tie - Array persistence
The tie package provides a framework for the creation of persistent Tcl array variables. It should be noted that the provided mechanism is generic enough to also allow its usage for the distribution of the contents of Tcl arrays over multiple threads and processes, i.e. communication.
This, persistence and communication, is accomplished by tying) a Tcl array variable to a data source. Examples of data sources are other Tcl arrays and files.
It should be noted that a single Tcl array variable can be tied to more than one data source. It is this feature which allows the framework to be used for communication as well. Just tie several Tcl arrays in many client processes to a Tcl array in a server and all changes to any of them will be distributed to all. Less centralized variants of this are of course possible as well.
This section describes the basic API used to establish and remove ties between Tcl array variables and data sources. This interface is the only one a casual user has to be concerned about. The following sections about the various internal interfaces can be safely skipped.
This command establishes a tie between the Tcl array whose name is provided by the argument arrayvarname and the data source identified by the dstype and its series of dsname arguments. All changes made to the Tcl array after this command returns will be saved to the data source for safekeeping (or distribution).
The result of the command is always a token which identifies the new tie. This token can be used later to destroy this specific tie.
The name of the Tcl array variable to connect the new tie to.
This argument specifies the type of the data source we wish to access. The dstype can be one of log, array, remotearray, file, growfile, or dsource; in addition, the programmer can register additional data source types. Each dstype is followed by one or more arguments that identify the data source to which the array is to be tied.
The series of dsname arguments coming after the dstype identifies the data source we wish to connect to, and has to be appropriate for the chosen type.
The command understands a number of additional options which guide the process of setting up the connection between Tcl array and data source.
This command dissolves one or more ties associated with the Tcl array named by arrayvarname. If no token is specified then all ties to that Tcl array are dissolved. Otherwise only the tie the token stands for is removed, if it is actually connected to the array. Trying to remove a specific tie not belonging to the provided array will cause an error.
It should be noted that while severing a tie will destroy management information internal to the package the data source which was handled by the tie will not be touched, only closed.
After the command returns none of changes made to the array will be saved to the data source anymore.
The result of the command is an empty string.
The name of a Tcl array variable which may have ties.
A handle representing a specific tie. This argument is optional.
This command returns a list of ties associated with the Tcl array variable named by arrayvarname. The result list will be empty if the variable has no ties associated with it.
This command returns a dictionary of registered types, and the class commands they are associated with.
This command returns the fully resolved class command for a type name. This means that the command will follow a chain of type definitions ot its end.
This package provides the six following types as examples and standard data sources.
This data source does not maintain any actual data, nor persistence. It does not accept any identifying arguments. All changes are simply logged to stdout.
This data source uses a regular Tcl array as the origin of the persistent data. It accepts a single identifying argument, the name of this Tcl array. All changes are mirrored to that array.
It accepts three identifying arguments, the name of the other Tcl array, the command prefix for the send-like accessor command, and an identifier for the remote entity hosting the array, in this order. All changes are mirrored to that array, via the command prefix. All commands will be executed in the context of the global namespace.
send-like means that the command prefix has to have send syntax and semantics. I.e. it is a channel over which we can send arbitrary commands to some other entity. The remote array data source however uses only the commands set, unset, array exists, array names, array set, and array get to retrieve and set values in the remote array.
The command prefix and the entity id are separate to allow the data source to use options like -async when assembling the actual commands.
Examples of command prefixes, listed with the id of the remote entity, without options. In reality only the part before the id is the command prefix:
The Tcl array is in a remote interpreter and is accessed via Tk's X communication.
The Tcl array is in a remote interpreter and is accessed through a socket.
The Tcl array is in a remote interpreter in a different thread of this process.
This data source uses a single file as origin of the persistent data. It accepts a single identifying argument, the path to this file. The file has to be both readable and writable. It may not exist, the data source will create it in that case. This (and only this) situation will require that the directory for the file exists and is writable as well.
All changes are saved in the file, as proper Tcl commands, one command per operation. In other words, the file will always contain a proper Tcl script.
If the file exists when the tie using it is set up, then it will be compacted, i.e. superfluous operations are removed, if the operations log stored in it contains either at least one operation clearing the whole array, or at least 1.5 times more operations than entries in the loaded array.
This data source is like file in terms of the storage medium for the array data, and how it is configured. In constrast to the former it however assumes and ensures that the tied array will never shrink. I.e. the creation of new array entries, and the modification of existing entries is allowed, but the deletion of entries is not, and causes the data source to throw errors.
This restriction allows us to simplify both file format and access to the file radically. For one, the file is read only once and the internal cache cannot be invalidated. Second, writing data is reduced to a simple append, and no compaction step is necessary. The format of the contents is the string representation of a dictionary which can be incrementally extended forever at the end.
This data source uses an explicitly specified data source object as the source for the persistent data. It accepts a single identifying argument, the command prefix, i.e. object command.
To use this type it is necessary to know how the framework manages ties and what data source objects are.
All changes are delegated to the specified object.
This section is of no interest to the casual user of ties. Only developers wishing to create new data sources have to know the information provided herein.
All ties are represented internally by an in-memory object which mediates between the tie framework and the specific data source, like an array, file, etc. This is the data source object.
Its class, the data source class is not generic, but specific to the type of the data source. Writing a new data source requires us to write such a class, and then registering it with the framework as a new type.
The following subsections describe the various APIs a data source class and the objects it generates will have to follow to be compatible with the tie framework.
Data source objects are normally automatically created and destroyed by the framework when a tie is created, or removed. This management can be explicitly bypassed through the usage of the "dsource" type. The data source for this type is a data source object itself, and this object is outside of the scope of the tie framework and not managed by it. In other words, this type allows the creation of ties which talk to pre-existing data source objects, and these objects will survive the removal of the ties using them as well.
After a data source class has been written it is necessary to register it as a new type with the framework.
Using this command causes the tie framework to remember the class command dsclasscmd of a data source class under the type name dstype.
After the call the argument dstype of the basic user command ::tie::tie will accept dstype as a type name and translate it internally to the appropriate class command for the creation of data source objects for the new data source.
Each data source class is represented by a single command, also called the class command, or object creation command. Its syntax is
The first argument of the class command is the name of the data source object to create. The framework itself will always supply the string %AUTO%, to signal that the class command has to generate not only the object, but the object name as well.
This is followed by a series of arguments identifying the data source the new object is for. These are the same dsname arguments which are given to the basic user command ::tie::tie. Their actual meaning is dependent on the data source class.
The result of the class command has to be the fully-qualified name of the new data source object, i.e. the name of the object command. The interface this command has to follow is described in the section DATA SOURCE OBJECT API
Please read the section DATA SOURCE CLASS first, to know how to generate new object commands.
Each object command for a data source object has to provide at least the methods listed below for proper inter-operation with the tie framework. Note that the names of most of the methods match the subcommands of the builtin array command.
This method is called when the object ds is destroyed. It now has to release all its internal resources associated with the external data source.
This command has to return a list containing the names of all keys found in the data source the object talks to. This is equivalent to array names.
This command has to return an integer number specifying the number of keys found in the data source the object talks to. This is equivalent to array size.
This command has to return a dictionary containing the data found in the data source the object talks to. This is equivalent to array get.
This command takes a dictionary and adds its contents to the data source the object talks to. This is equivalent to array set.
This command takes a pattern and removes all elements whose keys matching it from the data source. If no pattern is specified it defaults to *, causing the removal of all elements. This is nearly equivalent to array unset.
This command has to save the value in the data source the object talks to, under the key index.
The result of the command is ignored. If an error is thrown then this error will show up as error of the set operation which caused the method call.
This command has to remove the value under the key index from the data source the object talks to.
The result of the command is ignored. If an error is thrown then this error will show up as error of the unset operation which caused the method call.
This command has to return the value for the key index in the data source the object talks to.
And here a small table comparing the data source methods to the regular Tcl commands for accessing an array.
Regular Tcl Data source ----------- ----------- array names a ds names array size a ds size array get a ds get array set a dict ds set dict array unset a pattern ds unset ?pattern? ----------- ----------- set a($idx) $val ds setv idx val unset a($idx) ds unsetv idx $a($idx) ds getv idx ----------- -----------
This document, and the package it describes, will undoubtedly contain bugs and other problems. Please report such in the category tie of the Tcllib Trackers. Please also report any ideas for enhancements you may have for either package and/or documentation.