also, the decimal point is not included on whole numbers. For example, round(42) does not return 42.000000, and round(0.12) returns 0.12, not 0.120000.
MACHINE ACCURACY
One needs a function like equal to compare floats, because IEEE 754 single- and double precision implementations are not absolute - in contrast to the numbers they actually represent. In all machines non-integer numbers are only an approximation to the numeric truth. In other words, they're not commutative. For example, given two floats a and b, the result of a+b might be different than that of b+a. For another example, it is a mathematical truth that a * b = b * a, but not necessarily in a computer.
Each machine has its own accuracy, called the machine epsilon, which is the difference between 1 and the smallest exactly representable number greater than one. Most of the time only floats can be compared that have been carried out to a certain number of decimal places. In general this is the case when two floats that result from a numeric operation are compared - but not two constants. (Constants are accurate through to lexical conventions of the language. The Perl and C syntaxes for numbers simply won't allow you to write down inaccurate numbers.)
See also recipes 2.2 and 2.3 in the Perl Cookbook.
EXAMPLES
CALL RETURNS NUMBER
---- --------------
round('0.9957', 3) 0.996
round(42, 2) 42
round(0.12) 0.120000
round(0.99, 2) 0.99
round(0.991, 2) 0.99
round(0.99, 1) 1.0
round(1.096, 2) 1.10
round(+.99950678) 0.999510
round(-.00057260) -0.000573
round(-1.6804e-6) -0.000002
Compare and analyze two numbers, strings or references. Generates a list of messages describing exactly all unequal data. Hence, for any Perl data $a and $b one can assert:
croak "$a differs from $b" if deep_compare($a, $b);
When PRECISION is defined all numbers in A and B are rounded before actually comparing them. When TRACE_FLAG is true traces progress on STDOUT.
RESULT
Returns an array of messages, each describing unequal data, or data that cannot be compared because of type- or value-mismatching. The array is empty when deep comparison of A and B found no unequal numbers or strings, and only indifferent types.
EXAMPLES
The result is line-oriented, and for each mismatch it returns a single message:
Data::Rlist::deep_compare(undef, 1)
yields
<<undef>> cmp <<1>> stop! 1st undefined, 2nd defined (1)
Some more complex example. Deep-comparing two multi-level data structures A and B returned two messages:
'String literal' == REF(0x7f224) stop! type-mismatch (scalar versus REF)
'Greetings, earthlings!' == CODE(0x7f2fc) stop! type-mismatch (scalar versus CODE)
Somewhere in A a string "String literal" could not be compared, because the corresponding
element in B is a reference to a reference. Next it says that "Greetings, earthlings!" could not
be compared because the corresponding element in B is a code reference.
Actually, A and B are identical. B was written to disk (by write) and then read back as A (by read). So, why don't they compare anymore? Because in B the refs REF(0x7f224) and CODE(0x7f2fc) hide
\"String literal"
and
sub { 'Greetings, earthlings!' }
When writing B to disk write has dissolved the scalar- and the code-reference into "String
literal" and "Greetings, earthlings!". Of course, deep_compare will not do that, so A does
not compare to B anymore. Note that despite these two mismatches, deep_compare had continued
the comparison for all other elements in A and B. Hence the structures are otherwise identical.
Forks a process and waits for completion. The function will extract the exit-code, test whether the process died and prints status messages on STDERR. fork_and_wait hence is a handy wrapper around the built-in system and exec functions. Returns an array of three values:
($exit_code, $failed, $coredump)
$exit_code is -1 when the program failed to execute (e.g. it wasn't found or the current user has insufficient rights). Otherwise $exit_code is between 0 and 255. When the program died on receipt of a signal (like SIGINT or SIGQUIT) then $signal stores it. When $coredump is true the program died and a core-file was written.
Concatenates and forms all TEXT strings into a symbolic name that can be used as a pathname.
synthesize_pathname is a useful function to concatenate strings and nearby converting all
characters that do not qualify as filename-characters, into "_" and "-". Effectively this
function returns a symbolic name. The result cannot only be used as file- or URL name, but also
(coinstantaneously) as hash key, database name etc.
Make compile round all numbers to PLACES decimal places, by calling round on each scalar that looks like a number. By default PLACES is undef, which means floats are not rounded.
Causes compile to masquerade $Data::Rlist::RoundScientific. See round for what this means.
Defines how compile shall treat CODE reference. Legal values for TOKEN are 0 (the
default), "call" and "deparse".
A TOKEN value of 0 compiles subroutine references into the string "?CODE?". A value of "call"
calls the code, then compiles the return value. "deparse" serializes the code using
B::Deparse, which reproduces the Perl source. Note that it then makes sense to enable
"here_docs" (see below), because otherwise the deparsed code will be in one string with LFs
quoted as "\n". This causes no harm, but when opened in a text editor the data will be more
legible.
If enabled compile internally use multiple threads. Note that this makes only sense on machines with at least COUNT CPUs.
If enabled strings with at least two newlines in them are written as
here-document, when possible. To qualify as here-document a string has to have
at least two LFs ("\n"), one of which must terminate it.
When true (default) do not quote strings that look like identifiers (determined by is_symbol). When false quote all strings. Hash keys are not affected.
write_csv and write_conf interpret this flag differently. False means not to quote at all. True quotes only strings that don't look like numbers and that aren't yet quoted.
Use "eol_space" (linefeed) to ``distribute data on many lines.'' Insert a linefeed after every
NUMBERth array value; 0 disables outlining.
If enabled, and "outline_data" is also enabled, prints { and } on distinct lines when
compiling Perl hashes with at least one pair.
The comma-separator string to be used by write_csv. The default is ','.
Field-delimiter for read_csv. There is no default value. To read configuration files,
for example, you may use '\s*=\s*' or '\s+'; and to read CSV-files you may use
'\s*[,;]\s*'.
The following options format the generated Rlist; normally you don't want to modify them:
Count of physical, horizontal TAB characters to use at the begin-of-line per indentation level. Defaults to 1. Note that we don't use blanks, because they blow up the size of generated text without measure.
End-of-line string to use (the linefeed). For example, legal values are "", " ", "\n",
"\r\n" etc. The default is undef, which means to use the current value of $/.
Note that this is a compile-option that only affects compile. When parsing files the builtin readline function is called, which uses $/.
String to write after ( and {, and before } and ) when compiling arrays and hashes.
Comma and semicolon strings, which shall be at least "," and ";". No matter what,
compile will always print the "eol_space" string after the "semicolon_punct" string.
String to make up key/value-pairs. Defaults to " = ". Note that this is a compile option: the
parser always expects some "=" to designate a pair.
The OPTIONS parameter accepted by some package functions is either a hash-ref or the name of a predefined set:
Default if writing to a file.
Compact, no newlines/here-docs. Renders a ``string of data''.
Optimize the compiled Rlist for maximum readability.
Very compact, no whitespace at all. For very large Rlists.
Compile data in Perl syntax, using compile_Perl, not compile. The output then can be eval'd, but it cannot be read back.
Compile data as fast as possible, using compile_fast, not compile.
All functions that define an OPTIONS parameter implicitly call
complete_options to complete the argument from one of the predefined sets, and
"default". Therefore you may just define a ``lazy subset of options'' to these functions. For
example,
my $obj = new Data::Rlist(-data => $thing);
$obj->write('thing.rls', { scientific => 1, precision => 8 });
=x
Example:
use Data::Rlist qw/:floats :strings/;
Imports maybe_quote, quote, escape, unquote, unescape, unhere, is_random_text, is_number, is_symbol, split_quoted, and parse_quoted.
Imports predefined_options and complete_options.
Imports deep_compare, fork_and_wait and synthesize_pathname.
The following functions are implicitly imported into the callers symbol table. (But you may say require Data::Rlist instead of use Data::Rlist to prohibit auto-import. See also perlmod.)
These are aliases for Data::Rlist::read, Data::Rlist::read_csv and Data::Rlist::read_conf.
These are aliases for Data::Rlist::write, Data::Rlist::write_string Data::Rlist::write_csv and Data::Rlist::write_conf.
These are aliases for Data::Rlist::write_string_value. OutlineData applies the
predefined "outlined" options, while StringizeData applies
"string" and SqueezeData() "squeezed". When
specified, OPTIONS are merged into the predefined set by means of complete_options. For
example,
print "\n\$thing: ", OutlineData($thing, { precision => 12 });
rounds all numbers in $thing to 12 digits.
Just another way for
print OutlineData(DATA, OPTIONS);
These are aliases for keelhaul and deep_compare. For example,
use Data::Rlist;
.
.
my($copy, $as_text) = KeelhaulData($thing);
String- and number values:
"Hello, World!"
foo # compiles to { 'foo' => undef }
3.1415 # compiles to { 3.1415 => undef }
Array values:
(1, a, 4, "b u z") # list of numbers/strings
((1, 2),
(3, 4)) # list of list (4x4 matrix)
((1, a, 3, "foo bar"),
(7, c, 0, "")) # another list of lists
Here-document strings:
$hello = ReadData(\<<HELLO)
( <<DEUTSCH, <<ENGLISH, <<FRANCAIS, <<CASTELLANO, <<KLINGON, <<BRAINF_CK )
Hallo Welt!
DEUTSCH
Hello World!
ENGLISH
Bonjour le monde!
FRANCAIS
Ola mundo!
CASTELLANO
~ nuqneH { ~ 'u' ~ nuqneH disp disp } name
nuqneH
KLINGON
++++++++++[>+++++++>++++++++++>+++>+<<<<-]>++.>+.+++++++
..+++.>++.<<+++++++++++++++.>.+++.------.--------.>+.>.
BRAINF_CK
HELLO
Compiles $hello as
[ "Hallo Welt!\n", "Hello World!\n", "Bonjour le monde!\n", "Ola mundo!\n",
"~ nuqneH { ~ 'u' ~ nuqneH disp disp } name\n",
"++++++++++[>+++++++>++++++++++>+++>+<<<<-]>++.>+.+++++++\n..+++.>++.<<+++++++++++++++.>.+++.------.--------.>+.>.\n" ]
Configuration object as hash:
{
contribution_quantile = 0.99;
default_only_mode = Y;
number_of_runs = 10000;
number_of_threads = 10;
# etc.
}
Altogether:
Metaphysic-terms =
{
Numbers =
{
3.141592653589793 = "The ratio of a circle's circumference to its diameter.";
2.718281828459045 = <<___;
The mathematical constant "e" is the unique real number such that the value of
the derivative (slope of the tangent line) of f(x) = e^x at the point x = 0 is
exactly 1.
___
42 = "The Answer to Life, the Universe, and Everything.";
};
Words =
{
ACME = <<Value;
A fancy-free Company [that] Makes Everything: Wile E. Coyote's supplier of equipment and gadgets.
Value
<<Key = <<Value;
foo bar foobar
Key
[JARGON] A widely used meta-syntactic variable; see foo for etymology. Probably
originally propagated through DECsystem manuals [...] in 1960s and early 1970s;
confirmed sightings go back to 1972. [...]
Value
};
};
The Random Lists (Rlist) syntax is inspired by NeXTSTEP's Property Lists. But Rlist is simpler, more readable and more portable. The Perl, Python and C++ implementations are fast, stable and free. Markus Felten, with whom I worked a few month in a project at Deutsche Bank, Frankfurt in summer 1998, arrested my attention on Property lists. He had implemented a Perl variant of it (http://search.cpan.org/search).
The term ``Random'' underlines the fact that the language
has four primitive/anonymuous types;
the basic building block is a list, which is combined at random with other lists.
Hence the term Random does not mean aimless or accidental. Random Lists are arbitrary lists.
Rlists are not Perl syntax:
RLIST PERL
----- ----
5; { 5 => undef }
"5"; { "5" => undef }
5=1; { 5 => 1 }
{5=1;} { 5 => 1 }
(5) [ 5 ]
{} { }
; { }
() [ ]
=head2 Debugging Data (Finding Self-References)
To reduce recursive data structures (into true hierachies) set $Data::Rlist::MaxDepth to an integer above 0. It then defines the depth under which compile shall not venture deeper. The compilation of Perl data (into Rlist text) then continues, but on STDERR a message like the following is printed:
ERROR: compile2() broken in deep ARRAY(0x101aaeec) (depth = 101, max-depth = 100)
This message will also be repeated as comment when the compiled Rlist is written to a file. Furthermore $Data::Rlist::Broken is incremented by one. While the compilation continues, effectively any attempt to venture deeper as suggested by $Data::Rlist::MaxDepth will be blocked. See also broken.
Much work has been spent to optimize Data::Rlist for speed. Still it is implemented in pure Perl (no XS). A very rough estimation for Perl 5.8 is ``each MB takes one second per GHz''. For example, when the resulting Rlist file has a size of 13 MB, compiling it from a Perl script on a 3-GHz-PC requires about 5-7 seconds. Compiling the same data under Solaris, on a sparcv9 processor operating at 750 MHz, takes about 18-22 seconds.
The process of compiling can be speed up by calling quote explicitly on scalars. That is, before calling write or write_string. Big data sets may compile faster when for scalars, that certainly not qualify as symbolic name, quote is called in advance:
use Data::Rlist qw/:strings/;
$data{quote($key)} = $value;
.
.
Data::Rlist::write("data.rlist", \%data);
instead of
$data{$key} = $value;
.
.
Data::Rlist::write("data.rlist", \%data);
It depends on the case whether the first variant is faster: compile and
compile_fast both have to call is_random_text on each scalar. When the scalar is
already quoted, i.e., its first character is ", this test ought to run faster.
Note that internally is_random_text applies the precompiled regex $Data::Rlist::REValue. But for a given scalar $s the expression ($s !~ $Data::Rlist::REValue) can be up to 20% faster than the equivalent is_random_text($s).
Normally you don't have to care about strings, since un/quoting happens as required when reading/compiling Rlist or CSV text. A common problem, however, occurs when some text fragment (string) uses the same lexicography than numbers do.
Perl defines the string as the basic building block for all program data, then lets the program decide what strings mean. Analogical, in a printed book the reader has to decipher the glyphs and decide what evidence they hide. Printed text uses well-defined glyphs and typographic conventions, and finally the competence of the reader, to recognize numbers. But computers need to know the exact number type and format. Integer? Float? Hexadecimal? Scientific? Klingon? The Perl Cookbook recommends the use of a regular expression to distinguish number from string scalars (recipe 2.1).
In Rlist, string scalars that look like numbers need to be quoted explicitly. Otherwise, for
example, the string scalar "-3.14" appears as -3.14 in the output. Likewise "007324" is
compiled into 7324. Then the text quality is lost and the scalar is read back as a number. Of
course, this behavior is by intend, and in most cases this is just what you want. For hash keys,
however, it might be a problem. One solution is to prefix the string with "_":
my $s = '-9'; $s = "_$s";
Such strings do not qualify as a number anymore. In the C++ implementation it will then become
some std::string, not a double. But the leading "_" has to be removed by the reading
program. Perhaps a better solution is to explicitly call Data::Rlist::quote:
$k = -9;
$k = Data::Rlist::quote($k); # returns qq'"-9"'
use Data::Rlist qw/:strings/;
$k = 3.14_15_92;
$k = quote($k); # returns qq'"3.141592"'
Again, the need to quote strings that look like numbers is a problem evident only in the Perl implementation of Rlist, since Perl is a language with weak types. As a language with very strong typing C++ is quasi the antipode to Perl. With the C++ implementation of Rlist then there's no need to quote strings that look like numbers. See also write, is_number, is_symbol, is_random_text and http://en.wikipedia.org/wiki/American_Standard_Code_for_Information_Interchange.
Installing CPAN packages usually requires administrator privileges. Another way is to copy the Rlist.pm file into a directory of your choice, e.g. into . or ~/bin. Instead of use Data::Rlist;, however, you then use the following code. It finds Rlist.pm also in . and ~/bin, and then calls the Exporter manually:
BEGIN {
$0 =~ /[^\/]+$/;
push @INC, $`||'.', "$ENV{HOME}/bin";
require Rlist;
Data::Rlist->import();
Data::Rlist->import(qw/:floats :strings/);
}
Data::Rlist depends only on few other packages:
Exporter
Carp
strict
integer
Sys::Hostname
Scalar::Util # deep_compare() only
Text::Wrap # unhere() only
Text::ParseWords # split_quoted(), parse_quoted() only
Data::Rlist is free of $&, $` or $'. Reason: once Perl sees that you need one of these meta-variables anywhere in the program, it has to provide them for every pattern match. This may substantially slow your program (see also perlre).
This is supplement information for compile, the function internally called by write and write_string to ompile Rlist text from any Perl data. Typeglobs are a problem. Assume the variables $foo and %foo are in use. Then it is confusing that *foo does not return
(SCALAR => \$foo, HASH => \@foo)
Instead it simply says *foo is *main::foo:
$ perl -e 'print *foo'
*main::foo
$ perl -e '$foo = 42; @foo = (0); print *foo'
*main::foo
Typeglobs are an idiosyncracy of Perl. Typeglob objects are symbol table entries. Perl uses a symbol table per package (namespace) to map symbolic names like foo to Perl values. (Humans use abstract symbols to name things, because we can remember symbols better than numbers, or formulas that hide numbers :-)
The idiosyncracy is that different types need only one entry - one symbol can name all types of
Perl data (scalars, arrays, hashes) and nondata (functions, formats, I/O handles). For example,
the symbol foo is mapped to the typeglob *foo. Therein coexist $foo (the scalar value),
@foo (the list value), %foo (the hash value), &foo (the code value) and foo (the I/O
handle or the format specifier). There's no key "$foo" or "@foo" in the symbol table, only
"foo".
The symbol table is an ordinary hash, named like the package with two colons appended. The main symbol table's name is thus %main::, or %::. Internally this is called a stash (for symbol table hash). perl will create one stash per package.
In the C code that implements Perl, %:: is the global variable defstash (default stash). It holds items in the main package. But, as if it were a symbol in a stash, perl arranges it as typeglob-ref:
$ perl -e 'print \*::'
GLOB(0x10010f08)
But the root-stash defstash lists stashes from all other packages. For example, the symbol Data:: in stash %:: addresses the stash of package Data, and the symbol Rlist:: in the stash %Data:: addresses the stash of package Data::Rlist.
Like all hashes stashes contain string keys, which name symbols, and values which are typeglobs. In the C implementation of Perl typeglobs have the struct type GV, for Glob value. In the stashes, typeglobs are GV pointers.
The typeglob is interposed between the stash and the program's actual values for $foo, @foo etc.
The sigil * serves as wildcard for the other sigils %, @, $ and &. (A sigil is a symbol created for a specific magical purpose; the name derives from the latin sigilum = seal.)
\*names:: are actually stash-refs, but Perl calls them globs.
Modifying $foo in a Perl program won't change %foo. Each typeglob is merely a set of pointers to separate objects describing scalars, arrays, hashes, functions, formats and I/O handles. Normally only one pointer in *foo is non-null. Because typeglobs host pointers, *foo{ARRAY} is a way to say \@foo. To get a reference to the typeglob for symbol *foo you say *foo{GLOB}, or \*foo. But it is not quite clear why perl this is an error:
$ perl -e 'exists *foo{GLOB}'
exists argument is not a HASH or ARRAY element at -e line 1.
To define the scalar pointer in the typeglob *foo you simply say $foo = 42. But you may also assign a reference to the typeglob:
$ perl -e '$x = 42; *foo = \$x; print $foo'
42
Assigning a scalar alters the symbol, not the typeglob:
$ perl -e '$x = 42; *foo = $x; print *foo'
*main::42
$ perl -e '$x = 42; *foo = $x; print *42'
*main::42
Consider also:
$ perl -e 'print 1*9'
9
$ perl -e 'print *9'
*main::9
And also:
$ perl -e '*foo = 42; print $::{42}, *foo'
*main::42*main::42
IMHO it should not do that.
Maybe the best use of typeglobs are Typeglob-aliases. For example, *bar = *foo aliases the symbol bar in the stash. Then the symbols foo and bar point to the same typeglob! This means that when you declare sub foo {} after casting the alias, bar is foo. The penalty, however, is that the bar symbol cannot be easily removed from the stash. One way is to say local *bar, wich temporarily assigns a new typeglob to bar (in its stash) with all pointers zeroized.
What is this good for? This is not quite clear. Obviously an artefact from Perl4, it once made old scripts compatible with Perl5. In fact, local typeglob aliases seem to be faster than references, because no dereferencing is required. For example,
void f1 { my $bar = shift; ++$$bar }
void f2 { local *bar = shift; ++$bar }
f1(\$foo); # increments $foo
f1(*foo); # dto., but faster
Note, however, that my variables (lexical variables) are not stored in stashes, and do not use typeglobs. These variables are stored in a special array, the scratchpad, assigned to each block, subroutine, and thread. These are really private variables, and they cannot be localized. Each lexical variable occupies a slot in the scratchpad; hence is addressed by an integer index, not a symbol. my variables are like auto variables in C. They're also faster than locals, because they can be allocated at compile time, not runtime. Therefore you cannot declare *foo lexically:
$ perl -e 'my(*foo);'
Can't declare ref-to-glob cast in "my" at -e line 1, near ");"
Execution of -e aborted due to compilation errors.
The stash entry is arranged by perl on the fly, even with the use strict pragma in effect:
$ perl -e 'package nirvana; use strict; print *foo;'
*nirvana::foo
So the value of a typeglob is a full path into the perl stashes, down from the defstash. But what actually is *main::foo?
$ perl -e 'print "*foo is not interpolated"'
*foo is not interpolated
$ perl -e 'print "although ".*foo." could be a string"'
although *main::foo could be a string
$ perl -e 'print "*foo is \"*main::foo\"" if *foo eq "*main::foo"'
*foo is "*main::foo"
$ perl -e 'package nirvana; sub f { local *g=shift; print *g."=$g" }; package main; $foo=42; nirvana::f(*foo)'
*main::foo=42
Conclusion: with typeglobs you reach the bedrock of perl, where the spade bends back.
More on this perlguts, perlref, perldsc and perllol.
In C++ we use a flex/bison scanner/parser combination. Since each scalar and list is put into an Abstract Syntax Tree (AST), as separate object, we use a free store management that allows the allocation of huge amounts of tiny objects. We also use reference-counted smart-pointers, which allocate themselves on our fast free store.
So RAM will not be fragmented, and the allocation of RAM is significantly faster than with the default process heap.
For example, a 300 MB Rlist-file can be read from a C++ process which will not peak over 400-500 MB of process RAM.
There are no known bugs, this package is stable.
Deficiencies and TODOs:
The "deparse" functionality for the "code_refs" compile option has not
yet been implemented.
The "threads" compile option has not yet been implemented.
IEEE 754 notations of Infinite and NaN not yet implemented.
compile_Perl is experimental.
Data::Dumper. In contrast to the Data::Dumper, Data::Rlist scalars will be properly typed as number or string. Data::Dumper writes numbers always as quoted strings, for example
$VAR1 = {
'configuration' => {
'verbose' => 'Y',
'importance_sampling_loss_quantile' => '0.04',
'distribution_loss_unit' => '100',
'default_only' => 'Y',
'num_threads' => '5',
.
.
}
};
where Data::Rlist writes
{
configuration = {
verbose = Y;
importance_sampling_loss_quantile = 0.04;
distribution_loss_unit = 100;
default_only = Y;
num_threads = 5;
.
.
};
}
As one can see Data::Dumper writes the data right in Perl syntax, which means the dumped text can be simply eval'd, and the data can be restored very fast. Rlists are not quite Perl-syntax: a dedicated parser is required. But therefore Rlist text is portable and can be read from other programming languages, namely C++, where a fast flex/bison-parser in conjunction with a smart heap management is implemented. So C++ programs, like Perl programs, are able to handle Rlist files of several hundred MB.
With $Data::Dumper::Useqq enabled it was observed that Data::Dumper renders output significantly slower than compile. This is actually suprising, since Data::Rlist tests for each scalar whether it is numeric, and truely quotes/escapes strings. Data::Dumper quotes all scalars (including numbers), and it does not escape strings. This may also result in some odd behaviors. For example,
use Data::Dumper;
print Dumper "foo\n";
yields
$VAR1 = 'foo
';
while
use Data::Rlist;
PrintData "foo\n"
yields
{ "foo\n"; }
(Recall that parse always returns a list, as array- or hash-reference.)
Finally, Data::Rlist generates smaller files. With the default $Data::Dumper::Indent of 2 Data::Dumper's output is 4-5 times that of Data::Rlist's. This is because Data::Dumper recklessly uses blanks, instead of horizontal tabulators, which blows up file sizes without measure.
Copyright 1998-2007 Andreas Spindler
Maintained at CPAN (http://search.cpan.org/search) and the author's site (http://www.visualco.de). Please send mail to rlist@visualco.de.
This library is free software; you can redistribute it and/or modify it under the same terms as Perl itself, either Perl version 5.8.8 or, at your option, any later version of Perl 5 you may have available.
Contact the author for the C++ library at rlist@visualco.de.
Thank you for your attention.