docstrip_util - Docstrip-related utilities
The docstrip::util package is meant for collecting various utility procedures that are mainly useful at installation or development time. It is separate from the base package to avoid overhead when the latter is used to source code.
Like raw ".tcl" files, code lines in docstrip source files can be searched for package declarations and corresponding indices constructed. A complication is however that one cannot tell from the code blocks themselves which will fit together to make a working package; normally that information would be found in an accompanying ".ins" file, but parsing one of those is not an easy task. Therefore docstrip::util introduces an alternative encoding of such information, in the form of a declarative Tcl script: the catalogue (of the contents in a source file).
The special commands which are available inside a catalogue are:
Declares that the code for a package with name name and version version is made up from those modules in the source file which are selected by the terminals list of guard expression terminals. This code should preferably not contain a package provide command for the package, as one will be provided by the package loading mechanisms.
Declares that the code for a package is made up from those modules in the source file which are selected by the listed guard expression terminals. The name and version of this package is determined from package provide command(s) found in that code (hence there must be such a command in there).
Declares the fconfigure options that should be in force when reading the source; this can usually be ignored for pure ASCII files, but if the file needs to be interpreted according to some other -encoding then this is how to specify it. The command should normally appear first in the catalogue, as it takes effect only for commands following it.
Other Tcl commands are supported too — a catalogue is parsed by being evaluated in a safe interpreter — but they are rarely needed. To allow for future extensions, unknown commands in the catalogue are silently ignored.
To simplify distribution of catalogues together with their source files, the catalogue is stored in the source file itself as a module selected by the terminal 'docstrip.tcl::catalogue'. This supports both the style of collecting all catalogue lines in one place and the style of putting each catalogue line in close proximity of the code that it declares.
Putting catalogue entries next to the code they declare may look as follows
% First there's the catalogue entry
% \begin{tcl}
%<docstrip.tcl::catalogue>pkgProvide foo::bar 1.0 {foobar load}
% \end{tcl}
% second a metacomment used to include a copyright message
% \begin{macrocode}
%<*foobar>
%% This file is placed in the public domain.
% \end{macrocode}
% third the package implementation
% \begin{tcl}
namespace eval foo::bar {
# ... some clever piece of Tcl code elided ...
% \end{tcl}
% which at some point may have variant code to make use of a
% |load|able extension
% \begin{tcl}
%<*load>
load [file rootname [info script]][info sharedlibextension]
%</load>
%<*!load>
# ... even more clever scripted counterpart of the extension
# also elided ...
%</!load>
}
%</foobar>
% \end{tcl}
% and that's it!
The corresponding set-up with pkgIndex would be
% First there's the catalogue entry
% \begin{tcl}
%<docstrip.tcl::catalogue>pkgIndex foobar load
% \end{tcl}
% second a metacomment used to include a copyright message
% \begin{tcl}
%<*foobar>
%% This file is placed in the public domain.
% \end{tcl}
% third the package implementation
% \begin{tcl}
package provide foo::bar 1.0
namespace eval foo::bar {
# ... some clever piece of Tcl code elided ...
% \end{tcl}
% which at some point may have variant code to make use of a
% |load|able extension
% \begin{tcl}
%<*load>
load [file rootname [info script]][info sharedlibextension]
%</load>
%<*!load>
# ... even more clever scripted counterpart of the extension
# also elided ...
%</!load>
}
%</foobar>
% \end{tcl}
% and that's it!
This command is a sibling of the standard pkg_mkIndex command, in that it adds package entries to "pkgIndex.tcl" files. The difference is that it indexes docstrip-style source files rather than raw ".tcl" or loadable library files. Only packages listed in the catalogue of a file are considered.
The dir argument is the directory in which to look for files (and whose "pkgIndex.tcl" file should be amended). The pattern argument is a glob pattern of files to look into; a typical value would be *.dtx or *.{dtx,ddt}. Remaining arguments are option-value pairs, where the supported options are:
The package ifneeded scripts that are generated contain one package require docstrip command and one docstrip::sourcefrom command. If the catalogue entry was of the pkgProvide kind then the package ifneeded script also contains the package provide command.
Note that index_from_catalogue never removes anything from an existing "pkgIndex.tcl" file. Hence you may need to delete it (or have pkg_mkIndex recreate it from scratch) before running index_from_catalogue to update some piece of information, such as a package version number.
This command is an alternative to index_from_catalogue which creates Tcl Module (".tm") files rather than "pkgIndex.tcl" entries. Since this action is more similar to what docstrip classically does, it has features for putting pre- and postambles on the generated files.
The source argument is the name of the source file to generate ".tm" files from. The target argument is the directory which should count as a module path, i.e., this is what the relative paths derived from package names are joined to. The supported options are:
An existing file of the same name as one to be created will be overwritten.
This command returns a preamble in the classical docstrip style
## ## This is `TARGET', ## generated by the docstrip::util package. ## ## The original source files were: ## ## SOURCE (with options: `foo,bar') ## ## Some message line 1 ## line2 ## line3
if called as
docstrip::util::classical_preamble {##}\
"\nSome message line 1\nline2\nline3" TARGET SOURCE {foo bar}
The command supports preambles for files generated from multiple sources, even though modules_from_catalogue at present does not need that.
This command returns a postamble in the classical docstrip style
## Some message line 1 ## line2 ## line3 ## ## End of file `TARGET'.
if called as
docstrip::util::classical_postamble {##}\
"Some message line 1\nline2\nline3" TARGET SOURCE {foo bar}
In other words, the source and terminals arguments are ignored, but supported for symmetry with classical_preamble.
This command returns a list where every even index element is the name of a package provided by text when that is evaluated as a Tcl script, and the following odd index element is the corresponding version. It is used to do package indexing of extracted pieces of code, in the manner of pkg_mkIndex.
One difference to pkg_mkIndex is that the text gets evaluated in a safe interpreter. package require commands are silently ignored, as are unknown commands (which includes source and load). Other errors cause processing of the text to stop, in which case only those package declarations that had been encountered before the error will be included in the return value.
The setup-script argument can be used to customise the evaluation environment, if the code in text has some very special needs. The setup-script is evaluated in the local context of the packages_provided procedure just before the text is processed. At that time, the name of the slave command for the safe interpreter that will do this processing is kept in the local variable c. To for example copy the contents of the ::env array to the safe interpreter, one might use a setup-script of
$c eval [list array set env [array get ::env]]
Unlike the previous group of commands, which would use docstrip::extract to extract some code lines and then process those further, the following commands operate on text consisting of all types of lines.
The ddt2man command reformats text from the general docstrip format to doctools ".man" format (Tcl Markup Language for Manpages). The different line types are treated as follows:
The '%' and '%%' prefixes are removed, the rest of the text is kept as it is.
These are kept as they are. (Effectively this means that they will count as comment lines after a comment line and as code lines after a code line.)
example_begin and example_end commands are placed at the beginning and end of every block of consecutive code lines. Brackets in a code line are converted to lb and rb commands.
These are processed as usual, so they do not show up in the result but every line in a verbatim block is treated as a code line.
These are treated as code lines, except that the actual guard is emphasised.
At the time of writing, no project has employed doctools markup in master source files, so experience of what works well is not available. A source file could however look as follows
% [manpage_begin gcd n 1.0]
[see_also docstrip]
[see_also doctools]
[see_also doctools_fmt]
[keywords .ddt]
[keywords catalogue]
[keywords diff]
[keywords docstrip]
[keywords doctools]
[keywords documentation]
[keywords {literate programming}]
[keywords module]
[keywords {package indexing}]
[keywords patch]
[keywords source]
[keywords {Tcl module}]
% [moddesc {Greatest Common Divisor}]
% [require gcd [opt 1.0]]
% [description]
%
% [list_begin definitions]
% [call [cmd gcd] [arg a] [arg b]]
% The [cmd gcd] procedure takes two arguments [arg a] and [arg b] which
% must be integers and returns their greatest common divisor.
proc gcd {a b} {
% The first step is to take the absolute values of the arguments.
% This relieves us of having to worry about how signs will be treated
% by the remainder operation.
set a [expr {abs($a)}]
set b [expr {abs($b)}]
% The next line does all of Euclid's algorithm! We can make do
% without a temporary variable, since $a is substituted before the
% [lb]set a $b[rb] and thus continues to hold a reference to the
% "old" value of [var a].
while {$b>0} { set b [expr { $a % [set a $b] }] }
% In Tcl 8.3 we might want to use [cmd set] instead of [cmd return]
% to get the slight advantage of byte-compilation.
%<tcl83> set a
%<!tcl83> return $a
}
% [list_end]
%
% [manpage_end]
If the above text is fed through docstrip::util::ddt2man then the result will be a syntactically correct doctools manpage, even though its purpose is a bit different.
It is suggested that master source code files with doctools markup are given the suffix ".ddt", hence the "ddt" in ddt2man.
The guards command returns information (mostly of a statistical nature) about the ordinary docstrip guards that occur in the text. The subcmd selects what is returned.
List the guard expression terminals with counts. The format of the return value is a dictionary which maps the terminal name to the number of occurencies of it in the file.
List the guard expressions with counts. The format of the return value is a dictionary which maps the expression to the number of occurencies of it in the file.
List the syntactically incorrect guard expressions (e.g. parentheses do not match, or a terminal is missing). The return value is a list, with the elements in no particular order.
List the guard expressions. The return value is a list, with the elements in no particular order.
List the guard expressions with modifiers. The format of the return value is a dictionary where each index is a guard expression and each entry is a string with one character for every guard line that has this expression. The characters in the entry specify what modifier was used in that line: +, -, *, /, or (for guard without modifier:) space. This is the most primitive form of the information gathered by guards.
List the guard expression terminals. The return value is a list, with the elements in no particular order.
List the malformed guard lines (this does not include lines where only the expression is malformed, though). The format of the return value is a dictionary which maps line numbers to their contents.
This command tries to apply a diff file (for example a contributed patch) that was computed for a generated file to the docstrip source. This can be useful if someone has edited a generated file, thus mistaking it for being the source. This command makes no presumptions which are specific for the case that the generated file is a Tcl script.
patch requires that the source file to patch is kept as a list of lines in a variable, and the name of that variable in the calling context is what goes into the source-var argument. The terminals is the list of terminals used to extract the file that has been patched. The diff is the actual diff to apply (in a format as explained below) and the fromtext is the contents of the file which served as "from" when the diff was computed. Options can be used to further control the process.
The process works by "lifting" the hunks in the diff from generated to source file, and then applying them to the elements of the source-var. In order to do this lifting, it is necessary to determine how lines in the fromtext correspond to elements of the source-var, and that is where the terminals come in; the source is first extracted under the given terminals, and the result of that is then matched against the fromtext. This produces a map which translates line numbers stated in the diff to element numbers in source-var, which is what is needed to lift the hunks.
The reason that both the terminals and the fromtext must be given is twofold. First, it is very difficult to keep track of how many lines of preamble are supplied some other way than by copying lines from source files. Second, a generated file might contain material from several source files. Both make it impossible to predict what line number an extracted file would have in the generated file, so instead the algorithm for computing the line number map looks for a block of lines in the fromtext which matches what can be extracted from the source. This matching is affected by the following options:
The return value is in the form of a unified diff, containing only those hunks which were not applied or were only partially applied; a comment in the header of each hunk specifies which case is at hand. It is normally necessary to manually review both the return value from patch and the patched text itself, as this command cannot adjust comment lines to match new content.
An example use would look like
set sourceL [split [docstrip::util::thefile from.dtx] \n]
set terminals {foo bar baz}
set fromtext [docstrip::util::thefile from.tcl]
set difftext [exec diff --unified from.tcl to.tcl]
set leftover [docstrip::util::patch sourceL $terminals $fromtext\
[docstrip::util::import_unidiff $difftext] -metaprefix {#}]
set F [open to.dtx w]; puts $F [join $sourceL \n]; close $F
return $leftover
Here, "from.dtx" was used as source for "from.tcl", which someone modified into "to.tcl". We're trying to construct a "to.dtx" which can be used as source for "to.tcl".
The thefile command opens the file filename, reads it to end, closes it, and returns the contents (dropping a final newline if there is one). The option-value pairs are passed on to fconfigure to configure the open file channel before anything is read from it.
This command parses a unified (diff flags -U and --unified) format diff into the list-of-hunks format expected by docstrip::util::patch. The diff-text argument is the text to parse and the warning-var is, if specified, the name in the calling context of a variable to which any warnings about parsing problems will be appended.
The return value is a list of hunks. Each hunk is a list of five elements "start1 end1 start2 end2 lines". start1 and end1 are line numbers in the "from" file of the first and last respectively lines of the hunk. start2 and end2 are the corresponding line numbers in the "to" file. Line numbers start at 1. The lines is a list with two elements for each line in the hunk; the first specifies the type of a line and the second is the actual line contents. The type is - for lines only in the "from" file, + for lines that are only in the "to" file, and 0 for lines that are in both.
.ddt, Tcl module, catalogue, diff, docstrip, doctools, documentation, literate programming, module, package indexing, patch, source
Documentation tools
Copyright © 2003–2010 Lars Hellström <Lars dot Hellstrom at residenset dot net>