path: root/tcllib/modules/docstrip/docstrip.man

[manpage_begin docstrip n 1.2]
[see_also docstrip_util]
[keywords .dtx]
[keywords docstrip]
[keywords documentation]
[keywords LaTeX]
[keywords {literate programming}]
[keywords source]
[copyright "2003\u20132010 Lars Hellstr\u00F6m\
  <Lars dot Hellstrom at residenset dot net>"]
[moddesc {Literate programming tool}]
[titledesc {Docstrip style source code extraction}]
[category  {Documentation tools}]
[require Tcl 8.4]
[require docstrip [opt 1.2]]
[vset emdash \u2014]
[description]

[syscmd Docstrip] is a tool created to support a brand of Literate
Programming. It is most common in the (La)TeX community, where it
is being used for pretty much everything from the LaTeX core and up,
but there is nothing about [syscmd docstrip] which prevents using it
for other types of software.
[para]

In short, the basic principle of literate programming is that program
source should primarily be written and structured to suit the
developers (and advanced users who want to peek "under the hood"), not
to suit the whims of a compiler or corresponding source code consumer.
This means literate sources often need some kind of "translation" to an
illiterate form that dumb software can understand.
The [package docstrip] Tcl package handles this translation.
[para]

Even for those who do not whole-hartedly subscribe to the philosophy
behind literate programming, [syscmd docstrip] can bring greater
clarity to in particular:
[list_begin itemized]
  [item] programs employing non-obvious mathematics
  [item] projects where separate pieces of code, perhaps in
    different languages, need to be closely coordinated.
[list_end]
The first is by providing access to much more powerful typographical
features for source code comments than are possible in plain text.
The second is because all the separate pieces of code can be kept
next to each other in the same source file.
[para]

The way it works is that the programmer edits directly only one or
several "master" source code files, from which [syscmd docstrip]
generates the more traditional "source" files compilers or the like
would expect. The master sources typically contain a large amount of
documentation of the code, sometimes even in places where the code
consumers would not allow any comments. The etymology of "docstrip"
is that this [emph doc]umentation was [emph strip]ped away (although
"code extraction" might be a better description, as it has always
been a matter of copying selected pieces of the master source rather
than deleting text from it).
The [package docstrip] Tcl package contains a reimplementation of
the basic extraction functionality from the [syscmd docstrip]
program, and thus makes it possible for a Tcl interpreter to read
and interpret the master source files directly.
[para]

Readers who are not previously familiar with [syscmd docstrip] but
want to know more about it may consult the following sources.
[list_begin enumerated]
[enum]
  [emph {The tclldoc package and class}],
  [uri {http://ctan.org/tex-archive/macros/latex/contrib/tclldoc/}].
[enum]
  [emph {The DocStrip utility}],
  [uri {http://ctan.org/tex-archive/macros/latex/base/docstrip.dtx}].
[enum]
  [emph {The doc and shortvrb Packages}],
  [uri {http://ctan.org/tex-archive/macros/latex/base/doc.dtx}].
[enum]
  Chapter 14 of
  [emph {The LaTeX Companion}] (second edition),
  Addison-Wesley, 2004; ISBN 0-201-36299-6.
[list_end]

[section {File format}]

The basic unit [syscmd docstrip] operates on are the [emph lines] of
a master source file. Extraction consists of selecting some of these
lines to be copied from input text to output text. The basic
distinction is that between [emph {code lines}] (which are copied and
do not begin with a percent character) and [emph {comment lines}]
(which begin with a percent character and are not copied).

[example {
   docstrip::extract [join {
     {% comment}
     {% more comment !"#$%&/(}
     {some command}
     { % blah $blah "Not a comment."}
     {% abc; this is comment}
     {# def; this is code}
     {ghi}
     {% jkl}
   } \n] {}
}]
returns the same sequence of lines as
[example {
   join {
     {some command}
     { % blah $blah "Not a comment."}
     {# def; this is code}
     {ghi} ""
   } \n
}]

It does not matter to [syscmd docstrip] what format is used for the
documentation in the comment lines, but in order to do better than
plain text comments, one typically uses some markup language. Most
commonly LaTeX is used, as that is a very established standard and
also provides the best support for mathematical formulae, but the
[package docstrip::util] package also gives some support for
[term doctools]-like markup.
[para]

Besides the basic code and comment lines, there are also
[emph {guard lines}], which begin with the two characters '%<', and
[emph {meta-comment lines}], which begin with the two characters
'%%'. Within guard lines there is furthermore the distinction between
[emph {verbatim guard lines}], which begin with '%<<', and ordinary
guard lines, where the '%<' is not followed by another '<'. The last
category is by far the most common.
[para]

Ordinary guard lines conditions extraction of the code line(s) they
guard by the value of a boolean expression; the guarded block of
code lines will only be included if the expression evaluates to true.
The syntax of an ordinary guard line is one of
[example {
    '%' '<' STARSLASH EXPRESSION '>'
    '%' '<' PLUSMINUS EXPRESSION '>' CODE
}]
where
[example {
    STARSLASH  ::=  '*' | '/'
    PLUSMINUS  ::=  | '+' | '-'
    EXPRESSION ::= SECONDARY | SECONDARY ',' EXPRESSION
                 | SECONDARY '|' EXPRESSION
    SECONDARY  ::= PRIMARY | PRIMARY '&' SECONDARY
    PRIMARY    ::= TERMINAL | '!' PRIMARY | '(' EXPRESSION ')'
    CODE       ::= { any character except end-of-line }
}]
Comma and vertical bar both denote 'or'. Ampersand denotes 'and'.
Exclamation mark denotes 'not'. A TERMINAL can be any nonempty string
of characters not containing '>', '&', '|', comma, '(', or ')',
although the [syscmd docstrip] manual is a bit restrictive and only
guarantees proper operation for strings of letters (although even
the LaTeX core sources make heavy use also of digits in TERMINALs).
The second argument of [cmd docstrip::extract] is the list of those
TERMINALs that should count as having the value 'true'; all other
TERMINALs count as being 'false' when guard expressions are evaluated.
[para]

In the case of a '%<*[emph EXPRESSION]>' guard, the lines guarded are
all lines up to the next '%</[emph EXPRESSION]>' guard with the same
[emph EXPRESSION] (compared as strings). The blocks of code delimited
by such '*' and '/' guard lines must be properly nested.
[example {
   set text [join {
      {begin}
      {%<*foo>}
      {1}
      {%<*bar>}
      {2}
      {%</bar>}
      {%<*!bar>}
      {3}
      {%</!bar>}
      {4}
      {%</foo>}
      {5}
      {%<*bar>}
      {6}
      {%</bar>}
      {end}
   } \n]
   set res [docstrip::extract $text foo]
   append res [docstrip::extract $text {foo bar}]
   append res [docstrip::extract $text bar]
}]
sets $res to the result of
[example {
   join {
      {begin}
      {1}
      {3}
      {4}
      {5}
      {end}
      {begin}
      {1}
      {2}
      {4}
      {5}
      {6}
      {end}
      {begin}
      {5}
      {6}
      {end} ""
   } \n
}]

In guard lines without a '*', '/', '+', or '-' modifier after the
'%<', the guard applies only to the CODE following the '>' on that
single line. A '+' modifier is equivalent to no modifier. A '-'
modifier is like the case with no modifier, but the expression is
implicitly negated, i.e., the CODE of a '%<-' guard line is only
included if the expression evaluates to false.
[para]

Metacomment lines are "comment lines which should not be stripped
away", but be extracted like code lines; these are sometimes used for
copyright notices and similar material. The '%%' prefix is however
not kept, but substituted by the current [option -metaprefix], which
is customarily set to some "comment until end of line" character (or
character sequence) of the language of the code being extracted.
[example {
   set text [join {
      {begin}
      {%<foo> foo}
      {%<+foo>plusfoo}
      {%<-foo>minusfoo}
      {middle}
      {%% some metacomment}
      {%<*foo>}
      {%%another metacomment}
      {%</foo>}
      {end}
   } \n]
   set res [docstrip::extract $text foo -metaprefix {# }]
   append res [docstrip::extract $text bar -metaprefix {#}]
}]
sets $res to the result of
[example {
   join {
      {begin}
      { foo}
      {plusfoo}
      {middle}
      {#  some metacomment}
      {# another metacomment}
      {end}
      {begin}
      {minusfoo}
      {middle}
      {# some metacomment}
      {end} ""
   } \n
}]

Verbatim guards can be used to force code line
interpretation of a block of lines even if some of them happen to look
like any other type of lines to docstrip. A verbatim guard has the
form '%<<[emph END-TAG]' and the verbatim block is terminated by the
first line that is exactly '%[emph END-TAG]'.
[example {
   set text [join {
      {begin}
      {%<*myblock>}
      {some stupid()}
      {   #computer<program>}
      {%<<QQQ-98765}
      {% These three lines are copied verbatim (including percents}
      {%% even if -metaprefix is something different than %%).}
      {%</myblock>}
      {%QQQ-98765}
      {   using*strange@programming<language>}
      {%</myblock>}
      {end}
   } \n]
   set res [docstrip::extract $text myblock -metaprefix {# }]
   append res [docstrip::extract $text {}]
}]
sets $res to the result of
[example {
   join {
      {begin}
      {some stupid()}
      {   #computer<program>}
      {% These three lines are copied verbatim (including percents}
      {%% even if -metaprefix is something different than %%).}
      {%</myblock>}
      {   using*strange@programming<language>}
      {end}
      {begin}
      {end} ""
   } \n
}]
The processing of verbatim guards takes place also inside blocks of
lines which due to some outer block guard will not be copied.
[para]

The final piece of [syscmd docstrip] syntax is that extraction
stops at a line that is exactly "\endinput"; this is often used to
avoid copying random whitespace at the end of a file. In the unlikely
case that one wants such a code line, one can protect it with a
verbatim guard.

[section Commands]

The package defines two commands.

[list_begin definitions]
[call [cmd docstrip::extract] [arg text] [arg terminals] [
   opt "[arg option] [arg value] ..."
]]
  The [cmd extract] command docstrips the [arg text] and returns the
  extracted lines of code, as a string with each line terminated with
  a newline. The [arg terminals] is the list of those guard
  expression terminals which should evaluate to true.
  The available options are:
  [list_begin options]
  [opt_def -annotate [arg lines]]
    Requests the specified number of lines of annotation to follow
    each extracted line in the result. Defaults to 0. Annotation lines
    are mostly useful when the extracted lines are to undergo some
    further transformation. A first annotation line is a list of three
    elements: line type, prefix removed in extraction, and prefix
    inserted in extraction. The line type is one of: 'V' (verbatim),
    'M' (metacomment), '+' (+ or no modifier guard line), '-' (-
    modifier guard line), '.' (normal line). A second annotation line
    is the source line number. A third annotation line is the current
    stack of block guards. Requesting more than three lines of
    annotation is currently not supported.
  [opt_def -metaprefix [arg string]]
    The string by which the '%%' prefix of a metacomment line will
    be replaced. Defaults to '%%'. For Tcl code this would typically
    be '#'.
  [opt_def -onerror [arg keyword]]
    Controls what will be done when a format error in the [arg text]
    being processed is detected. The settings are:
    [list_begin definitions]
    [def [const ignore]]
      Just ignore the error; continue as if nothing happened.
    [def [const puts]]
      Write an error message to [const stderr], then continue
      processing.
    [def [const throw]]
      Throw an error. The [option -errorcode] is set to a list whose
      first element is [const DOCSTRIP], second element is the
      type of error, and third element is the line number where
      the error is detected. This is the default.
    [list_end]
  [opt_def -trimlines [arg boolean]]
    Controls whether [emph spaces] at the end of a line should be
    trimmed away before the line is processed. Defaults to true.
  [list_end]

  It should be remarked that the [arg terminals] are often called
  "options" in the context of the [syscmd docstrip] program, since
  these specify which optional code fragments should be included.

[call [cmd docstrip::sourcefrom] [arg filename] [arg terminals] [
   opt "[arg option] [arg value] ..."
]]
  The [cmd sourcefrom] command is a docstripping emulation of
  [cmd source]. It opens the file [arg filename], reads it, closes it,
  docstrips the contents as specified by the [arg terminals], and
  evaluates the result in the local context of the caller, during
  which time the [cmd info] [method script] value will be the
  [arg filename]. The options are passed on to [cmd fconfigure] to
  configure the file before its contents are read. The
  [option -metaprefix] is set to '#', all other [cmd extract]
  options have their default values.
[list_end]


[section {Document structure}]

The file format (as described above) determines whether a master
source code file can be processed correctly by [syscmd docstrip],
but the usefulness of the format is to no little part also dependent
on that the code and comment lines together constitute a well-formed
document.
[para]

For a document format that does not require any non-Tcl software, see
the [cmd ddt2man] command in the [package docstrip::util] package. It
is suggested that files employing that document format are given the
suffix [file .ddt], to distinguish them from the more traditional
LaTeX-based [file .dtx] files.
[para]

Master source files with [file .dtx] extension are usually set up so
that they can be typeset directly by [syscmd latex] without any
support from other files. This is achieved by beginning the file
with the lines
[example_begin]
   % \iffalse
   %<*driver>
   \documentclass{tclldoc}
   \begin{document}
   \DocInput{[emph filename.dtx]}
   \end{document}
   %</driver>
   % \fi
[example_end]
or some variation thereof. The trick is that the file gets read twice.
With normal LaTeX reading rules, the first two lines are comments and
therefore ignored. The third line is the document preamble, the fourth
line begins the document body, and the sixth line ends the document,
so LaTeX stops there [vset emdash] non-comments below that point in
the file are never subjected to the normal LaTeX reading rules. Before
that, however, the \DocInput command on the fifth line is processed,
and that does two things: it changes the interpretation of '%' from
"comment" to "ignored", and it inputs the file specified in the
argument (which is normally the name of the file the command is in).
It is this second time that the file is being read that the comments
and code in it are typeset.
[para]

The function of the \iffalse ... \fi is to skip lines two to seven
on this second time through; this is similar to the "if 0 { ... }"
idiom for block comments in Tcl code, and it is needed here because
(amongst other things) the \documentclass command may only be
executed once. The function of the <driver> guards is to prevent this
short piece of LaTeX code from being extracted by [syscmd docstrip].
The total effect is that the file can function both as a LaTeX
document and as a [syscmd docstrip] master source code file.
[para]

It is not necessary to use the tclldoc document class, but that does
provide a number of features that are convenient for [file .dtx]
files containing Tcl code. More information on this matter can be
found in the references above.

[manpage_end]