[Setools] [gjallar] Beach 2007 report

[goran at krampe.se]

Hi folks and especially Gjallar hackers! (this post mainly for
participants in "Beach 2007")

Ok, we are off to a good start!

I have focused on trying to get you guys going on one or a few tasks
each - mainly picked in priority - and I think you are all getting into
the codebase and starting to produce - which is not easy on such a short
period of time, I know.

The main priorities for the rest of Beach 2007 are (modulo great
contradicting feedback of course):

1. Robustness and reasonable scalability.
2. A few selected areas of functionality added.
3. Adding a bunch of UIs on top of the model, primarily for
administration.
4. Getting filters much more operational.
5. Offline mechanisms.


Let's go through them one by one:

There are a range of issues to work on for #1 and it involves more
testing and getting Gjallar working in a multi-image setup. There are
today a few "hacks" in the codebase that will break when moving to a
multi-image-Seaside setup which we will need to address. The numbers for
scalability are probably about 500 users, 30 concurrent, 100k cases.

#2 is focused on case linking/spawning, filters of course, builtin
statistics, some historic views, print views and a few other odds and
ends. I am adding a Priority field to bugs.gjallar.se so it will be
clear which ones to work on. This set of functionality is tailored to a
large extent to Micronic needs.

#3 mainly revolves around exposing functionality we already have in the
model through *reasonable* UIs - we don't need to make them perfect. Not
making them perfect means that we shouldn't end up in Ajax timesinks -
Ajax is good, but it also produces side effects. Also, there is no point
at all in making fancy *admin* UIs because there are so few users seeing
them. If we need to spend time on Ajaxification/beautification it should
be done on the UIs that *all users* use on an *everyday basis*.

#4 is on my table and I hope I will get them working good enough during
this period. Filters are the basis for several areas, My View (with its
case panels), View cases (the generic stackable search UI),
Subscriptions (which use filters for fine grained selections), Reports
(which again use filters to make the selection), Statistics (again,
filters for selection) and so on. The model is more or less in place and
the UIs are taking shape. It is rather complex stuff, the filters
compile code that tries to use Magma indexes iin #where:-clauses etc.

#5 is Gjallar's "secret weapon". Good offline functionality that works
is our killer feature. It is important to get it Right. Getting it Right
means getting the txn model beneath clean and verified. This is what
Balazs and Levente are working on and I think/hope they are making good
progress. Another important area we need to address is attachment
handling for offline installations (rsync?) and the mirror selection and
generation - which probably needs revisiting in the face of larger
scales. Also, free text indices needs to be handled.

As you see we have lots of work and the idea is to do these things in
the right priority order since we probably will be forced to drop things
at the end of june. ;)

We are working in parallell and I am keeping track of what everyone are
doing. As you know all our tasks *should* be in bugs.gjallar.se and I
will go through them later today to verify this. I am also adding a
Priority field and will expose it in the tables so that you can sort on
it (stumbled on a bug which prevented that), and possibly also in a
filter.

Of the 5 top areas listed above we are working on all fronts except
scalability. I don't want to change our disposition right now so we will
see the next few days how to address that.

We are also thinking of making an interim release in order to look over
packaging etc so that we can do it reasonably quick in late june. 

Ok, cya on IRC and keep those debuggers warm! :)

regards, Göran