[Crit me now!]


An Issue-Tracking System for Knowledge Workers

Ka-Ping Yee

Implementation Guide


  1. Introduction
  2. The Layer Cake
  3. Hyperdatabase
    1. Dates and Date Arithmetic
    2. Nodes and Classes
    3. Identifiers and Designators
    4. Property Names and Types
    5. Interface Specification
    6. Application Example
  4. Roundup Database
    1. Reserved Classes
      1. Users
      2. Messages
      3. Files
    2. Item Classes
    3. Interface Specification
    4. Default Schema
  5. Detector Interface
    1. Interface Specification
    2. Detector Example
  6. Command Interface
    1. Interface Specification
    2. Usage Example
  7. E-mail User Interface
    1. Message Processing
    2. Nosy Lists
    3. Setting Properties
    4. Workflow Example
  8. Web User Interface
    1. Views and View Specifiers
    2. Displaying Properties
    3. Index Views
      1. Index View Specifiers
      2. Filter Section
      3. Index Section
      4. Sorting
    4. Item Views
      1. Item View Specifiers
      2. Editor Section
      3. Spool Section
  9. Deployment Scenarios
  10. Acknowledgements

1. Introduction

This document presents a description of the components of the Roundup system and specifies their interfaces and behaviour in sufficient detail to guide an implementation. For the philosophy and rationale behind the Roundup design, see the first-round Software Carpentry submission for Roundup. This document fleshes out that design as well as specifying interfaces so that the components can be developed separately.

2. The Layer Cake

Lots of software design documents come with a picture of a cake. Everybody seems to like them. I also like cakes (i think they are tasty). So i, too, shall include a picture of a cake here.

E-mail Client

Web Browser

Detector Scripts


E-mail User Interface

Web User Interface

Detector Interface

Command Interface

Roundup Database Layer

Hyperdatabase Layer

Storage Layer

The colourful parts of the cake are part of our system; the faint grey parts of the cake are external components.

I will now proceed to forgo all table manners and eat from the bottom of the cake to the top. You may want to stand back a bit so you don't get covered in crumbs.

3. Hyperdatabase

The lowest-level component to be implemented is the hyperdatabase. The hyperdatabase is intended to be a flexible data store that can hold configurable data in records which we call nodes.

The hyperdatabase is implemented on top of the storage layer, an external module for storing its data. The storage layer could be a third-party RDBMS; for a "batteries-included" distribution, implementing the hyperdatabase on the standard bsddb module is suggested.

3.1. Dates and Date Arithmetic

Before we get into the hyperdatabase itself, we need a way of handling dates. The hyperdatabase module provides Timestamp objects for representing date-and-time stamps and Interval objects for representing date-and-time intervals.

As strings, date-and-time stamps are specified with the date in international standard format (yyyy-mm-dd) joined to the time (hh:mm:ss) by a period ("."). Dates in this form can be easily compared and are fairly readable when printed. An example of a valid stamp is "2000-06-24.13:03:59". We'll call this the "full date format". When Timestamp objects are printed as strings, they appear in the full date format with the time always given in GMT. The full date format is always exactly 19 characters long.

For user input, some partial forms are also permitted: the whole time or just the seconds may be omitted; and the whole date may be omitted or just the year may be omitted. If the time is given, the time is interpreted in the user's local time zone. The Date constructor takes care of these conversions. In the following examples, suppose that yyyy is the current year, mm is the current month, and dd is the current day of the month; and suppose that the user is on Eastern Standard Time.

Date intervals are specified using the suffixes "y", "m", and "d". The suffix "w" (for "week") means 7 days. Time intervals are specified in hh:mm:ss format (the seconds may be omitted, but the hours and minutes may not).

The Date class should understand simple date expressions of the form stamp + interval and stamp - interval. When adding or subtracting intervals involving months or years, the components are handled separately. For example, when evaluating "2000-06-25 + 1m 10d", we first add one month to get 2000-07-25, then add 10 days to get 2000-08-04 (rather than trying to decide whether 1m 10d means 38 or 40 or 41 days).

Here is an outline of the Date and Interval classes.

class Date:
    def __init__(self, spec, offset):
        """Construct a date given a specification and a time zone offset.

        'spec' is a full date or a partial form, with an optional
        added or subtracted interval.  'offset' is the local time
        zone offset from GMT in hours.

    def __add__(self, interval):
        """Add an interval to this date to produce another date."""

    def __sub__(self, interval):
        """Subtract an interval from this date to produce another date."""

    def __cmp__(self, other):
        """Compare this date to another date."""

    def __str__(self):
        """Return this date as a string in the yyyy-mm-dd.hh:mm:ss format."""

    def local(self, offset):
        """Return this date as yyyy-mm-dd.hh:mm:ss in a local time zone."""

class Interval:
    def __init__(self, spec):
        """Construct an interval given a specification."""

    def __cmp__(self, other):
        """Compare this interval to another interval."""
    def __str__(self):
        """Return this interval as a string."""

Here are some examples of how these classes would behave in practice. For the following examples, assume that we are on Eastern Standard Time and the current local time is 19:34:02 on 25 June 2000.

>>> Date(".")
<Date 2000-06-26.00:34:02>
>>> _.local(-5)
>>> Date(". + 2d")
<Date 2000-06-28.00:34:02>
>>> Date("1997-04-17", -5)
<Date 1997-04-17.00:00:00>
>>> Date("01-25", -5)
<Date 2000-01-25.00:00:00>
>>> Date("08-13.22:13", -5)
<Date 2000-08-14.03:13:00>
>>> Date("14:25", -5)
<Date 2000-06-25.19:25:00>
>>> Interval("  3w  1  d  2:00")
<Interval 22d 2:00>
>>> Date(". + 2d") - Interval("3w")
<Date 2000-06-07.00:34:02>

3.2. Nodes and Classes

Nodes contain data in properties. To Python, these properties are presented as the key-value pairs of a dictionary. Each node belongs to a class which defines the names and types of its properties. The database permits the creation and modification of classes as well as nodes.

3.3. Identifiers and Designators

Each node has a numeric identifier which is unique among nodes in its class. The nodes are numbered sequentially within each class in order of creation, starting from 1. The designator for a node is a way to identify a node in the database, and consists of the name of the node's class concatenated with the node's numeric identifier.

For example, if "spam" and "eggs" are classes, the first node created in class "spam" has id 1 and designator "spam1". The first node created in class "eggs" also has id 1 but has the distinct designator "eggs1". Node designators are conventionally enclosed in square brackets when mentioned in plain text. This permits a casual mention of, say, "[patch37]" in an e-mail message to be turned into an active hyperlink.

3.4. Property Names and Types

Property names must begin with a letter.

A property may be one of five basic types:

None is also a permitted value for any of these property types. An attempt to store None into a String property stores the empty string; an attempt to store None into a Multilink property stores an empty list.

3.5. Interface Specification

The hyperdb module provides property objects to designate the different kinds of properties. These objects are used when specifying what properties belong in classes.

class String:
    def __init__(self):
        """An object designating a String property."""

class Date:
    def __init__(self):
        """An object designating a Date property."""

class Link:
    def __init__(self, classname):
        """An object designating a Link property that links to
        nodes in a specified class."""

class Multilink:
    def __init__(self, classname):
        """An object designating a Multilink property that links
        to nodes in a specified class."""

Here is the interface provided by the hyperdatabase.

class Database:
    """A database for storing records containing flexible data types."""

    def __init__(self, storagelocator, journaltag):
        """Open a hyperdatabase given a specifier to some storage.

        The meaning of 'storagelocator' depends on the particular
        implementation of the hyperdatabase.  It could be a file name,
        a directory path, a socket descriptor for a connection to a
        database over the network, etc.

        The 'journaltag' is a token that will be attached to the journal
        entries for any edits done on the database.  If 'journaltag' is
        None, the database is opened in read-only mode: the Class.create(),
        Class.set(), and Class.retire() methods are disabled.

    def __getattr__(self, classname):
        """A convenient way of calling self.getclass(classname)."""

    def getclasses(self):
        """Return a list of the names of all existing classes."""

    def getclass(self, classname):
        """Get the Class object representing a particular class.

        If 'classname' is not a valid class name, a KeyError is raised.

class Class:
    """The handle to a particular class of nodes in a hyperdatabase."""

    def __init__(self, db, classname, **properties):
        """Create a new class with a given name and property specification.

        'classname' must not collide with the name of an existing class,
        or a ValueError is raised.  The keyword arguments in 'properties'
        must map names to property objects, or a TypeError is raised.

    # Editing nodes:

    def create(self, **propvalues):
        """Create a new node of this class and return its id.

        The keyword arguments in 'propvalues' map property names to values.
        The values of arguments must be acceptable for the types of their
        corresponding properties or a TypeError is raised.  If this class
        has a key property, it must be present and its value must not
        collide with other key strings or a ValueError is raised.  Any other
        properties on this class that are missing from the 'propvalues'
        dictionary are set to None.  If an id in a link or multilink
        property does not refer to a valid node, an IndexError is raised.

    def get(self, nodeid, propname):
        """Get the value of a property on an existing node of this class.

        'nodeid' must be the id of an existing node of this class or an
        IndexError is raised.  'propname' must be the name of a property
        of this class or a KeyError is raised.

    def set(self, nodeid, **propvalues):
        """Modify a property on an existing node of this class.
        'nodeid' must be the id of an existing node of this class or an
        IndexError is raised.  Each key in 'propvalues' must be the name
        of a property of this class or a KeyError is raised.  All values
        in 'propvalues' must be acceptable types for their corresponding
        properties or a TypeError is raised.  If the value of the key
        property is set, it must not collide with other key strings or a
        ValueError is raised.  If the value of a Link or Multilink
        property contains an invalid node id, a ValueError is raised.

    def retire(self, nodeid):
        """Retire a node.
        The properties on the node remain available from the get() method,
        and the node's id is never reused.  Retired nodes are not returned
        by the find(), list(), or lookup() methods, and other nodes may
        reuse the values of their key properties.

    def history(self, nodeid):
        """Retrieve the journal of edits on a particular node.

        'nodeid' must be the id of an existing node of this class or an
        IndexError is raised.

        The returned list contains tuples of the form

            (date, tag, action, params)

        'date' is a Timestamp object specifying the time of the change and
        'tag' is the journaltag specified when the database was opened.
        'action' may be:

            'create' or 'set' -- 'params' is a dictionary of property values
            'link' or 'unlink' -- 'params' is (classname, nodeid, propname)
            'retire' -- 'params' is None

    # Locating nodes:

    def setkey(self, propname):
        """Select a String property of this class to be the key property.

        'propname' must be the name of a String property of this class or
        None, or a TypeError is raised.  The values of the key property on
        all existing nodes must be unique or a ValueError is raised.

    def getkey(self):
        """Return the name of the key property for this class or None."""

    def lookup(self, keyvalue):
        """Locate a particular node by its key property and return its id.

        If this class has no key property, a TypeError is raised.  If the
        'keyvalue' matches one of the values for the key property among
        the nodes in this class, the matching node's id is returned;
        otherwise a KeyError is raised.

    def find(self, propname, nodeid):
        """Get the ids of nodes in this class which link to a given node.
        'propname' must be the name of a property in this class, or a
        KeyError is raised.  That property must be a Link or Multilink
        property, or a TypeError is raised.  'nodeid' must be the id of
        an existing node in the class linked to by the given property,
        or an IndexError is raised.

    def list(self):
        """Return a list of the ids of the active nodes in this class."""

    def count(self):
        """Get the number of nodes in this class.

        If the returned integer is 'numnodes', the ids of all the nodes
        in this class run from 1 to numnodes, and numnodes+1 will be the
        id of the next node to be created in this class.

    # Manipulating properties:

    def getprops(self):
        """Return a dictionary mapping property names to property objects."""

    def addprop(self, **properties):
        """Add properties to this class.

        The keyword arguments in 'properties' must map names to property
        objects, or a TypeError is raised.  None of the keys in 'properties'
        may collide with the names of existing properties, or a ValueError
        is raised before any properties have been added.

3.6. Application Example

Here is an example of how the hyperdatabase module would work in practice.

>>> import hyperdb
>>> db = hyperdb.Database("foo.db", "ping")
>>> db
<hyperdb.Database "foo.db" opened by "ping">
>>> hyperdb.Class(db, "status", name=hyperdb.String())
<hyperdb.Class "status">
>>> _.setkey("name")
>>> db.status.create(name="unread")
>>> db.status.create(name="in-progress")
>>> db.status.create(name="testing")
>>> db.status.create(name="resolved")
>>> db.status.count()
>>> db.status.list()
[1, 2, 3, 4]
>>> db.status.lookup("in-progress")
>>> db.status.retire(3)
>>> db.status.list()
[1, 2, 4]
>>> hyperdb.Class(db, "issue", title=hyperdb.String(), status=hyperdb.Link("status"))
<hyperdb.Class "issue">
>>> db.issue.create(title="spam", status=1)
>>> db.issue.create(title="eggs", status=2)
>>> db.issue.create(title="ham", status=4)
>>> db.issue.create(title="arguments", status=2)
>>> db.issue.create(title="abuse", status=1)
>>> hyperdb.Class(db, "user", username=hyperdb.Key(), password=hyperdb.String())
<hyperdb.Class "user">
>>> db.issue.addprop(fixer=hyperdb.Link("user"))
>>> db.issue.getprops()
{"title": <hyperdb.String>, "status": <hyperdb.Link to "status">,
 "user": <hyperdb.Link to "user">}
>>> db.issue.set(5, status=2)
>>> db.issue.get(5, "status")
>>> db.status.get(2, "name")
>>> db.issue.get(5, "title")
>>> db.issue.find("status", db.status.lookup("in-progress"))
[2, 4, 5]
>>> db.issue.history(5)
[(<Date 2000-06-28.19:09:43>, "ping", "create", {"title": "abuse", "status": 1}),
 (<Date 2000-06-28.19:11:04>, "ping", "set", {"status": 2})]
>>> db.status.history(1)
[(<Date 2000-06-28.19:09:43>, "ping", "link", ("issue", 5, "status")),
 (<Date 2000-06-28.19:11:04>, "ping", "unlink", ("issue", 5, "status"))]
>>> db.status.history(2)
[(<Date 2000-06-28.19:11:04>, "ping", "link", ("issue", 5, "status"))]

For the purposes of journalling, when a Multilink property is set to a new list of nodes, the hyperdatabase compares the old list to the new list. The journal records "unlink" events for all the nodes that appear in the old list but not the new list, and "link" events for all the nodes that appear in the new list but not in the old list.

4. Roundup Database

The Roundup database layer is implemented on top of the hyperdatabase and mediates calls to the database. Some of the classes in the Roundup database are considered item classes. The Roundup database layer adds detectors and user nodes, and on items it provides mail spools, nosy lists, and superseders.

4.1. Reserved Classes

Internal to this layer we reserve three special classes of nodes that are not items.

4.1.1. Users

Users are stored in the hyperdatabase as nodes of class "user". The "user" class has the definition:

hyperdb.Class(db, "user", username=hyperdb.String(),

4.1.2. Messages

E-mail messages are represented by hyperdatabase nodes of class "msg". The actual text content of the messages is stored in separate files. (There's no advantage to be gained by stuffing them into the hyperdatabase, and if messages are stored in ordinary text files, they can be grepped from the command line.) The text of a message is saved in a file named after the message node designator (e.g. "msg23") for the sake of the command interface (see below). Attachments are stored separately and associated with "file" nodes. The "msg" class has the definition:

hyperdb.Class(db, "msg", author=hyperdb.Link("user"),

The "author" property indicates the author of the message (a "user" node must exist in the hyperdatabase for any messages that are stored in the system). The "summary" property contains a summary of the message for display in a message index.

4.1.3. Files

Submitted files are represented by hyperdatabase nodes of class "file". Like e-mail messages, the file content is stored in files outside the database, named after the file node designator (e.g. "file17"). The "file" class has the definition:

hyperdb.Class(db, "file", user=hyperdb.Link("user"),

The "user" property indicates the user who submitted the file, the "name" property holds the original name of the file, and the "type" property holds the MIME type of the file as received.

4.2. Item Classes

All items have the following standard properties:


Also, two Date properties named "creation" and "activity" are fabricated by the Roundup database layer. By "fabricated" we mean that no such properties are actually stored in the hyperdatabase, but when properties on items are requested, the "creation" and "activity" properties are made available. The value of the "creation" property is the date when an item was created, and the value of the "activity" property is the date when any property on the item was last edited (equivalently, these are the dates on the first and last records in the item's journal).

4.3. Interface Specification

The interface to a Roundup database delegates most method calls to the hyperdatabase, except for the following changes and additional methods.

class Database:
    # Overridden methods:

    def __init__(self, storagelocator, journaltag):
        """When the Roundup database is opened by a particular user,
        the 'journaltag' is the id of the user's "user" node."""

    def getclass(self, classname):
        """This method now returns an instance of either Class or
        ItemClass depending on whether an item class is specified."""

    # New methods:

    def getuid(self):
        """Return the id of the "user" node associated with the user
        that owns this connection to the hyperdatabase."""

class Class:
    # Overridden methods:

    def create(self, **propvalues):
    def set(self, **propvalues):
    def retire(self, nodeid):
        """These operations trigger detectors and can be vetoed.  Attempts
        to modify the "creation" or "activity" properties cause a KeyError.

    # New methods:

    def audit(self, event, detector):
    def react(self, event, detector):
        """Register a detector (see below for more details)."""

class ItemClass(Class):
    # Overridden methods:

    def __init__(self, db, classname, **properties):
        """The newly-created class automatically includes the "messages",
        "files", "nosy", and "superseder" properties.  If the 'properties'
        dictionary attempts to specify any of these properties or a
        "creation" or "activity" property, a ValueError is raised."""

    def get(self, nodeid, propname):
    def getprops(self):
        """In addition to the actual properties on the node, these
        methods provide the "creation" and "activity" properties."""

    # New methods:

    def addmessage(self, nodeid, summary, text):
        """Add a message to an item's mail spool.

        A new "msg" node is constructed using the current date, the
        user that owns the database connection as the author, and
        the specified summary text.  The "files" and "recipients"
        fields are left empty.  The given text is saved as the body
        of the message and the node is appended to the "messages"
        field of the specified item.

    def sendmessage(self, nodeid, msgid):
        """Send a message to the members of an item's nosy list.

        The message is sent only to users on the nosy list who are not
        already on the "recipients" list for the message.  These users
        are then added to the message's "recipients" list.

4.4. Default Schema

The default schema included with Roundup turns it into a typical software bug tracker. The database is set up like this:

pri = Class(db, "priority", name=hyperdb.String(), order=hyperdb.String())
pri.create(name="critical", order="1")
pri.create(name="urgent", order="2")
pri.create(name="bug", order="3")
pri.create(name="feature", order="4")
pri.create(name="wish", order="5")

stat = Class(db, "status", name=hyperdb.String(), order=hyperdb.String())
stat.create(name="unread", order="1")
stat.create(name="deferred", order="2")
stat.create(name="chatting", order="3")
stat.create(name="need-eg", order="4")
stat.create(name="in-progress", order="5")
stat.create(name="testing", order="6")
stat.create(name="done-cbb", order="7")
stat.create(name="resolved", order="8")

Class(db, "keyword", name=hyperdb.String())

Class(db, "issue", fixer=hyperdb.Multilink("user"),

(The "order" property hasn't been explained yet. It gets used by the Web user interface for sorting.)

The above isn't as pretty-looking as the schema specification in the first-stage submission, but it could be made just as easy with the addition of a convenience function like Choice for setting up the "priority" and "status" classes:

def Choice(name, *options):
    cl = Class(db, name, name=hyperdb.String(), order=hyperdb.String())
    for i in range(len(options)):
        cl.create(name=option[i], order=i)
    return hyperdb.Link(name)

5. Detector Interface

Detectors are Python functions that are triggered on certain kinds of events. The definitions of the functions live in Python modules placed in a directory set aside for this purpose. Importing the Roundup database module also imports all the modules in this directory, and the init() function of each module is called when a database is opened to provide it a chance to register its detectors.

There are two kinds of detectors:

When the Roundup database is about to perform a create(), set(), or retire() operation, it first calls any auditors that have been registered for that operation on that class. Any auditor may raise a Reject exception to abort the operation.

If none of the auditors raises an exception, the database proceeds to carry out the operation. After it's done, it then calls all of the reactors that have been registered for the operation.

5.1. Interface Specification

The audit() and react() methods register detectors on a given class of nodes.

class Class:
    def audit(self, event, detector):
        """Register an auditor on this class.

        'event' should be one of "create", "set", or "retire".
        'detector' should be a function accepting four arguments.

    def react(self, event, detector):
        """Register a reactor on this class.

        'event' should be one of "create", "set", or "retire".
        'detector' should be a function accepting four arguments.

Auditors are called with the arguments:

audit(db, cl, nodeid, newdata)
where db is the database, cl is an instance of Class or ItemClass within the database, and newdata is a dictionary mapping property names to values. For a create() operation, the nodeid argument is None and newdata contains all of the initial property values with which the node is about to be created. For a set() operation, newdata contains only the names and values of properties that are about to be changed. For a retire() operation, newdata is None.

Reactors are called with the arguments:

react(db, cl, nodeid, olddata)
where db is the database, cl is an instance of Class or ItemClass within the database, and olddata is a dictionary mapping property names to values. For a create() operation, the nodeid argument is the id of the newly-created node and olddata is None. For a set() operation, olddata contains the names and previous values of properties that were changed. For a retire() operation, nodeid is the id of the retired node and olddata is None.

5.2. Detector Example

Here is an example of detectors written for a hypothetical project-management application, where users can signal approval of a project by adding themselves to an "approvals" list, and a project proceeds when it has three approvals.

# Permit users only to add themselves to the "approvals" list.

def check_approvals(db, cl, id, newdata):
    if newdata.has_key("approvals"):
        if cl.get(id, "status") == db.status.lookup("approved"):
            raise Reject, "You can't modify the approvals list " \
                          "for a project that has already been approved."
        old = cl.get(id, "approvals")
        new = newdata["approvals"]
        for uid in old:
            if uid not in new and uid != db.getuid():
                raise Reject, "You can't remove other users from the "
                              "approvals list; you can only remove yourself."
        for uid in new:
            if uid not in old and uid != db.getuid():
                raise Reject, "You can't add other users to the approvals "
                              "list; you can only add yourself."

# When three people have approved a project, change its
# status from "pending" to "approved".

def approve_project(db, cl, id, olddata):
    if olddata.has_key("approvals") and len(cl.get(id, "approvals")) == 3:
        if cl.get(id, "status") == db.status.lookup("pending"):
            cl.set(id, status=db.status.lookup("approved"))

def init(db):
    db.project.audit("set", check_approval)
    db.project.react("set", approve_project)

Here is another example of a detector that can allow or prevent the creation of new nodes. In this scenario, patches for a software project are submitted by sending in e-mail with an attached file, and we want to ensure that there are text/plain attachments on the message. The maintainer of the package can then apply the patch by setting its status to "applied".

# Only accept attempts to create new patches that come with patch files.

def check_new_patch(db, cl, id, newdata):
    if not newdata["files"]:
        raise Reject, "You can't submit a new patch without " \
                      "attaching a patch file."
    for fileid in newdata["files"]:
        if db.file.get(fileid, "type") != "text/plain":
            raise Reject, "Submitted patch files must be text/plain."

# When the status is changed from "approved" to "applied", apply the patch.

def apply_patch(db, cl, id, olddata):
    if cl.get(id, "status") == db.status.lookup("applied") and \
        olddata["status"] == db.status.lookup("approved"):
        # ...apply the patch...

def init(db):
    db.patch.audit("create", check_new_patch)
    db.patch.react("set", apply_patch)

6. Command Interface

The command interface is a very simple and minimal interface, intended only for quick searches and checks from the shell prompt. (Anything more interesting can simply be written in Python using the Roundup database module.)

6.1. Interface Specification

A single command, roundup, provides basic access to the hyperdatabase from the command line.

Property values are represented as strings in command arguments and in the printed results:

When multiple nodes are specified to the roundup get or roundup set commands, the specified properties are retrieved or set on all the listed nodes.

When multiple results are returned by the roundup get or roundup find commands, they are printed one per line (default) or joined by commas (with the -list) option.

6.2. Usage Example

To find all messages regarding in-progress issues that contain the word "spam", for example, you could execute the following command from the directory where the database dumps its files:

shell% for issue in `roundup find issue status=in-progress`; do
> grep -l spam `roundup get $issue messages`
> done

Or, using the -list option, this can be written as a single command:

shell% grep -l spam `roundup get \
    \`roundup find -list issue status=in-progress\` messages`

7. E-mail User Interface

The Roundup system must be assigned an e-mail address at which to receive mail. Messages should be piped to the Roundup mail-handling script by the mail delivery system (e.g. using an alias beginning with "|" for sendmail).

7.1. Message Processing

Incoming messages are examined for multiple parts. In a multipart/mixed message or part, each subpart is extracted and examined. In a multipart/alternative message or part, we look for a text/plain subpart and ignore the other parts. The text/plain subparts are assembled to form the textual body of the message, to be stored in the file associated with a "msg" class node. Any parts of other types are each stored in separate files and given "file" class nodes that are linked to the "msg" node.

The "summary" property on message nodes is taken from the first non-quoting section in the message body. The message body is divided into sections by blank lines. Sections where the second and all subsequent lines begin with a ">" or "|" character are considered "quoting sections". The first line of the first non-quoting section becomes the summary of the message.

All of the addresses in the To: and Cc: headers of the incoming message are looked up among the user nodes, and the corresponding users are placed in the "recipients" property on the new "msg" node. The address in the From: header similarly determines the "author" property of the new "msg" node. The default handling for addresses that don't have corresponding users is to create new users with no passwords and a username equal to the address. (The web interface does not permit logins for users with no passwords.) If we prefer to reject mail from outside sources, we can simply register an auditor on the "user" class that prevents the creation of user nodes with no passwords.

The subject line of the incoming message is examined to determine whether the message is an attempt to create a new item or to discuss an existing item. A designator enclosed in square brackets is sought as the first thing on the subject line (after skipping any "Fwd:" or "Re:" prefixes).

If an item designator (class name and id number) is found there, the newly created "msg" node is added to the "messages" property for that item, and any new "file" nodes are added to the "files" property for the item.

If just an item class name is found there, we attempt to create a new item of that class with its "messages" property initialized to contain the new "msg" node and its "files" property initialized to contain any new "file" nodes.

Both cases may trigger detectors (in the first case we are calling the set() method to add the message to the item's spool; in the second case we are calling the create() method to create a new node). If an auditor raises an exception, the original message is bounced back to the sender with the explanatory message given in the exception.

7.2. Nosy Lists

A standard detector is provided that watches for additions to the "messages" property. When a new message is added, the detector sends it to all the users on the "nosy" list for the item that are not already on the "recipients" list of the message. Those users are then appended to the "recipients" property on the message, so multiple copies of a message are never sent to the same user. The journal recorded by the hyperdatabase on the "recipients" property then provides a log of when the message was sent to whom.

7.3. Setting Properties

The e-mail interface also provides a simple way to set properties on items. At the end of the subject line, propname=value pairs can be specified in square brackets, using the same conventions as for the roundup set shell command.

8. Web User Interface

The web interface is provided by a CGI script that can be run under any web server. A simple web server can easily be built on the standard CGIHTTPServer module, and should also be included in the distribution for quick out-of-the-box deployment.

The user interface is constructed from a number of template files containing mostly HTML. Among the HTML tags in templates are interspersed some nonstandard tags, which we use as placeholders to be replaced by properties and their values.

8.1. Views and View Specifiers

There are two main kinds of views: index views and item views. An index view displays a list of items of a particular class, optionally sorted and filtered as requested. An item view presents the properties of a particular item for editing and displays the message spool for the item.

A view specifier is a string that specifies all the options needed to construct a particular view. It goes after the URL to the Roundup CGI script or the web server to form the complete URL to a view. When the result of selecting a link or submitting a form takes the user to a new view, the Web browser should be redirected to a canonical location containing a complete view specifier so that the view can be bookmarked.

8.2. Displaying Properties

Properties appear in the user interface in three contexts: in indices, in editors, and as filters. For each type of property, there are several display possibilities. For example, in an index view, a string property may just be printed as a plain string, but in an editor view, that property should be displayed in an editable field.

The display of a property is handled by functions in a displayers module. Each function accepts at least three standard arguments -- the database, class name, and node id -- and returns a chunk of HTML.

Displayer functions are triggered by <display> tags in templates. The call attribute of the tag provides a Python expression for calling the displayer function. The three standard arguments are inserted in front of the arguments given. For example, the occurrence of

    <display call="plain('status', max=30)">
in a template triggers a call to
    plain(db, "issue", 13, "status", max=30)
when displaying item 13 in the "issue" class. The displayer functions can accept extra arguments to further specify details about the widgets that should be generated. By defining new displayer functions, the user interface can be highly customized.

Some of the standard displayer functions include:

8.3. Index Views

An index view contains two sections: a filter section and an index section. The filter section provides some widgets for selecting which items appear in the index. The index section is a table of items.

8.3.1. Index View Specifiers

An index view specifier looks like this (whitespace has been added for clarity):


The index view is determined by two parts of the specifier: the layout part and the filter part. The layout part consists of the query parameters that begin with colons, and it determines the way that the properties of selected nodes are displayed. The filter part consists of all the other query parameters, and it determines the criteria by which nodes are selected for display.

The filter part is interactively manipulated with the form widgets displayed in the filter section. The layout part is interactively manipulated by clicking on the column headings in the table.

The filter part selects the union of the sets of items with values matching any specified Link properties and the intersection of the sets of items with values matching any specified Multilink properties.

The example specifies an index of "issue" nodes. Only items with a "status" of either "unread" or "in-progres" or "resolved" are displayed, and only items with "topic" values including both "security" and "ui" are displayed. The items are grouped by priority, arranged in ascending order; and within groups, sorted by activity, arranged in descending order. The filter section shows filters for the "status" and "topic" properties, and the table includes columns for the "title", "status", and "fixer" properties.

Associated with each item class is a default layout specifier. The layout specifier in the above example is the default layout to be provided with the default bug-tracker schema described above in section 4.4.

8.3.2. Filter Section

The template for a filter section provides the filtering widgets at the top of the index view. Fragments enclosed in <property>...</property> tags are included or omitted depending on whether the view specifier requests a filter for a particular property.

Here's a simple example of a filter template.

<property name=status>
    <display call="checklist('status')">
<property name=priority>
    <display call="checklist('priority')">
<property name=fixer>
    <display call="menu('fixer')">

8.3.3. Index Section

The template for an index section describes one row of the index table. Fragments enclosed in <property>...</property> tags are included or omitted depending on whether the view specifier requests a column for a particular property. The table cells should contain <display> tags to display the values of the item's properties.

Here's a simple example of an index template.

    <property name=title>
        <td><display call="plain('title', max=50)"></td>
    <property name=status>
        <td><display call="plain('status')"></td>
    <property name=fixer>
        <td><display call="plain('fixer')"></td>

8.3.4. Sorting

String and Date values are sorted in the natural way. Link properties are sorted according to the value of the "order" property on the linked nodes if it is present; or otherwise on the key string of the linked nodes; or finally on the node ids. Multilink properties are sorted according to how many links are present.

8.4. Item Views

An item view contains an editor section and a spool section. At the top of an item view, links to superseding and superseded items are always displayed.

8.4.1. Item View Specifiers

An item view specifier is simply the item's designator:


8.4.2. Editor Section

The editor section is generated from a template containing <display> tags to insert the appropriate widgets for editing properties.

Here's an example of a basic editor template.

    <td colspan=2>
        <display call="field('title', size=60)">
        <display call="field('fixer', size=30)">
        <display call="menu('status')>
        <display call="field('nosy', size=30)">
        <display call="menu('priority')>
    <td colspan=2>
        <display call="note()">

As shown in the example, the editor template can also request the display of a "note" field, which is a text area for entering a note to go along with a change.

When a change is submitted, the system automatically generates a message describing the changed properties. The message displays all of the property values on the item and indicates which ones have changed. An example of such a message might be this:

title: Polly Parrot is dead
priority: critical
status: unread -> in-progress
fixer: (none)
keywords: parrot,plumage,perch,nailed,dead

If a note is given in the "note" field, the note is appended to the description. The message is then added to the item's message spool (thus triggering the standard detector to react by sending out this message to the nosy list).

8.4.3. Spool Section

The spool section lists messages in the item's "messages" property. The index of messages displays the "date", "author", and "summary" properties on the message nodes, and selecting a message takes you to its content.

9. Deployment Scenarios

The design described above should be general enough to permit the use of Roundup for bug tracking, managing projects, managing patches, or holding discussions. By using nodes of multiple types, one could deploy a system that maintains requirement specifications, catalogs bugs, and manages submitted patches, where patches could be linked to the bugs and requirements they address.

10. Acknowledgements

My thanks are due to Christy Heyl for reviewing and contributing suggestions to this paper and motivating me to get it done, and to Jesse Vincent, Mark Miller, Christopher Simons, Jeff Dunmall, Wayne Gramlich, and Dean Tribble for their assistance with the first-round submission.