Ir al contenido principal

Ralsina.Me — El sitio web de Roberto Alsina

Publicaciones sobre programming (publicaciones antiguas, página 21)

The python spreadsheet: Another look (Traxter DSL)

I apol­o­gize in ad­vance for any ug­ly am­a­teurism in this post. It's my first at­tempt at a do­main spe­cif­ic lan­guage :-)

Yes­ter­day I post­ed about us­ing Py­Cells to write a spread­sheet in Python.

Sad­ly, I can't fig­ure out the prob­lem with my code, and the Py­Cells mail­ing list seems to be pret­ty much dead.

So, I start­ed think­ing... what oth­er ways are to achieve my goal? And de­cid­ed to go me­dieval on this prob­lem.

By that I mean that I will do it the most tra­di­tion­al way pos­si­ble... with a twist.

The tra­di­tion­al way is, of course, to write one or more of lex­er/­parser/in­ter­preter/­com­pil­er for the for­mu­la lan­guage.

Mind you, I don't in­tend to do any­thing com­plete, much less Ex­cel-­com­pat­i­ble (see Ex­cel for­mu­la parsers are hell in this same blog.

So, let's start with a toy lan­guage, sup­port­ing the fol­low­ing:

  • As­sign­­ment to a var­i­able

  • Clas­sic 4-op arith­met­ic­s.

  • Func­­tion calls

  • Cell ranges

That's enough for a toy spread­sheet, and it should be easy to ex­tend.

Here's a de­scrip­tion of the gram­mar for such a lan­guage, writ­ten us­ing Ape­ri­ot [1]:

# This is a simple language for arithmetic expressions


     plus   "+"
     times  "*"
     minus  "-"
     div    "/"
     equal  "="
     colon ":"
     comma ","
     semicolon ";"

     lpar  "("
     rpar  ")"




LIST             -> ASSIGNMENT                : "[$1]"
                  | ASSIGNMENT semicolon LIST : "[$1]+$3"
                  | ASSIGNMENT semicolon : "[$1]"

ASSIGNMENT       -> label equal EXPR : "($1,$3)"

ARGLIST          -> ARG comma ARGLIST : "[$1]+$3"
                  | ARG          : "[$1]"

ARG              -> RANGE       : "$1"
                  | EXPR        : "$1"
                  | label       : "$1"

EXPR             -> TERM              : "$1"
                  | TERM plus EXPR    : "(\'+\',$1,$3)"
                  | TERM minus EXPR   : "(\'-\',$1,$3)"

TERM             -> FACTOR               : "$1"
                  | FACTOR times TERM    : "(\'*\',$1,$3)"
                  | FACTOR div TERM      : "(\'/\',$1,$3)"

FACTOR           -> number           : "$1.val()"
                  | lpar EXPR rpar  : "(\'group\',$2)"
                  | FUNCALL     : "$1"
                  | label               : "$1"
                  | minus FACTOR    : "-$2"

FUNCALL          ->  label lpar ARGLIST rpar : "(\'funcall\',$1,$3)"

RANGE            -> label colon label   : "(\'range\',$1,$3)"

This trans­forms this:


In­to this:

[(<aperiot.lexer.Identifier instance at 0xb7af10ac>,
    <aperiot.lexer.Identifier instance at 0xb7af142c>,
      <aperiot.lexer.Identifier instance at 0xb7af15cc>,
      <aperiot.lexer.Identifier instance at 0xb7af144c>)]),
 (<aperiot.lexer.Identifier instance at 0xb7b4c72c>, ('+', 2, 2))]

Which is sort of a tree with all the ex­pres­sions in pre­fix no­ta­tion in them.

Now, here is the twist: I will "com­pile" this tree in­to.... python code. So I can use eval to do the eval­u­a­tion, just like in the orig­i­nal python spread­sheet recipe.

So this is sort of a pre­pro­ces­sor:

  • The us­er writes ex­cel-­­like for­­mu­las.

  • The spread­­sheet stores python code ob­­tained through com­pi­la­­tion.

  • The spread­­sheet evals the python code.

Of course we have the same prob­lem as usu­al: cell de­pen­den­cies, which is the rea­son why I start­ed play­ing with Py­Cells in the first place!

But... well, here's an­oth­er trick: since I am com­pil­ing, I know when­ev­er there is a vari­able ref­er­enced in the code. And I can re­mem­ber them :-)

So, I can turn this:


In­to this:

[['A1=SUM(a1,a2,a3)*2;', set(['a1', 'a3', 'a2'])],
 ['A3=2+2;', set([])]]

The "com­piled" python code and a de­pen­den­cy set. And voila, this spread­sheet will prop­a­gate cor­rect­ly.

Here's the com­pil­er... in about 60 lines of python [2]. And since the whole point of this lan­guage is to track de­pen­den­cies... let's call it Trax­ter.

Of course, this is a toy right now. But it's a toy with po­ten­tial!

from pprint import pprint
from aperiot.parsergen import build_parser
import aperiot
import cellutils
import sys


def addOp(*args):
        return '+'.join([compile_token(a) for a in args])
def mulOp(*args):
        return '*'.join([compile_token(a) for a in args])
def subOp(*args):
        return '-'.join([compile_token(a) for a in args])
def divOp(*args):
        return '/'.join([compile_token(a) for a in args])

def groupOp(*args):
        return '(%s)'%compile_token(args[0])

def funcOp(*args):
        return '%s(%s)'%(args[0].symbolic_name,
                         ','.join([compile_token(a) for a in args[1]]))

def rangeOp(*args):
        return ','.join([compile_token(a) for a in cellutils.cellrange(c1,c2)])


def compile_token(token):
        if isinstance (token,aperiot.lexer.Identifier):
                return v
        if isinstance(token,list) or isinstance(token,tuple):
            return apply(operators[token[0]],token[1:])
        return str(token)

def compile_assignment(tokens):
        return '%s=%s;'%(target,compiled)

myparser = build_parser('traxter')
pprint (assign_list)

for assignment in assign_list:

print compiled

PyCells: The Python SpreadSheet redux

In 2004 I saw a recipe about how to make a "spread­sheet" in python in 10 lines of code:

class SpreadSheet:
    _cells = {}
    tools = {}
    def __setitem__(self, key, formula):
        self._cells[key] = formula
    def getformula(self, key):
        return self._cells[key]
    def __getitem__(self, key ):
        return eval(self._cells[key],, self)

It's shock­ing. And it work­s, too:

>>> from math import sin, pi
>>>, pi=pi, len=len)
>>> ss = SpreadSheet()
>>> ss['a1'] = '5'
>>> ss['a2'] = 'a1*6'
>>> ss['a3'] = 'a2*7'
>>> ss['a3']
>>> ss['b1'] = 'sin(pi/4)'
>>> ss['b1']
>>> ss.getformula('b1')

I was so awed, I wrote a PyQt ver­sion . Of course there is a catch in that code: it sucks if you are try­ing to write a spread­sheet with it.

Why? Be­cause it does­n't store re­sult­s, but on­ly for­mu­las.

For ex­am­ple:


If you ask for the val­ue of A2, you get 4. If you set A1 to 7, what's the val­ue of A2?

Well, it's noth­ing yet, be­cause it's on­ly cal­cu­lat­ed when you ask for it. But sup­pose you are try­ing to dis­play that sheet... you need to know A2's val­ue changed when you set A1!

That's cell de­pen­den­cies, and while that sim­ple code han­dles them in a way, it to­tal­ly sucks in an­oth­er.

So, I went ahead and cod­ed around it suc­cess­ful­ly. Of course the code was not so pret­ty any­more (although a large part of the ug­ly­ness is just for mak­ing it work with Python 2.3 and rel­a­tive cell­s).

Then yes­ter­day, while look­ing at the ex­cel for­mu­la pars­er mad­ness I saw a ref­er­ence to Py­Cells, a python port of Cells from CLOS.

Here is a blog com­ment­ing on Py­cells:

It ba­si­cal­ly takes the con­cept of a cell in a spread­sheet that get up­dat­ed au­to­mat­i­cal­ly to pro­gram­ming where there are a lot of in­ter­nal da­ta states that are de­pen­dent on one an­oth­er in a chain, or a com­plex graph of de­pen­den­cies. Like, the col­or of a but­ton de­pends on whether you se­lect­ed a ra­dio but­ton or not. Or, shut down the mo­tor if the sen­sor reads above 100 de­grees (ex­am­ple giv­en in tex­t).

Al­most ev­ery­one us­es that anal­o­gy... how­ev­er, no mat­ter how hard I looked, I could­n't find any­one who had ac­tu­al­ly tried writ­ing a spread­sheet us­ing Py­Cell­s! Not even as an ex­am­ple!

So here it is:

import cells

class Cell(cells.Model):
    def value(self,prev):
        print "eval ",self.formula
        return eval(self.formula, {},
    def __init__(self, ss, *args, **kwargs):
        cells.Model.__init__(self, *args, **kwargs)

class ssDict:
        def __init__(self):

        def __getitem__(self,key):

        def __setitem__(self,key,v):
                if not

if __name__ == "__main__":
        ss['a1'] ='5'

        print "a1: ", ss['a1']
        print "a2: ", ss['a2']

        ss['a1'] = '7'
        print "a1: ", ss['a1']
        print "a2: ", ss['a2']

And here you can see it run­ning:

[ralsina@monty cells]$ python
a1:  eval  5
a2:  eval  2*a1
eval  7
eval  2*a1
a1:  7
a2:  14

See how when I set a1 to 7 I get "e­val 7" and "e­val 2*a1"? That's be­cause it's prop­a­gat­ing changes the right way. And that's why this would work as a ba­sis for a spread­sheet.

Excel formula parsers are hell

On 2004 I wrote a spread­sheet in python, which was about a 25KB down­load (com­pressed). It was pret­ty func­tion­al!.

The main prob­lem that would nev­er let that pro­gram be a re­al ap­pli­ca­tion was that it used python as a for­mu­la lan­guage in­stead of the tra­di­tion­al Lo­tus/Ex­cel thing.

Yes­ter­day I de­cid­ed to look at it again, and see if there was a way around it [1]. My main as­set was that I know a fair bit more about parsers now than I did then, and how hard could it be?

Well, it turns out that lan­guage is a whole lot hard­er than I knew!

I found this page about a pars­er

Here's an ex­am­ple from it:

(R[14]C[11] *IF(OR(AND(R[39]C[11]>=55,R[40]C[11]>=20),AND(R[40]C[11]>=20,
R11C3="YES")), R[45]C[11],R[43]C[11])),0))

Yes, that's a valid Ex­cel for­mu­la. No, I have no idea what it does.

And here's some ex­tra nuggets about the lan­guage (quot­ed from the same page):

  • For some rea­­son Ex­­cel us­es a com­­ma as the range union op­er­a­­tor. This means that its on­­ly un­am­bigu­ous use be­tween ranges is out­­­side of a func­­tion call or with­­in a sub­­ex­pres­­sion (i.e., be­tween paren­the­s­es).

  • For some oth­­er rea­­son Ex­­cel us­es a space (or mul­ti­­ple spaces) as the range in­­ter­sec­­tion op­er­a­­tor. While not as am­bigu­ous as the com­­ma, it does re­quire some con­sid­er­a­­tion.

And it goes on, and on, and on....

But hey, he wrote a nice parser!

Just for kick­s, here's the best BNF I could find.

I re­al­ly don't en­vy the KSpread au­thors if they try to be com­pat­i­ble to this garbage!

But why is this lan­guage so gnarly?

  • Be­­cause it has been grow­ing or­­gan­i­­cal­­ly for 30 years?

  • Be­­cause of back­­wards com­­pat­i­­bil­i­­ty?

  • Be­­cause it was nev­er de­signed?

  • All of the above?

I mean, it can't be in­ten­tion­al! Noone ac­tu­al­ly meant it to be like this, right?

Why I use Arch Linux

I have been an Arch Lin­ux for a while now, and I am still lik­ing it.

Here's the good side of it:

  • It's small (one CD)

  • It's sim­­ple (it comes with very lit­tle)

  • It has a de­­cent pack­­age se­lec­­tion (if you con­sid­er AUR, more about that lat­er)

  • It us­es pret­­ty much un­­patched up­­stream soft­­ware

  • It's a bi­­na­ry dis­­tro (ex­­cept for AUR. Again, more about it lat­er)

  • It's pret­­ty sta­ble (no crash­es I can re­mem­ber)

  • It has rolling re­leas­es (un­­like, say, Fe­­do­ra or De­bian)

  • It's easy to keep up­­­dat­ed (like all of them nowa­­days)

  • It's not ide­o­log­i­­cal­­ly dog­­mat­ic, but prag­­mat­ic (yes, there are NVidia driver­s, and test-­­drive games, and what­ev­er)

  • It does­n't seem to be a one-guy joint

And the bad side:

  • Up­­­dates some­­times break things (about twice a year)

  • Ad­min tools are be­tween un­ex­is­­tant and dis­­join­t­ed

And of course, there is the very very good side: AUR

AUR is a co­mu­ni­ty repos­i­to­ry. And there is a rather large com­mu­ni­ty. And pack­ag­ing things for Arch is so easy, and putting things in AUR is so sim­ple, even I find time to con­trib­ute (my pack­ages).

And it's a calm com­mu­ni­ty, and pret­ty much, in­stead of com­pil­ing my ran­dom un­known pack­ages for my­self, I save the steps to build them and stick them in a PKG­BUILD and up­load them. Takes two min­utes for most things.

It's a throw­back to the old days of Lin­ux: qui­et, com­pe­tent (or learn­ing) peo­ple do­ing things, shar­ing, you use them, you give back­... I had not felt that way with a dis­tro for years.

Small software released: RA-WebPass

I just re­leased a wee piece of soft­ware, called RA-Web­Pass which is sim­ply a web­page that you can use to change lin­ux sys­tem pass­word­s.

Ba­si­cal­ly, I wrote it to­day out of my frus­tra­tion with cus­tomers ask­ing me "how can I change the pass­word for the FTP server". Ok, this is how you can.

It's sim­ple and does­n't have too many de­pen­den­cies, so it should be rather se­cure, but don't trust me on that, and you need to run it as root, so be very very care­ful.

RA-Web­Pass home page

On a re­lat­ed note: it's quite sat­is­fy­ing to write some­thing in two hours and just re­lease it :-)

Contents © 2000-2024 Roberto Alsina