Code generation fails for a simple model

[Lennart Ochel]

Code generation fails for the following simple model:

model Unnamed
  block activationCon
    parameter input Integer nIn;
    parameter input Integer nOut;
    input Real tIn[:];
    input Real tOut[:];
    input Integer tIntIn[:];
    input Integer tIntOut[:];
    input Integer arcType[:];
    input Real arcWeightIn[:];
    input Real arcWeightOut[:];
    input Integer arcWeightIntIn[:];
    input Integer arcWeightIntOut[:];
    input Real minTokens[:];
    input Real maxTokens[:];
    input Integer minTokensInt[:];
    input Integer maxTokensInt[:];
    input Boolean firingCon;
    input Boolean fed[:];
    input Boolean emptied[:];
    input Boolean disPlaceIn[:];
    input Boolean disPlaceOut[:];
    input Real testValue[:];
    input Integer testValueInt[:];
    input Integer normalArc[:];
    input Boolean testChange[:];
    output Boolean active;
    output Boolean weaklyInputActiveVec[nIn];
    output Boolean weaklyOutputActiveVec[nOut];
  algorithm 
    active:=true;
    weaklyInputActiveVec:=fill(false, nIn);
    weaklyOutputActiveVec:=fill(false, nOut);
    
    for i in 1:nIn loop
      if disPlaceIn[i] then
        if arcType[i]==1 and not (tIntIn[i]-arcWeightIntIn[i]  >= minTokensInt[i]) then
          active:=false;
        elseif arcType[i]==2 and not (tIntIn[i] > testValueInt[i]) then
          active:=false;
        elseif arcType[i]==3 and not (tIntIn[i] < testValueInt[i]) then
          active:=false;
        end if;
      else
        if arcType[i]==1 or normalArc[i]==2 then
           if not (tIn[i]>minTokens[i] or (tIn[i]<=minTokens[i] and fed[i])) then
              active:=false;
           elseif tIn[i]<=minTokens[i] and fed[i] then
              weaklyInputActiveVec[i]:=true;
           end if;
        end if;
        if arcType[i]==2 then
            if not (tIn[i] > testValue[i]) then
              active:=false;
            end if;
             if testChange[i] and fed[i] and normalArc[i]==2 then
               weaklyInputActiveVec[i]:=true;
             end if;
        elseif arcType[i]==3 and not (tIn[i] < testValue[i]) then 
          active:=false;
        end if;
      end if;
    end for;
    
    for i in 1:nOut loop
       if disPlaceOut[i] then
         if not (tIntOut[i]+arcWeightIntOut[i]<=maxTokensInt[i]) then
          active:=false;
         end if;
       else
        if not (tOut[i]<maxTokens[i] or (tOut[i]>=maxTokens[i] and emptied[i])) then
          active:=false;
        elseif tOut[i]>=maxTokens[i] and emptied[i] then
          weaklyOutputActiveVec[i]:=true;
        end if;
       end if;
    end for;
    active:=active and firingCon;
    weaklyOutputActiveVec:=weaklyOutputActiveVec and fill(firingCon,nOut);
    weaklyInputActiveVec:=weaklyInputActiveVec and fill(firingCon,nIn);
  end activationCon;

  activationCon activation(nIn=0,
                           nOut=0,
                           tIn=fill(0, 0),
                           tOut=fill(0, 0),
                           tIntIn=fill(0, 0),
                           tIntOut=fill(0, 0),
                           arcType=fill(0, 0),
                           arcWeightIn=fill(0, 0),
                           arcWeightOut=fill(0, 0),
                           arcWeightIntIn=fill(0, 0),
                           arcWeightIntOut=fill(0, 0),
                           minTokens=fill(0, 0),
                           maxTokens=fill(0, 0),
                           minTokensInt=fill(0, 0),
                           maxTokensInt=fill(0, 0),
                           firingCon=false,
                           fed=fill(false, 0),
                           emptied=fill(false, 0),
                           disPlaceIn=fill(false, 0),
                           disPlaceOut=fill(false, 0),
                           testValue=fill(0, 0),
                           testValueInt=fill(0, 0),
                           normalArc=fill(0, 0),
                           testChange=fill(false, 0));
end Unnamed;

Error message:

Unnamed.c: In function 'eqFunction_4':
Unnamed.c:354: error: '$Pactivation$PweaklyInputActiveVec' undeclared (first use in this function)
Unnamed.c:354: error: (Each undeclared identifier is reported only once
Unnamed.c:354: error: for each function it appears in.)
Unnamed.c:532: error: '$Pactivation$PweaklyOutputActiveVec' undeclared (first use in this function)
mingw32-make: *** [Unnamed.o] Error 1
mingw32-make: *** Waiting for unfinished jobs....
Unnamed_06inz.c: In function 'eqFunction_2':
Unnamed_06inz.c:340: error: '$Pactivation$PweaklyInputActiveVec' undeclared (first use in this function)
Unnamed_06inz.c:340: error: (Each undeclared identifier is reported only once
Unnamed_06inz.c:340: error: for each function it appears in.)
Unnamed_06inz.c:518: error: '$Pactivation$PweaklyOutputActiveVec' undeclared (first use in this function)
mingw32-make: *** [Unnamed_06inz.o] Error 1
Compilation process exited with code 2

[Adrian Pop]

Note that when array sizes are zero some things start to disappear from the DAE.
So this might not actually be a codegen issue. Have a look at the flattened code.

[Willi Braun]

Yes, Adrian is right, the corresponding variable is removed from the DAE.
Somewhere we need to analyse the algorithm section, and detect that the for loop doesn't do anything, so it should removed. The question is where we want to do that, I guess not in code generation phase.

[Adrian Pop]

The flattened code of the model looks like:

class Unnamed
  parameter input Integer activation.nIn = 0;
  parameter input Integer activation.nOut = 0;
  input Boolean activation.firingCon = false;
  output Boolean activation.active;
equation
  activation.tIn = {};
  activation.tOut = {};
  activation.tIntIn = {};
  activation.tIntOut = {};
  activation.arcType = {};
  activation.arcWeightIn = {};
  activation.arcWeightOut = {};
  activation.arcWeightIntIn = {};
  activation.arcWeightIntOut = {};
  activation.minTokens = {};
  activation.maxTokens = {};
  activation.minTokensInt = {};
  activation.maxTokensInt = {};
  activation.fed = {};
  activation.emptied = {};
  activation.disPlaceIn = {};
  activation.disPlaceOut = {};
  activation.testValue = {};
  activation.testValueInt = {};
  activation.normalArc = {};
  activation.testChange = {};
algorithm
  activation.active := true;
  for i in 1:activation.nIn loop
    if {}[i] then
      if {}[i] == 1 and not 0 >= {}[i] then
        activation.active := false;
      elseif {}[i] == 2 and not {}[i] > {}[i] then
        activation.active := false;
      elseif {}[i] == 3 and not {}[i] < {}[i] then
        activation.active := false;
      end if;
    else
      if {}[i] == 1 or {}[i] == 2 then
        if not ({}[i] > {}[i] or {}[i] <= {}[i] and {}[i]) then
          activation.active := false;
        elseif {}[i] <= {}[i] and {}[i] then
          activation.weaklyInputActiveVec[i] := true;
        end if;
      end if;
      if {}[i] == 2 then
        if not {}[i] > {}[i] then
          activation.active := false;
        end if;
        if {}[i] and {}[i] == 2 then
          activation.weaklyInputActiveVec[i] := true;
        end if;
      elseif {}[i] == 3 and not {}[i] < {}[i] then
        activation.active := false;
      end if;
    end if;
  end for;
  for i in 1:activation.nOut loop
    if {}[i] then
      if not 2 * {}[i] <= {}[i] then
        activation.active := false;
      end if;
    else
      if not ({}[i] < {}[i] or {}[i] >= {}[i] and {}[i]) then
        activation.active := false;
      elseif {}[i] >= {}[i] and {}[i] then
        activation.weaklyOutputActiveVec[i] := true;
      end if;
    end if;
  end for;
  activation.active := activation.active and activation.firingCon;
end Unnamed;

[Adrian Pop]

The question is if we should remove the for loops which are from 1 to 0 during front-end.
The problem is that they are based on parameters, but also the arrays we're removing are based on parameters.

I think that if we remove arrays of 0 size we should also remove for-loops of 0 size.

[Adrian Pop]

I'll have a look at this, as it seems to be a front-end issue.
We either should not remove zero size arrays from the DAE or remove everything that is zero sized, including for and while loops.

[Adrian Pop]

Fixed in r18612.

[Lennart Ochel]

Thanks for that fix! Thus, much more PNlib tests are working (see r18613).