[mkws-moved-to-github.git] / notes / developers.txt

Notes for developers

These notes are collected by Heikki, mostly from skype chats with Wolfram
and Mike. I collected them for my own use, but I hope they will turn out
to be helpful to anyone who needs to get started with mkws.


Libraries
---------

* We are using jquery as a browser indepent layer to access the dom, so we
don't have to worry about IE bugs.  Wolfram looked why we are using
jquery.json-2.4.js ... it turns out we needed it because the standard functions
in IE8 are broken.

* jasmine is a test framework (mkws dev). you will not use jasmine in a
production site.  the nice thing with the jasmine test framework is that it
will work with any browser. I can start a virtual machine with IE8, open the
test page, wait 3 seconds for success and shutdown windows.

* handlebar is a template engine



Include files
-------------

The whitepaper says to include mkws-complete.js. This file is made by concatenating
a number of files (see Makefile). For us developers, it is easier to include the
raw files, as in

    <script type="text/javascript" src="tools/htdocs/jquery-1.10.0.min.js"></script>
    <script type="text/javascript" src="tools/htdocs/pz2.js"></script>
    <script type="text/javascript" src="tools/htdocs/handlebars-v1.1.2.js"></script>
    <script type="text/javascript" src="tools/htdocs/jquery.json-2.4.js"></script>
    <script type="text/javascript" src="tools/htdocs/mkws.js"></script>

You can also include the css directly in your test page:
    <style type="text/css">
      #mkwsTermlists div.facet {
      float:left;
      width: 30%;
      margin: 0.3em;
      }
      #mkwsStat {
      text-align: right;
      }
    </style>


Most (all?) code work happens in mkws.js.


Unit tests
----------

if you want understand the test than you can look at mkws/test/spec/mkws-config.js
and mkws/test/spec/mkws-pazpar2.js . See also mkws/test/README.txt


Structure of mkws.js
--------------------
(This will soon be out of date, but should provide some kind of starting
point even then)

First page is just helper functions for the Handlebars template library, which we
use to generate some of the HTML output. (Down the line, we will use this more
heavilty -- right now it's only used for records).

Then we define the mkws object, which contains all global state -- which hopefully
there is not much of. It is one of only two objects we place in the global namespace:
the other is mkws_config, which is a hash supplied by the application if it wants
to override default configs.

Next is a very short function defined to allow us to publish and subscribe events.
That is not yet used: shifting much of the code to do so is a big part of what I
am working on right now.

Next, a very short stanza of code that just makes sure mkws_config is defined:
simple applications won't bother to define it at all since they override node
of the defaults.

Next, a factory method for making widget objects. At present this is trivial
because we are only now starting to need a representation of individual widgets
as JS objects. More of the functionality will get moved into these objects over
the next week.

Next, a factory method for making widget-team objects. This is where all the
awesomeness is at the moment. A team is a bunch of widgets that work together
to achieve a common goal, e.g. the search-box, search-button and results-pane
widgets.

HTML elements are defined as belonging to the same team if they have an
mkwsTeam_NAME class for the same NAME. You can have multiple teams (as in
two-teams.html that I linked to earlier) which are completely independent of
each other.

I guess you're familiar with the JS idiom where the factory function for a kind
of object also acts as a namespace where all the object's member-variables live,
invisible to the outside world? That's what we do here. All the member variables
have names of the form m_NAME.

Now I sugges you skip over all the team-object code for now -- we'll return to it
later. For now, page down to "// wrapper to call team() after page load" which is
the next thing after the end of that function (or class, if you like).

You're familiar with this JS idiom?
  (function() { code ... })();
 Runs the code immediately, but within its own namespace. That's what we do for
all the remaining code in mkws.js. In this case, we pass the jQuery object into
that namespace under the name `j' for reasons that are frankly opaque to me.

There's still a few places in the code where oddities live on, either from jsdemo
or from work Wolfram's done, where I don't really understand why it's that way
but I'm scared to change it. In this case, IIRC, it's something to do with
protecting our copy of the jQuery object, or something.

Aaanyway, within that namespaced area, where's what we do.

First, we set up the mkws.debug() function, which just logs to the console in a
form that doesn't explode IE. I have plans for this function, make it understand
debugging levels a bit like log4j or maybe more like yaz-log where there are
named logging types that are not in a sequence.

(You will notice that the teams have a debug() function which delegates to this
but adds some other useful team-specific stuff.)

 Next up: the utility function mkws.handle_node_with_team(). We use a LOT of nodes
that have their team-name in a class (as in "mkwsTeam_NAME" outlined above).
All the utility does is parse out that team-name, and the widget-type, from the
classes, and pass them through to the callback.

mkws.resize_page() does what it says. Gets called when window-size changes, and
allows us to move the facers to the side or the bottom as the screen is wide or
narrow (e.g. when you turn your iPad 90 degrees)

(Aside: I thought we'd have to iterate over all teams to move their facet lists
but it turns out we don't: jQuery just Does The Right Thing if you call
   $(".mkwsTermlistContainer1").hide();
or similar and there are multiple hits.)

Next come a bunch of JS functions that are invoked from the MKWS UIs --
swithching between target and record views, stepping through pages of results,
etc. All of these are team-specific, but the global code in the HTML can't
invoke a team's member function directly. So these stub functions just invoke
the relevant member of the appropriate team.

default_mkws_config() fills in the mkws_config structure from hardwired defaults.
This is the wrong way round: instead, whenever we want to find a config value, we
should default our way up a tree, starting with the individual widget's config,
falling back to the team's config if the widget doesn't define that value, then
the global config, and finally the default. I'll make that change once widget
objects are fully real.

authenticate_session() authenticates onto the SP when we're using it (rather
than raw pp2). It's a bit sellotape-and-string, to be honest, just does a wget.
It would be better if this was supported by pz2.js

run_auto_searches() is what makes pages like
  http://example.indexdata.com/auto3.html
work. THere are two places it's invoked from. Either directly when all the HTML
has been set up if we're using raw pp2; or when SP authentication has been
completed if we're using that. As with the UI functions, it just delegates down
into the teams.

Finally, code that runs when the page has finished loading -- this is really
the main() function

The first thing it does is patch up the HTML so that old-style MKWS apps work
using new-style elements. This is the code you just fixed.

Straight after that, more fixup: elements that have an mkws* class but no
team are given an extra class kwsTeam_AUTO. This is the ONLY thing that's special
about the team "AUTO" -- it has no other privileges.

 Very near the end now: we walk through all nodes with an mkws* class, and create
the team and widget objects using the factories we described earlier. Jason is
worried this will be slow, hence the instrumentation. It's not :-)

Last of all: start things off! SP-auth if used, otherwise straight to the
auto-searches.


OK, want to plough into the team object?

... bearing in mind that some of this should be moved out of the team into the
top-level code, and some other parts will be moved down into individual widgets
once we have them.

OK. So we start with a bunch of member variables for state. Many of them will be
hauntingly familiar to anyone who's worked on jsdemo :-)

A new one is m_debug_time, which is a structure used for keeping track of elapsed
time. It's nice: it lets debug messages for each time note how long it's been
since the last message in that same team, which means you can see how slow-ass
our network operations can be. That's implemented in debug(), which is the very
next thing in the file.

Then a bunch of code to do with setting initial values from defaults.

The stuff about languages is a great example of code that should be at the top
level, not in the team: it deals only with global objects, yet gets run n times
if there are n teams. (I am adding a ###-comment to this effect right now.)

Then we make the pz2 object, which is our only channel of communication with the
server outside of the HTTP GET hack in authenticate_session(). The callback
functions are all closures with access to this team's member variables. Are you
somewhat familiar with pz2.js already?

Well, it's all driven by callbacks. You register them at object-creation time;
then later when you call (say) m_paz.search(), it will invoke your my_onsearch()
function when the search-response arrives.

There are some oddities in the order things get done. For example, is m_perpage
set AFTER this object is created, rather than at the same time as its stablemate
m_sort up above? No reason. So plenty of tedious, error-prone, cleaning up of
this sort to be done.

Then come the pz2 callbacks. my_onshow() is an interesting representative. The
team-name is passed back by pz2 (which has used it as the SP window-ID to keep
sessions separate). I used to think it needed to be passed back like this so
the invoked callback functions could know what team they're being called for,
but now I realise they don't need to, because they're closures that can see
m_teamName. Oh well.

Indeed, you can see that my_onshow() doesn't even USE the teamName parameter.

Anyway, the point is to find specific elements that belong to the right team,
so they can be populated with relevant data. We do that for the pager, and of
course the record-list.

The meat of the record-list is filled in by invocations of renderSummary(),
which loads a Handlebars template and uses that to generate the HTML. Needless
to say, loadTemplate() caches compiled templates.

my_onstat() and my_onterm() do similar things -- you can figure out the details.

add_single_facet(), target_filtered() and others are uninteresting utility
functions.

my_onrecord() and my_onbytarget() are more of the same. There is some nastiness
in my_onrecord() to handle poping up a full-record div in the right place and
getting rid of any old ones. It doesn't work quite right.

onFormSubmitEventHandler() is more interesting. This is a JS event handle (for
form-submit, duh) but how does it know what team to work for? It checks `this'
to see that classes the HTML element has, and so finds the relevant mkwsTeam_NAME
class. Then it can fire newSearch() on the relevant team. But it now occurs to
me that this too is a closure so it doesn't need to do any of that. It can just
start the search in its own team.

[Why can that simplifying change be made here but not in, say, switchView()?
Because the former is assigned to a DOM object from within the JS code, so acts
as a closure; but the latter is invoked by a fragment of JS text which the user
clicks on, when there is no context.]

An oddity of newSearch(): it's not defined as
    function newSearch()
like most of the other member functions, but
    that.newSearch = function()
That's because, unlike most other member function, it gets explicitly invoked
on a team object:
    mkws.teams[tname].newSearch(val);

But in fact that won't be necessary once I fix the invoker
(onFormSubmitEventHandler) to be aware of its own context, so that can simplify,
too.

The next interesting method is triggerSearch(). You can see that it assembles
the pp2filter and pp2limit values for the search invocation by walking the
array m_filters[], which is built by click on facets.

That's done in a slightly clumsy way. I might make a Filters object at some
point with some nice clean methods.

BTW., more unnecessary context-jockying with the windowid parameter. I don't
need it, I have m_teamName.

loadSelect() is another fine example of a method that appears in a random
place. It's just one of the HTML-generation helpers.

Now we come to a bunch of externally invoked functions, that.limitTarget() etc.
These are the meat that are called by the stubs in the top-level code -- remember
those one-liners?

They change state in various ways based on the user's clicks. The first four
({un,}limit{Query,Target}()) do so by tweaking the m_filters[] array.

More HTML-generation helpers: redraw_navi(), drawPager() -- note the inconsistent
multi-word naming scheme

We are *completely* schizophrenic over whether to use camelCase or
underscore_separated

Then more UI functions (that.X, that.Y)

Anyway, onwards ... loadTemplate() you already know about.

defaultTemplate() is the hardwired defaults, used for applications that don't
define their own templates. For an application that does, see
http://example.indexdata.com/lolcat.html

As you can imagine, Lynn was WAY impressed by lolcat.html

mkws_html_all() is a big ugly function that generates a buttload of HTML.
Basically, it spots where you've used a magic name like <class="mkwsSearch">
and expands it to contain the relevant HTML.

Then it's just utility functions: parseQueryString() to read URL params,
mkws_set_lang() and friends generate more HTML.

All these HTML-generation functions should of course be together. Many of them
should also use Handlebars templates, so that clever applications can redefine
them. In places, too, they still need fixing to use CSS classes instead of inline
markup, <b> and suchlike.

that.run_auto_search() is interesting. It gets the term to search for from the
"autosearch" attribute on the relevant element, but there are special forms for
that string. !param!NAME gets the query from the URL parameter called NAME;
!path!NUM gets it from the NUM'th last component of the URL path. There may be
more of these in future.

Once it's got that it's a pretty straightforward invocation of newSearch()

M(string) yields the translation of string into the currently selected language.
We use it a lot in the HTML generation, and it's one part of that process that's
more cumbersome in Handlebars. The problem is that M() is a closure so it could
in principle know what the language of THIS team is, and it could be different
for different teams, even though that's not the case at the moment. But Handlebars
helpers are set once for all time, so they can't be team-specific, which means
they can only refer to the globally selected lan

And that, really, is the end of the team object (and so of mkws.js). TA-DAH!