Re^3: Generic RPN Translator available?

[saintmike]

Update: Looks like I spoke too soon:

"sin(ab)"

yields

sin, a, b

instead of the correct

sin, ab

Hmmm ...

Actually, I was looking for exactly the solution browserUK provided. Outstanding job, browserUK! It even works for nested function calls like

"max(min(a,b-c), c, min(d-f,e) ) * atan( pi*4, -1 )"

which gets transformed to

a, b, c, -, min, c, d, f, -, e, min, max, pi, 4, *, -1, atan, *

Now, anyone up for a Parse::Descent grammar, just for kicks :) ?

[BrowserUk]

Fixed. Update: But it still has another bug!..

P:\test>423305 -XLATE
1+2+3
         1, 2, ADD, 3, ADD

a+b+c
         a, b, ADD, c, ADD

abc+def+Efg_hij
         abc, def, ADD, Efg_hij, ADD

2.0+1e-2/0.01E-21
         2.0, 1e-2, ADD, 0.01E-21, DIV

max( a, b, c, d ) * atan( pi*4, -1 )
         a, b, c, d, 4, max, pi, 4, MULT, -1, 2, atan, MULT

sin( cos( x ) - tan( y ) ) + f( g( z ) )
         x, cos, y, tan, SUBT, sin, z, g, f, ADD

2*(somevar+other) + max(this, that)
         2, somevar, other, ADD, MULT, this, that, 2, max, ADD

A+(B*C-D)/E
         A, B, C, MULT, D, SUBT, ADD, E, DIV

(a*(b)-c^(3.4e-2))
         a, b, MULT, c, SUBT, 3.4e-2, POW

5^((-2e-3+x)*sin(p+4.0)/fred)
         5, -2e-3, x, ADD, p, 4.0, ADD, sin, MULT, fred, DIV, POW

sin(a) + sin(ab) + sin( a, b )
         a, sin, ab, sin, ADD, a, b, 2, sin, ADD

Func_1( 1, Func_2( Func3( 1* 2 * 3) * aFunc( 3 ) )+FuNc(4,5,6), -2,e, 
+-10, -2e-10 ) +1
         1, 1, 2, MULT, 3, MULT, Func3, 3, aFunc, MULT, Func_2, 4, 5, 
+6, 3, FuNc, ADD, -2, e, -10, -2e-10, 6, Func_1, 1, ADD

Anyone have a ready source of expressions plus their RPN forms? Or a clever way of verifying them?

Read more... (3 kB)

[BrowserUk]

An updated version of the above with all the bugs I am aware of fixed.

It attempts to verify the results by reversing them. The caveat being that as far as I can see, there is no easy way to put back parenthesis around subexpressions? Hence it will flag expressions that contain parenthesised subexpression as not being correctly.

In every case I have tested, the expression has been reversed correctly except for the replacement of the parens.

It's not a perfect test method, but the best I have come up with. Anyone have a better one?

Results

Read more... (2 kB)

Code:

Read more... (4 kB)

[saintmike]

Hmm, looks like precedence of * and / over + and - isn't handled correctly:

a+b*c/2

should yield

a,b,c,*,2,/,+

but results in

a,b,+,c,*,2,/

instead. I guess we really need grammar, especially if you into consideration that an operator like '**' or '^' needs even higher precedence.