gecko-dev/webtools/tinderbox2/Goals

USERS
-----

It is helpful to consider the types of users for this system and their
special needs.

  * Project Managers: not very technical will be administering the
    project and need a GUI to help change various types of project
    information (tree state, message of the day, etc).  They will also
    need summary pages which will show them the current status of all
    the projects which they are working on.  They may occasionally
    "drill down" into the detailed status page but this will not be
    their primary view of the Tinderbox2 system.

  * Build Administrator: A system administrator who will be in charge
    of setting up the build machines, configuring Tinderbox2 and other
    build systems (bugzilla, cvs, bonsai, etc).  A GUI would not be
    helpful as local customizations may require small changes to the code.

    Configurations need to be kept (mostly) in files which are separate
    from the Tinderbox2 source code so that they can be version controlled
    and will not get stepped on when Tinderbox2 is upgraded.

  * Developers: need to view the "state of development" and add notices
    to the notice board.

Improvements needed from Tinderbox1
-----------------------------------

  * Highly configurable design with multiple Version Control systems
    possible (bonsai, raw cvs, perforce, continuous, clearcase) and
    multiple modes of running possible (with no version control system
    with no builds display).

  * Clear programatic interfaces and better separation of functionality
    into separate modules.

  * Modules should not have circular dependences.  Care should be taken
    on adding 'use' statement as you should worry very much about which
    modules depend on which other modules.  It is important to consider
    which modules need to work in isolation and which modules need to
    share data.

  * It should be possible to add hooks so that users get beeped when the
    next good build goes through or that trouble tickets are automatically
    opened when the builds fail.

  * Greater flexibility in setting status of builds.  We may need more
    graduations of failure then just 'busted' or 'test-failed' to
    distinguish the types of tests which have failed.

  * Generated HTML must be readable and help isolate programming errors.

  * All programmable configuration parameters should be stored easy change
    and configure for novice users.

  * Make better use of the Perl data structures to mirror the way we wish
    to use the data.  This will allow easier maintainability and allow for
    more expansion of features.

  * Display should work on many different browsers.

  * Popup windows should not be Netscape specific.

  * Permanent data should be stored via datadumper so that the data and
    data structures are easy to read and debug. Currently this is a
    performance bottle neck with a large percentage of cpu time during
    testing being spent in Data::Dumper::Dump.  The Perl module Storable
    is much faster and there is no need to add additional module
    requirements at this time, this will be configurable.

    # dprofpp says that:
    #	%64.8 of elapsed real time which is 66.25 seconds 
    #	(out of 102.15 Seconds)
    #	was spent in 3 calls to TinderDB::VC::apply_db_updates()

    #	%58.0 of user time which is 11.05 seconds 
    #	(out of 19.03 User/102.15 Elapsed Seconds)
    #	was spend in 32878 calls to Data::Dumper::_dump()

    #	System Time was negligible at 2.49 Seconds

  * All errors should be trapped and sent to log files.  Strange program
    states should be explicitly checked for.

  * Databases should update atomically, no information should be lost due
    to race conditions.

  * All column modules (processmail, build, VC, Notices) should be able to
    be run individually.  Modules should accept well defined text files as
    input and produce text files as output.  This will greatly enhance the
    ability to test each module in isolation and to quickly port modules
    to new architectures.

  * The source code should be able to run using the standard Perl
    libraries, as it can be difficult for some users to add libraries
    which are not included in the perl distribution onto production
    machines.

  * Put CVS keywords into all the source files so that when the software
    is deployed, there is no doubt what version was checked out and where
    the files are stored in the local version control system.

  * Pay closer attention to security.  Use taint Perl to ensure that
    harmful input does not get used in unexpected ways.  Keep the flow of
    control clear and simple.

  * Allow for use of any text browser which can display tables.
    Currently this is true for the browser
    links (http://artax.karlin.mff.cuni.cz/~mikulas/links/) but not the
    browser lynx.