.ds CP 2003-2004
.ds TC \'tconfpy\'
.TH TCONFPY 3 "TundraWare Inc."
.SH NAME tconfpy.py
Configuration File Support For Python Applications
.SH SYNOPSIS
It is common to provide an external "configuration file" when writing
sophisticated applications. This gives the end-user the ability to
easily change program options by editing that file.
\*(TC is a Python module for parsing such configuration files. \*(TC
understands and parses a configuration "language" which has a rich set
of string-substitution, variable name, conditional, and validation
features.
By using \*(TC, you unburden your program from the major responsibility
of configuration file parsing and validation, while providing your
users a rich set of configuration features.
.SH DOCUMENT ORGANIZATION
This document is divided into 4 major sections:
.B PROGRAMMING USING THE \*(TC API
discusses how to call the configuration file parser, the options
available when doing this, and what the parser returns. This is
the "Programmer's View" of the module and provides in-depth
descriptions of the API, data structures, and options available to
the programmer.
.B CONFIGURATION LANGUAGE REFERENCE
describes the syntax and semantics of the configuration language
recognized by \*(TC. This is the "User's View" of the package, but
both programmers and people writing configuration files will find this
helpful.
.B ADVANCED TOPICS
describes some ways to combine the various \*(TC features to
do some fairly nifty things.
.B INSTALLATION
explains how to install this package on various platforms. This
information can also be found in the \'READ-1ST.txt\' file distributed
with the package.
.SH PROGRAMMING USING THE \*(TC API
\*(TC is a Python module and thus available for use by any Python
program. This section discusses how to invoke the \*(TC parser, the
options available when doing so, and what the parser returns to
the calling program.
One small note is in order here. As a matter of coding style and
brevity, the code examples here assume the following Python import
syntax:
.nf
from tconfpy import *
.fi
If you prefer the more pedestrian:
.nf
import tconfpy
.fi
you will have to prepend all references to a \*(TC object with
\'tconfpy.\'. So \'retval=ParseConfig(...\' becomes
\'retval = tconfpy.ParseConfig(...\' and so on.
You will also find the test driver code provided in the \*(TC package
helpful as you read through the following sections. \'test-tc.py\' is
a utility to help you learn and exercise the \*(TC API. Perusing the
code therein is helpful as an example of the topics discussed below.
.SS API Overview
The \*(TC API consists of a single call. Only the configuration file
to be processed is a required parameter, all the others are optional
and default as described below:
.nf
from tconfpy import *
retval = ParseConfig(cfgfile,
InitialSymTbl={},
AllowNewVars=True,
AllowNewNamespaces=True,
Debug=False,
LiteralVars=False
)
where:
.fi
.TP
.B cfgfile (Required Parameter - No Default)
The the name of a file containing configuration information
.TP
.B InitialSymTbl (Default: {})
A pre-populated symbol table (a Python dictionary). As described
below, this must contain valid \'VarDescriptor\' entries for each
symbol in the table.
.TP
.B AllowNewVars (Default: True)
Allow the user to create new variables in the configuration file.
.TP
.B AllowNewNamespaces (Default: True)
Allow new namespaces to be created in the configuration file.
.TP
.B Debug (Default: False)
If set to True, \*(TC will provide detailed debugging information
about each line processed when it returns.
.TP
.B LiteralVars (Default: False)
If set to True this option enables variable substitutions within
\'.literal\' blocks of a configuration file. See the section in the
language reference below on \'.literal\' usage for details.
.TP
.B retval
An object of type \'tconfpy.RetObj\' used to return parsing results.
.SS Reasons For Passing An Initial Symbol Table
The simplest way to parse a configuration file is just to call
the parser with the name of that file:
.nf
retval = ParseConfig("myconfigfile")
.fi
Assuming your configuration file is valid, \'ParseConfig()\' will
return a symbol table populated with all the variables defined in the
file and their associated values. This symbol table will have
.B only
the symbols defined in that file (plus a few built-in and pre-defined
symbols needed internally by \*(TC).
However, the API provides a way for you to pass a "primed" symbol
table to the parser that contains pre-defined symbols/values of
your own choosing. Why on earth would you want to do this? There
are a number of reasons:
.nf
1) You may wish to write a configuration file which somehow depends
on a pre-defined variable that only the calling program can know:
.if [APPVERSION] == 1.0
# Set configuration for older application releases
.else
# Set configuration for newer releases
.endif
In this example, only the calling application can know its own
version, so it sets the variable APPVERSION in a symbol table
which is passed to \'ParseConfig()\'.
2) You may wish to "protect" certain variable names be creating
them ahead of time and marking them as "Read Only". This is
useful when you want a variable to be available for use
within a configuration file, but you do not want users to
be able to change its value. In this case, the variable can
be referenced in a string substitution or conditional test,
but cannot be changed.
3) You may want to place limits on what values can be assigned to
a particular variable. When a variable is newly defined in a
a configuration file, it just defaults to being a string
variable without any limits on its length or content. But
variables that are created by a program have access to the
variable's "descriptor". By setting various attribues
of the variable descriptor you can control variable type,
content, and range of values. In other words, you can have
\*(TC "validate" what values the user assigns to particular
variables. This substantially simplifies your application because
no invalid variable value will ever be returned from the parser.
.fi
.SS How To Create An Initial Symbol Table
A \*(TC "Symbol Table" is really nothing more than a Python
dictionary. The key for each dictionary entry is the variable's name
and the value is a \*(TC-specific object called a "variable
descriptor". Creating new variables in the symbol table involves
nothing more than this:
.nf
from tconfpy import *
# Create an empty symbol table
MySymTable = {}
# Create descriptor for new variable
MyVarDes = VarDescriptor()
# Code to fiddle with descriptor contents goes here
MyVarDes.Value = "MyVal"
# Now load the variable into the symbol table
MySymTable["MyVariableName"] = MyVarDes
.fi
The heart of this whole business the \'VarDescriptor\' object. It
"describes" the value and properties of a variable. These
descriptor objects have the following attributes and defaults:
.nf
VarDescriptor.Value = ""
VarDescriptor.Writeable = True
VarDescriptor.Type = TYPE_STRING
VarDescriptor.Default = ""
VarDescriptor.LegalVals = []
VarDescriptor.Min = None
VarDescriptor.Max = None
.fi
When \*(TC encounters a new variable in a configuration file, it just
instantiates one of these descriptor objects with these defaults for
that variable. That is, variables newly-defined in a configuration
file are entered into the symbol table as string types, with an
initial value of "" and with no restriction on content or length.
But, when you create variables under program control to "prime" an
initial symbol table, you can modify the content of any of these
attributes for each variable. These descriptor attributes are what
\*(TC uses to validate subsequent attempts to change the variable's
value in the configuration file. In other words, modifying a
variable's descriptor tells \*(TC just what you'll accept as "legal"
values for that variable.
Each attribute has a specific role:
.TP
.B VarDescriptor.Value (Default: Empty String)
Holds the current value for the variable.
.TP
.B VarDescriptor.Writeable (Default: True)
Sets whether or not the user can change the variable's value. Setting
this attribute to False makes the variable
.B Read Only.
.TP
.B VarDescriptor.Type (Default: TYPE_STRING)
One of TYPE_BOOL, TYPE_COMPLEX, TYPE_FLOAT, TYPE_INT, or TYPE_STRING.
This defines the type of the variable. Each time \*(TC sees
a value being assigned to a variable in the configuration file, it
checks to see if that variable already exists in the symbol table.
If it does, the parser checks the value being assigned and makes sure
it matches the type declared for that variable. For example,
suppose you did this when defining the variable, \'foo\':
.nf
VarDescriptor.Type = TYPE_INT
.fi
Now suppose the user puts this in the configuration file:
.nf
foo = bar
.fi
This will cause a type mismatch error because \'bar\' cannot be coerced
into an integer type - it is a string.
As a general matter, for existing variables, \*(TC attempts to coerce
the right-hand-side of an assignment to the type declared for that
variable. The least fussy operation here is when the variable is
defined as TYPE_STRING because pretty much everything can be coerced
into a string. For example, here is how \'foo = 3+8j\' is treated for
different type declarations:
.nf
VarDescriptor.Type VarDescriptor.Value
------------------ -------------------
TYPE_BOOL Type Error
TYPE_COMPLEX 3+8j (A complex number)
TYPE_FLOAT Type Error
TYPE_INT Type Error
TYPE_STRING \'3+8j\' (A string)
.fi
This is why the default type for newly-defined variables in
the configuration file is TYPE_STRING: they can accept
pretty much
.B any
value.
.TP
.B VarDescriptor.Default (Default: Empty String)
This is a place to store the default value for a given variable. When
a variable is newly-defined in a configuration file, \*(TC places the
first value assigned to that variable into this attribute. For
variables already in the symbol table, \*TC does nothing to this
attribute. This attribute is not actually used by \*(TC for anything.
It is provided as a convenience so that the calling program can easily
"reset" every variable to its default value if desired.
.TP
.B VarDescriptor.LegalVals (Default: [])
Sometimes you want to limit a variable to a specific set of values.
That's what this attribute is for. \'LegalVals\' explictly lists
every legal value for the variable in question. If the list is
empty,then this validation check is skipped.
.B IMPORTANT:
If you change the content of \'LegalVals\', make sure it is always a
Python list. \*(TC's validation logic presumes this attribute
to be a list and will blow up nicely if it is not.
The exact semantics of LegalVals varies depending on
the type of the variable.
.nf
Variable Type What LegalVals Does
------------- -------------------
Boolean Nothing - Ignored
Integer, Float, Complex List of numeric values the
user can assign to this variable
Examples: [1, 2, 34]
[3.14, 2.73, 6.023e23]
[3.8-4j, 5+8j]
String List of Python regular expressions.
User must assign a value to this
variable that matches at least
one of these regular expressions.
Example: [r'a+.*', r'^AnExactString$']
.fi
The general semantic here is "If Legal Vals is not an empty list, the
user must assign a value that matches one of the items in LegalVals."
One special note applies to \'LegalVals\' for string variables. \*(TC
always assumes that this list contains Python regular expressions.
For validation, it grabs each entry in the list, attempts to compile
it as a regex, and checks to see if the value the user wants to set
matches. If you define an illegal regular expression here, \*(TC will
catch it and produce an appropriate error.
.TP
.B VarDescriptor.Min and VarDescriptor.Max (Default: None)
These set the minimum and maxium legal values for the variables,
but the semantics vary by variable type:
.nf
Variable Type What Min/Max Do
------------- ---------------
Boolean, Complex Nothing - Ignored
Integer, Float Set Minimum/Maxium allowed values.
String Set Minimum/Maximum string length
.fi
In all cases, if you want these tests skipped, set \'Min\' or \'Max\'
to the Python None.
.P
All these various validations are logically "ANDed" together.
i.e., A new value for a variable must be allowed
AND of the appropriate type AND one of the legal values AND
within the min/max range.
\*(TC makes no attempt to harmonize these validation
conditions with each other. If you specify a value in
\'LegalVals\' that is, say, lower than allowed by
\'Min\' you will always get an error when the user sets
the variable to that value: It passed the \'LegalVals\'
validation but failed it for \'Min\'.
One last note here concerns Boolean variables. Booleans
are actually stored in the symbol table as True or False.
However, \*(TC accepts user statements in a number of
formats to do this:
.nf
Boolean True Boolean False
------------ -------------
foo = 1 foo = 0
foo = True foo = False
foo = Yes foo = No
foo = On foo = Off
.fi
This is the one case where \*(TC is insensitive to case -
"tRUE", "TRUE", and "true" are all accepted, for example.
.SS How The \*(TC Parser Validates The Initial Symbol Table
When you pass an initial symbol table to the parser, \*(TC does some
basic validation that the table contents properly conform to the
\'VarDescriptor\' format and generates error messages if it finds
problems. However, the program does
.B not
check your specifications to see if they make sense. For instance
if you define an integer with a minimum value of 100 and a maximum
value of 50, \*(TC cheerfully accepts these limits even they they
are impossible. You'll just be unable to do anything with that
variable - any attempt to change its value will cause an error
to be recorded. Similarly, if you put a value in \'LegalVals\' that
is outside the range of \'Min\' to \'Max\', \*(TC will accept
it quietly.
.SS The \'AllowNewVars\' Option
By default, \*(TC lets the user define any new variables they
wish in a configuration file, merely by placing a line in the
file in this form:
.nf
Varname = Value
.fi
However, you can disable this capability by calling the parser like
this:
.nf
retval = ParseConfig("myconfigfile", AllowNewVars=False)
.fi
This means that the configuration file can "reference"
any pre-defined variables, and even change their values
(if they are not Read-Only), but it cannot create
.B new
variables.
This feature is primarily intended for use when you pass an initial
symbol table to the parser and you do not want any other variables
defined by the user. Why? There are several possible uses for
this option:
.nf
1) You know every configuration variable name the calling program
will use ahead of time. Disabling new variable names keeps
the configuration file from getting cluttered with variables
that the calling program will ignore anyway, thereby
keeping the file more readable.
2) You want to insulate your user from silent errors caused
by misspellings. Say your program looks for a configuration
variable called \'MyEmail\' but the user enters something
like \'myemail = foo@bar.com\'. \'MyEmail\' and \'myemail\'
are entirely different variables and only the former is
recognized by your calling program. By turning off new
variable creation, the user's inadvertent misspelling of
the desired variable name will be flagged as an error.
.fi
Note, however, that there is one big drawback to disabling new
variable creation. \*(TC processes the configuration file on a
line-by-line basis. No line "continuation" is supported. For really
long variable values and ease of maintenance, it is sometimes helpful
to create "intermediate" variables what hold temporary values used to
construct a variable actually needed by the calling program. For
example:
.nf
inter1 = Really, really, really, really, long argument #1
inter2 = Really, really, really, really, long argument #2
realvar = command [inter1] [inter2]
.fi
If you disable new variable creation you can't do this
anymore unless all the variables \'inter1\', \'inter2\',
and \'realvar\' are predefined in the initial symbol
table passed to the parser.
.SS The \'AllowNewNamespaces\' Option
By default, \*(TC supports the use of an arbitrary number of
lexical namespaces. They can be predefined in an initial
symbol table passed to the parser and/or created in the configuration
file as desired. (The details are described in a later section
of this document.)
There may be times, however, when you do not want users creating new
namespaces on their own. The reasons are much the same as for
preventing the creation of new variables in the option above:
Maintaining simplicity and clarity in the configuration file and
preventing "silent" errors due to misspellings.
In this case, call the API with \'AllowNewNamespaces=False\' and the
creation of new namespaces in the configuration file will be disabled.
Any attempt to create a new namespaces via either the
\'[new-ns-name]\' or \'NAMESPACE=new-ns-name\' methods will cause a
parse error to be generated.
It is important to realize that this option only disables the
creation of
.B new
namespaces. As discussed in the later section on namespace
processing, it is possible to pass an initial symbol table to the
parser which has one or more pre-defined namespaces in it. Each of
these pre-defined namespaces is available for use throughout the
configuration file even if \'AllowNewNamespaces\' is set to False.
.SS The \'Debug\' Option
\*(TC has a fairly rich set of debugging features built into its
parser. It can provide some detail about each line parsed as well
as overall information about the parse. Be default, debugging is
turned off. To enable debugging, merely set \'Debug=True' in the
API call:
.nf
retval = ParseConfig("myconfigfile", Debug=True)
.fi
.SS The \'LiteralVars\' Option
\*(TC supports the inclusion of literal text anywhere in a
configuration file via the \'.literal\' directive. This
directive effectively tells the \*(TC parser to pass
every line it encounters "literally" until it sees a
corresponding \'.endlinteral\' directive. By default,
\*(TC does
.B exactly
this. However, \*(TC has very powerful variable substitution
mechanisms. You may want to embed variable references in a
"literal" block and have them replaced by \*(TC.
Here is an example:
.nf
MyEmail = me@here.com # This defines variable MyEmail
.literal
printf("[MyEmail]"); /* A C Statement */
.endliteral
.fi
By default, \'ParseConfig()\' will leave everything within
the \'.literal\'/\'.endliteral\' block unchanged. In our
example, the string:
.nf
printf("[MyEmail]"); /* A C Statement */
.fi
would be in the list of literals returned by \'ParseConfig()\'.
However, we can ask \*(TC to do variable replacement
.B within
literal blocks by setting \'LiteralVars=True\' in the
\'ParseConfig()\' call:
.nf
retval = ParseConfig("myconfigfile", LiteralVars=True)
.fi
In this example, \*(TC would return:
.nf
printf("me@here.com"); /* A C Statement */
.fi
At first glance this seems only mildly useful, but it is
actually very handy. As described later in this document,
\*(TC has a rich set of conditional operators and string
sustitution facilities. You can use these features along
with literal block processing and variable substitution
within those blocks. This effectively lets you use
\*(TC as a preprocessor for
.B any
other language or text.
.SS How \*(TC Processes Errors
As a general matter, when \*(TC encounters an error in the
configuration file currently being parsed, it does two things. First,
it adds a descriptive error message into the list of errors returned
to the calling program (see the next section). Secondly, in many
cases, noteably during conditional processing, it sets the parser
state so the block in which the error occurred is logically False.
This does not happen in every case, however. If you are having
problems with errors, enable the Debugging features of the package and
look at the debug output. It provides detailed information about what
caused the error and why.
.SS \*(TC Return Values
When \*(TC is finished processing the configuration file it returns an
object which contains the entire results of the parse. This includes
a symbol table, any relevant error or warning messages, debug information
(if you requested this via the API), and any "literal" lines encountred
in the configuration.
The return object is an instance of the class \'twander.RetObj\' which is
nothing more than a container to hold return data. In the simplest
case, we can parse and extract results like this:
.nf
from tconfpy import *
retval = ParseConfig("myconfigfile", Debug=True)
.fi
\'retval\' now contains the results of the parse:
.nf
retval.Errors - A Python list containing error messages. If this list
is empty, you can infer that there were no parsing
errors - i.e., The configuration file was OK.
retval.Warnings - A Python list containing warning messages. These
describe minor problems not fatal to the parse
process, but that you really ought to clean up in
the configuration file.
retval.SymTable - A Python dictionary which lists all the defined
symbols and their associated values. A "value"
in this case is always an object of type
tconfpy.VarDescriptor (as described above).
retval.Literals - As described below, the \*(TC configuration language
supports a \'.literal\' directive. This directive
allows the user to embed literal text anywhere in
the configuration file. This effectively makes
\*(TC useful as a preprocessor for any other
language or text. retval.Literals is a Python list
containing all literal text discovered during the parse.
retval.Debug - A Python list containing detailed debug information for
each line parsed as well as some brief summary
information about the parse. retval.Debug defaults
to an empty list and is only populated if you set
\'Debug=True\' in the API call that initiated the
parse (as in the example above).
.fi
.SH CONFIGURATION LANGUAGE REFERENCE
\*(TC recognizes a full-featured configuration language that includes
variable creation and value assignment, a full preprocessor with
conditionals, type and value enforcement, and lexical namespaces.
This section of the document describes that language and provides
examples of how each feature can be used.
.SS General Rules For The \*(TC Configuration Language
.SS Variables And Variable References
.SS Predefined Variables
.SS The \'.include\' Directive
.SS Conditional Directives
.SS \'.literal\. - Using \*(TC As A Preprocessor For Other Languages
.SS Type And Value Enforcement
.SS Lexical Namespaces
.SH ADVANCED TOPICS
.SS Guaranteeing A Correct Base Configuration
.SS Enforcing Mandatory Configurations
.SS Iterative Parsing
.SH INSTALLATION
There are three ways to install \*(TC depending on your preferences
and type of system. In each of these installation methods you must be
logged in with root authority on Unix-like systems or as the
Administrator on Win32 systems.
.SS Preparation - Getting And Extracting The Package
For the first two installation methods, you must first download the
latest release from:
.nf
http://www.tundraware.com/Software/tconfpy/
.fi
Then unpack the contents by issuing the following command:
.nf
tar -xzvf py-tconfpy-X.XXX.tar.gz (where X.XXX is the version number)
.fi
Win32 users who do not have tar installed on their system can find
a Windows version of the program at:
.nf
http://unxutils.sourceforge.net/
.fi
.SS Install Method #1 - All Systems (Semi-Automated)
Enter the directory created in the unpacking step above. Then issue
the following command:
.nf
python setup.py install
.fi
This will install the \*(TC module and compile it.
You will manually have to copy the 'test-tc.py' program to a directory
somewhere in your executable path. Similarly, copy the documentation
files to locations appropriate for your system.
.SS Install Method #2 - All Systems (Manual)
Enter the directory created in the unpacking step above. Then,
manually copy the tconfpy.py file to a directory somewhere in your
PYTHONPATH. The recommended location for Unix-like systems is:
.nf
.../pythonX.Y/site-packages
.fi
For Win32 systems, the recommended location is:
.nf
...\\PythonX.Y\\lib\\site-packages
.fi
Where X.Y is the Python release number.
You can pre-compile the \*(TC module by starting Python interactively
and then issuing the command:
.nf
import tconfpy
.fi
Manually copy the 'test-tc.py' program to a directory
somewhere in your executable path. Copy the documentation
files to locations appropriate for your system.
.SS Install Method #3 - FreeBSD Only (Fully-Automated)
Make sure you are logged in as root, then:
.nf
cd /usr/ports/devel/py-tconfpy
make install
.fi
This is a fully-automated install that puts both code and
documentation where it belongs. After this command has completed
you'll find the license agreement and all the documentation (in the
various formats) in:
.nf
/usr/local/share/doc/py-tconfpy
.fi
The 'man' pages will have been properly installed so either of these
commands will work:
.nf
man tconfpy
man test-tc
.fi
.SS Bundling \*(TC With Your Own Programs
If you write a program that depends on \*(TC you'll need to ensure
that the end-users have it installed on their systems. There are two
ways to do this:
1) Tell them to download and install the package as described above.
This is not recommended since you cannot rely on the technical
ability of end users to do this correctly.
2) Just include 'tconfpy.py' in your program distribution directory.
This ensures that the module is available to your program
regardless of what the end-user system has installed.
.SH OTHER
\*(TC requires Python 2.3 or later.
.SH BUGS AND MISFEATURES
None known as of this release.
.SH COPYRIGHT AND LICENSING
\*(TC is Copyright(c) \*(CP TundraWare Inc. For terms of use, see
the tconfpy-license.txt file in the program distribution. If you
install \*(TC on a FreeBSD system using the 'ports' mechanism, you
will also find this file in /usr/local/share/doc/py-tconfpy.
.SH AUTHOR
.nf
Tim Daneliuk
tconfpy@tundraware.com
tundra