%TOC%
++Name+
mystcl? catalytcl? kinetcl? attcl? systematcl? optcl? genetcl? grammatcl? elliptcl? gnostcl? identcl? linguistcl? practcl? phonetcl? cortcl? synaptcl? or maybe an acronym? tenet: tcl extention network

++Requirements+
+++Use cases+
development will proceed with the perspective of the repository being used in the following ways:

  * programmers looking for specific functions
  programmers looking for new features which might be useful (browsing)
  programmers looking for a specific package
  programmers looking for package docs
  users looking for a dependency (this part needs to be very simple)

+++Metadata+
pkg metadata file must be both human and computer readable


|+Reqd|+Auto|+Data|
|  X  ||package name|
|  X  ||descr (short description)|
|||description|
|||categories/tags|
|  X  |  X  |package provides (might differ from human readable name)|
|  X  |  X  |filenames (does it have to be a single file?)|
|  X  |  X  |dependencies (needs to be in a format we can xref)|
|  X  |  X  |version|
|  X  ||platforms|
|  X  |  X  |type (binary/source/tcl/etc)|
|  X  ||installer (configure/make,script,binary installer, none, etc)|
||  X  |released date stamp|
||  X  |uploaded date stamp|
|||author|
|||relevant links|
|||change log|
||  X  |commands provided|
|  X  |  X  |list of files installed by package|
||  X  |total size of installed files?|
|  X  |  X  |checksum|

+++Issues+
  * where do the docs go?

+++Required Repository Functionality+
  * must support binaries, source, pure tcl, starkits, .tm files.
  must handle multiple package types (source/binary/etc) and platforms for a single package. we could offer a small tcl installer that package maintainers could use, but the intention here is not to invent another packaging format or system.
  can contain multiple versions of packages. only latest is displayed by default but full version history available
  package documentation
  RESTful api for package management apps
  browse by package names, tags, dates, platform, license
  searchable by commands, name, tag, author, description, docs
  editing of metadata by maintainers after initial creation
  easy user contributions (include guidelines/howto)
  append comments to package pages (like php manual)
  rss feeds: tag, package, everything

+++Optional Repository Functionality+
Do not preclude these in the initial design but do not get overly ambitious
  * screenshots (per version or per package?)
  hosted version control
  trackers
  wiki per package

+++Reference Client Features+
a client isnt quite within scope of the project but it might be good to release a basic usable reference client at the same time as the repository. there is also the possibility of an upload client which could possibly provide some usability enhancements over a web interface.

  * command line or gui
  list installed packages and versions
  search available packages
  update installed package(s)
  install new package and dependencies
  remove package

+++Advanced Client Features+
  * install in user sandbox and/or for the whole system
  ability for site administrators to customize the package list

++Questions+
  * how to resolve name collisions? I think this will be up to the client

++Policies+
  * certain metadata bits are required (keep it minimal)
  versions cant be replaced, only incremented. certain information is mutable without bumping the version, such as docs. data that is definitely *not* changable: name, license, released, provides

++Site Organization+
The main sections should be as follows

  * browse
  ** name
     date
     tag
     license
  * search
  ** everything
     advanced
  * package home
    ** released, license, provides, requires, etc
       documentation
       older versions
    *** edit
  * package edit
  * account home
    ** user info
    ** projects
      *** new
  ** settings
  * help/faqs
    ** about
    ** contributing

++Workflow+
+++Add new package+
  1 create a user account if needed
  upload files (does it have to be a single file?)
  app decodes files and captures as much metadata as possible
  user fills out web form adding or correcting metadata
  done
+++Update a package+
2 cases:

  1 new upstream release, with a new version number etc.
  1 a packager may have fixed a mistake in an earlier upload and needs to replace it. So, same release version number built from the same source release, but it should still count as an "upgrade".  

Most Linux distributions append an auxilliary "release" or "build" number to the upstream version number to handle this case.  Example:

  1 upstream releases package foo version 3.4
  1 packager builds binary foo-3.4-1.tar.gz, signs it and uploads it to the repository
  1 packager finds mistake in packaging, needs to apply patch, or whatever
  1 packager fixes problem, builds binary foo-3.4-2.tar.gz and uploads that.

+++Remove a package+
+++Find a package+
  1 use a client which uses the api (the client will present its own search, list, or other interface)
or
  1 hit the front page
  search, or browse by tag, etc
+++Download and install a package+
  1 find a package
  select a platform
  select a version
  select a package type (binary, source, pure tcl, etc) (optional)

++DB Schema+
  create table file (
  id integer primary key,
  filename varchar,
  size integer,
  checksum varchar,
  instance integer,
  foreign key (instance) references instance (id)
  );

  create table bundle (
  id integer primary key,
  name varchar,
  descr varchar,
  description text,
  provides varchar,
  -- key value pairs, eg home, docs, cvs, etc. should this be a table?
  links text,
  );

  create table instance (
  id integer primary key,
  bundle integer,
  version
  -- (binary/source/tcl/etc)
  type varchar,
  -- this only allows for one platform, but platform table could have entries like "unix", "windows", "all"
  platform integer,
  installer (configure/make,script,binary installer, none, etc)
  documentation text,
  uploaded_by integer,
  uploaded datetime,
  modified datetime,
  modified_by integer,
  released date,
  -- list of installed files?
  files text,
  -- sum of installed files
  size integer,
  commands varchar,
  author integer,
  license varchar,
  -- may create non "real" instances for purpose of depending on them
  real integer,
  foreign key (bundle) references bundle (id),
  foreign key (platform) references platform (id),
  foreign key (author,uploaded_by,modified_by) references user (id,id,id)
  );

  create table dependencies (
  instance integer,
  package varchar,
  -- exact, greater than, conflicts, etc
  type varchar,
  -- tip 268 style version spec
  version varchar,
  foreign key (instance) references instance (id)
  );

  create table platform (
  id
  name
  identifier
  );

  create table user (
  id integer primary key,
  username varchar unique not null,
  password varchar,
  realname varchar,
  email varchar,
  registered datetime,
  login datetime,
  -- may create non "real" users for purpose of authors
  real integer
  );

  create table perms (
  bundle integer,
  user integer,
  -- perm string is single letter, or word? multiple words/letters per row or multiple rows?
  perm varchar,
  foreign key (bundle) references bundle (id),
  foreign key (user) references user (id)
  );

  create table tags (
  bundle integer,
  tag varchar,
  foreign key (bundle) references bundle (id)
  );

  create table comments (
  bundle integer,
  -- what to do about anonymous comments? maybe nonreal anonymous user entry
  user integer,
  comment text,
  posted datetime,
  foreign key (bundle) references bundle (id)
  );

++Controller Actions+
add new package

add new version

update version

add user

update user

search

edit maintainers

get metadata from uploaded package

_seems like a short list, what am I missing?_

++Activity Diagrams+
+++Upload+
Requirements
++++Requirements+
  * data
  metadata
  authentication
  authorization
  result

++++Methods+
  * data - http post, scp?
  metadata - api (included in post), tpm file, web form
  authentication - http basic, cookie
++++API+
  result - api: token or error via xml, web form: direct display
  1 authenticated api call made, file and optional metadata supplied, token returned

  api call made with token, returns not ready
  instance entry is created in incoming db table. file is decoded to try to automatically determine any missing metadata. db row filled in
  api call made with token, returns ok or more info needed

  if ok then temp row is moved to real table

  if more info needed then temp row expires after X days. user can log in and provide token to fill in missing data. missing data is returned with more info result so client can prompt user

++++Web+
  1 logged in user presented with new instance form
  file attached, optional form filled out, submitted
  instance entry is created in incoming db table. file is decoded to try to automatically determine any missing metadata. db row filled in
  results page refreshes every X seconds until ready
  if ok then temp row is moved to real table
  if more info needed then prompt user

++Other Input+
Anonymous on Nov 11 2007 wrote:

Some things to consider:
  1 MD5 or some other checksum so that a person can confirm they got the right thing
  Some sort of authentication so that one can track what changes were made by whom (in case of bugs, trojans, etc.)
  Packages should be able to be installed without the installer having to create special directories (don't assume things are going into /usr/local, etc.)
  Packages should include doc - or at least a pointer to where to find doc
  Authority location for package should be required
  Orphan packages (where author has disappeared or ceased to be interested in supporting package) should be marked as such
  Users of the repository should be able to relatively simply be able to determine what items are new


jenglish> It would be *really nice* to be able to say "make upload" and not have to type anything else other than maybe a GPG passphrase to get a new build uploaded.


Colin> server needs to be able to answer 'is anything in this set newer than X'

Colin> reduce accesses


muonics would like to see a namespace registry


nem and colin suggest allowing for an http vfs