~azzar1/unity/add-show-desktop-key

IVLE Design - Subversion Web Access
===================================

    Author: Matt Giuca
    Date: 3/12/2007

We will provide students with a web interface to Subversion. A major focus of
the web interface is to shield new students from the details of actually
managing a repository and just provide them with what looks like a normal file
browser, which happens to backup their files.

We are considering writing our own web file browser to Subversion.

The original concept is this (as implemented in the IVLE / PynApple prototype):

* Provide an interface to just the repository. All accesses are done directly
  to the repository, the interface knows nothing about workspaces.
* Editing the files within the environment automatically commits the changes
  to the repo.
* Allow import / export of files.
* Execute the Python programs in a special "run" directory which protects the
  user's workspace.
* Presumably allow some kind of rollback to previous versions.

This was problematic because it did not allow us to browse the directories
outside of subversion (such as the execution directory).

Our new design is to write a web-based _workspace_ browser. This will give
students a more hands-on experience using a repository, but more importantly
let them see what temporary files are being created and clean them up. This
simplifies the system and reduces the amount of "magic" we have to perform
around the students.

A major advantage of this is that we no longer have to have a separate file
browser from the repository browser. There is no "execution" directory - the
student just works entirely within their own workspace. (If necessary this
browser could be made to explore parts of the file system outside of a
workspace, simply disabling the SVN features).

This means the student can have their permanent files (say "flower-in.png") in
the same directory as the temporary files (say "flower-out.png") and not need
to worry about changing directories or any of those complications.

Terminology shift
-----------------

In order to shield new students from the immediate reality of a "repository",
we will present the system as a simple extension of a normal file system they
are used to. (This section is written in a deliberately obtuse style; this is
the "story" we will tell new students).

The student is given a "workspace", with no mention of a repository. Inside
the workspace they have "permanent files" and "temporary files". Permanent
files are those which exist in both the (somewhat hidden) repository, and the
workspace, while temporary files exist only in the workspace.

Unlike originally proposed, editing and saving documents will *not*
auto-commit. We hope to augment the editor with a "save" button (which saves
to the workspace) and a "save-and-commit" button (which saves and commits the
file automatically, possibly asking for a log message).

The paradigm we present to students is: the system saves a permanent backup of
your file when you hit "save-and-commit". In fact, each time you hit
"save-and-commit" it saves a separate backup, so later you can go back to any
older version. Files backed up permanently are the so-called "permanent files"
while files which haven't been committed permanently are "temporary files".

Editing a permanent file and saving it but not committing it results in an
"out-of-date" file. This means you have made changes which aren't yet
permanently backed up. The commit button either in the editor or the file
browser will ensure these changes are permanently saved.

A useful reason we have temporary files is you can muck around with them
however you like, and if you don't like the changes, you can "roll back" or
"clean up" by hitting the cleanup button. The cleanup button does *two* things
- it "undoes" all the changes you made to out-of-date files, going back to
their last permanent version, and it also deletes any temporary files.

(Note: I think it would be a good idea to separate these. Have a "rollback"
button which is seen as a large-scale undo button, which just reverts all
out-of-date files; that is reverts uncommitted changes to files in the repo,
and a "cleanup" button which deletes all temporary files; that is deletes any
files which have not been added).

(Another note: One thing which confuses *many* students, even 3rd and 4th
year, is the difference between "add" and "commit". Hopefully we can simplify
this so that the add button also commits? We need to decide if we want to
enforce atomicity - that is, if we expect students to avoid committing one
file at a time creating broken revisions - because things like the
"save-and-commit" button on the editor, and "add-and-commit" in the file
browser really discourage atomicity. I would argue for the best.)

Note that files could also be out-of-date if the Python program somehow
mangles them. In this case they should be reverted. Because of this, the
environment could portray out-of-date files as somehow "damaged".

A final note that the current "save" button in EditArea saves to the local
disk - ie. it is really a "download" button. Will need to modify this so
"save" just writes to the workspace, and have a separate download button
(possibly only through the file browser interface).

Multiple levels of expertise
----------------------------

We have identified at least two modes to place the interface (particularly,
the file browser) into when considering the expertise levels of the users.

We plan to have new students go into a "dumbed down" mode ("novice mode")
which will hide certain features and details, in particular the subversion
features.

Essentially the novice mode will give students the impression that this is
just a file system, with no revision control. There will be just a single
"save" button in the editor which will save to the student's workspace.

The "adept" mode by contrast will provide two save buttons - "save as draft"
(or save as temporary) and "save and commit". These behave as described above.

The question is, what does the novice mode "save" button do. It could either
write to the workspace only (effectively meaning students are without revision
control until they move into adept mode), or it could auto-commit each save.
If it autocommits we may or may not let the students in novice mode browse
their history (but if necessary, a lab demonstrator could go into adept mode
to retrieve old versions).

Autocommitting each save has advantages:

* Students can be helped to retrieve old versions if they mess up or lose
  their work.
* We can check their history when we mark projects.

Disadvantages:

* CPU drain from high volume of commits (see "unresolved issues" at the bottom).
* Possibly hard to explain once they move into adept mode that "save no longer
  saves your history unless you click save+commit" - which is why I'm calling
  it "save as draft" and having NO button called simply "save".

I have to say I am inclined to give even novice users both buttons. It isn't a
terribly difficult concept to distinguish between "I am just mucking around
with this file" and "I want to keep this".

Alternatively, if we do dumb it down, perhaps instead of committing each save,
we just save to their workspace and do automatic commits from the workspace
every so often (timewise, or number-of-saves-wise).

Use cases
---------

We have identified three levels of proficiency, which we will expect students
to progress through as they go through the course.

1. Novice user (students in their first or second week with no prior
    experience to revision control).
2. Familiar user (students who have grasped the concept of revision control
    and are expected to use the system as normal developers use revision
    control for single-developer projects).
3. Group user (students who are undertaking group work and need to use
    advanced multi-user features such as conflict-and-merge handling).

### Novice user ###

#### Use case: First hello world ####

Background: The student has used IVLE's tutorial section to write simple
Python programs and also played around in the console, but they haven't used
the file system yet. They presumably are proficient managing and editing normal
files and folders as in Windows at home, school, etc.

1. Student navigates to the "files" screen. They see an empty directory.
2. Student clicks "new file". They are immediately taken to the editor with an
    untitled file. (We do not ask for filenames until they save, as most GUI
    programs behave).
3. Student types in the Python "Hello world" program, and clicks "save".
    (Tempting to print a warning about temporary files but this will probably just
    confuse them at first).
4. A dialog box asks for the filename. They type in "hello.py". As with
    familiar GUI apps, they can always choose to "Save As" later.
5. Student can now "run" the program from within the editor. (They could have
    done this before they saved - we should be able to handle running code
    which the student has not saved).
6. Returning to the file browser, student sees "hello.py" with a "temporary
    file" icon next to it. At this stage this icon can be ignored but if the
    student enquires, a demonstrator explains that temporary files are SAVED
    but you can make permanent backups of files in case you mess them up by
    clicking "commit".
7. Student clicks "commit" and the temporary file icon goes away.

Note that apart from the commit issue, this whole usage example is supposed to
feel like what the students are used to with Windows applications such as file
Explorer and Microsoft Word, Notepad, etc. (Sorry to use Microsoft metaphors
but let's face it, that's what they know).

Students will call directories "folders". Do we need to explain that they are
the same thing? Should IVLE call them "folders"? Presumably the icons will
make this rather obvious!

#### Use case: Normal workflow, single-file programs ####

Background: For the first few weeks, students will be primarily editing
programs which are single-file Python programs, and will treat the file
browser as just a directory hierarchy with backup facilities. Assume the
student already has a Python file created as above, and it is a permanent file.

1. Student opens the file browser where their file is stored. They click
    "edit" and are taken to the editor screen.
2. Student edits the file, saving it frequently and running it. They are aware
    of the "temporary file" icon on the screen.
3. Student is happy that the changes they made are positive. They click "save
    and commit". For novice students, this simply removes the "temporary file"
    icon, and the student can sleep knowing they have a permanent backup to
    revert to.
4. Advanced students may have turned on the "ask for log messages" checkbox in
    their settings. When they click "save and commit", a dialog box appears
    which prompts for a log message. (This is a more advanced feature, after
    students realise the value of leaving themselves short notes for when they
    browse the history).

Note I am thinking of having a little icon space in the editor which shows
either:

* A "unsaved" icon if the user has made changes without saving (like how vim
  writes a "+" in the title bar).
* A "temporary file" icon if the user has saved it without committing.
* No icon if the user has just committed.

Unresolved issues
-----------------

### Disk quota ###
If we are keeping a repository, this means students completely
lack the ability to delete their permanent files (for the purpose of freeing
up space). This means that if students reach their quota, they can't fix it.

Solutions:

* Only put a quota on the head (do not penalise students for large files in
  repo). This could be abused.
* Allow students to request (through an admin) that a file is permanently
  historically erased from the repository.

### Novice mode CPU usage ###

In novice mode there will be signficantly more commits than adept mode because
of the commit each save policy (if we go down that road). Each commit takes up
a little bit of disk space (permanently), but also is problematic because it
takes awhile to commit and many students all using novice mode could overwhelm
the system.