I was able to hack a bit at Stupid­Sheet this morn­ing, and there are some re­al ad­vances.

In fac­t, it's a bare­ly func­tion­al spread­sheet al­ready! Hel­l, it has at least one fea­ture OOcalc lack­s, and one both OOcalc and KSpread missed.

Check it out in this video (9.5M­B):

Sor­ry, dead file

Sor­ry if the au­dio is out of sync and/or too low. And you prob­a­bly can't stream it, you need to down­load it first. And maybe it will 404 on you. If that's the case, wait a few min­utes and in­sist.

For my spread­sheet pro­jec­t, I had to re­do some­thing I had for­got­ten about: cell dis­place­men­t. I did that once when the for­mu­la lan­guage was python.

At the time, I parsed the python us­ing the to­k­enize mod­ule and Ka-Ping Yee's re­gur­gi­tate.

Python->­To­ken­s->Dis­place cell­s->re­gur­gi­tate->Python

Since I have lots of oth­er things to do, I de­cid­ed to do it the same way, and wrote the equiv­a­lent of re­gur­gi­tate for Trax­ter, my for­mu­la lan­guage.

It turns out it was not re­al­ly hard, but I had to re­do parts of the pars­er so it kept more in­for­ma­tion about the source.

Af­ter that it was sim­ple, you see, Trax­ter com­piles to Python. That means all I had to do was an­oth­er (very sim­i­lar) back­end, and there it is, a Trax­ter-­to-­Trax­ter com­pil­er. Or de­com­pil­er. Or some­thing. And rel­a­tive cell ref­er­ences are work­ing again (and now with the right syn­tax).

It promis­es:

• Na­­tive look&feel for your apps on all ma­jor plat­­forms

• Works for graph­i­­cal and ter­mi­­nal apps

• Open Source

• Low over­­head

• Uni­ver­sal bi­­na­ries that work on Win­­dows, Mac and Lin­ux

Seems too good to be true, but it is sup­posed to be re­leased next month.

• The we­b­site

• The press re­lease

• The videos

• The whitepa­per (which reads most­­ly like an ad)

Af­ter hack­ing for about two hours the cell de­pen­den­cies yes­ter­day us­ing dict­s, I found my­self say­ing "how can I check if the de­pen­den­cy graph is cycli­cal?"

And of course it hit me. That's the de­pen­den­cy graph. Use a graph li­brary!

I looked for about 2 min­utes be­fore find­ing one that's prob­a­bly not op­ti­mal, but is damn sim­ple and in the pub­lic do­main: graph_lib.py by Nathan Den­ny.

First in­ter­est­ing da­ta point, the first two lines on the file (yes, I found out there is a lat­er ver­sion):

#--Version 1.0.0
#--Nathan Denny, May 27, 1999

Yup. In two days this piece of code is turn­ing 8 years old and un­touched. But it works just fine and dandy!

Here's the piece of code that makes the en­gine run which has grown from a hum­ble 10 LOC to al­most a whoop­ing 40 LOC:

class SpreadSheet(QtCore.QObject):
    _cells = {}
    tools = {}
    graph=Graph()

    def __init__(self,parent):
        QtCore.QObject.__init__(self,parent)
        for name in dir(functions):
                self.tools[name]=eval('functions.'+name)


    def __setitem__(self, key, formula):
        key=key.lower()
        c=traxcompile('%s=%s;'%(key,formula))
        self._cells[key] = [c[key][0],False,compile(c[key][0],"Formula for %s"%key,'eval')]

        # Dependency graph
        if not self.graph.has_node(key):
                self.graph.add_node(key)
        for edge in self.graph.in_arcs(key):
                self.graph.delete_edge(edge)
        for cell in c[key][1]:
                self.graph.add_edge(cell,key)
        try:
                print 'GRAPH(TOPO): ',self.graph.topological_sort()
                self._cells[key][1]=False
                print 'GRAPH(BFS) : ',self.graph.bfs(key)
                for cell in self.graph.bfs(key)[1:]:
                        self.emit(QtCore.SIGNAL('changed'),(cell))
        except Graph_topological_error:
                # We made the graph cyclic
                # So, mark this cell as evil
                self._cells[key][1]=True
                # And remove all incoming edges to go back to
                # status quo
                for edge in self.graph.in_arcs(key):
                        self.graph.delete_edge(edge)

    def getformula(self, key):
        key=key.lower()
        return self._cells[key][0]
    def __getitem__(self, key ):
        print self._cells[key]
        if self._cells[key][1]:
                return "ERROR: cyclic dependency"
        else:
                print 'evaluating [%s]: '%key,type(self._cells[key][0]),self._cells[key][0]
              return eval(self._cells[key][0], self.tools, self)

This en­gine sup­port­s:

• Us­er de­fined func­­tions (just add them in for­­mu­la.py)

• Cell de­pen­­den­­cies with de­tec­­tion of ci­­cles, of any length

• Un­lim­it­ed size spread­­sheets

• No­ti­­fi­­ca­­tion of cell changes to oth­­er mod­­ules

• Au­­to­­mat­ic re­­cal­cu­la­­tion

• A cus­­tom for­­mu­la lan­guage (ok, that's not in this piece of code ;-)

The whole project is now 1167 LOC, of which 591 are gen­er­at­ed or graph_lib, which means ev­ery­thing in­clud­ing the trax­ter "com­pil­er" and the UI is less than 600 lines of code.

Not too shab­by.

My goal is a func­tion­al spread­sheet with a work­ing GUI and sup­port­ing the most com­mon func­tions and pieces of the for­mu­la lan­guage in .... 2000 LOC, and I am will­ing to work on it for about two week­s.

Let's see how it turns out. Of course any­one with 5 free min­utes can con­trib­ute his favourite spread­sheet func­tion (I al­ready have sum ;-)

Adding some­thing else to my plate is prob­a­bly not a very good idea, but what the heck, I can make it sleep an­oth­er three years if I lose in­ter­est.

So: I am writ­ing a re­al spread­sheet in python.

Prob­a­bly nev­er go­ing to be use­ful for cor­po­ra­tions, but it should be at least as fea­ture­ful as Google's and it should be amaz­ing­ly smal­l.

Here are the com­po­nents:

• Trax­ter: my spread­­sheet-­­for­­mu­la-­­like-lan­guage with de­pen­­den­­cy track­­ing that com­piles to python.

• PyQt (hey, it has a grid wid­get)

• Python (Of course)

The sta­tus right now:

• It's al­­most as func­­tion­al as it was 2.5 years ago

• Ex­­cept for bro­ken rel­a­­tive cel­l­s.

• But with the be­gin­n­ing of a re­al for­­mu­la lan­guage.

• With au­­to­­mat­ic re­­cal­cu­la­­tion and cyclic de­pen­­den­­cy check­­s.

• Adding things is dead sim­­ple. Here's the im­­ple­­men­­ta­­tion of SUM (not up­­load­­ed yet, though):

def sum(*args):
  ac=0
  for v in args:
    ac+=v
  return ac

All the range stuff hap­pens be­hind the scenes (although you may get a func­tion called with thou­sands of args... I won­der how well Python han­dles that).

You can check it at the google code project (Use the SVN).