bad code generation for arrays

[Lennart Ochel]

Bad simulation code gets generated for the example bug_2271_b from our testsuite.
location: /testsuite/simulation/modelica/algorithms_functions/bug_2271.mos

error message

Simulation execution failed for model: bug_2271_b
array s->index[0][1] == 4 incorrect, a->dim_size[0] == 3
bug_2271_b: util/real_array.c:322: indexed_assign_real_array: Assertion `index_spec_fit_base_array(dest_spec, dest)' failed.
Aborted (core dumped)

[Adrian Pop]

Lennart, there is no model in directory:
testsuite/simulation/modelica/algorithms_functions/
that starts with b*. Have you forgot to add it?

[Adrian Pop]

Sorry, I'm stupid, I did not do svn up :)

[Henning Kiel]

x[1:n]:=fill(0.0,n);

this creates a temporary array t1[2] and makes an (indexed) assign from the fill() to t1 and then copies t1 verbatim to x
=> index starts with "1". The copy procedure is problematic if t1 and x have more than one dimension and dimensions differ.

x[n+1:M]:=fill(1.0,N-n);

this creates a temporary array t2[3] and makes an indexed assign from the fill() to t2 and then copies t2 verbatim to x
=> index starts with 3 and is thus too large for t2
=> later t2 is copied verbatim to x which (a) does not use the index spec and (b) overwrites the content copied from t1 earlier.

temporary arrays should not be filled with index.
When copying the temporary arrays to the destination x, the index should be considered.

template indexedAssign(DAE.Type ty, String exp, DAE.ComponentRef cr,
  String ispec, Context context, Text &varDecls)
::=
  let type = expTypeArray(ty)
  let cref = contextArrayCref(cr, context)
  match context
  case FUNCTION_CONTEXT(__) then
    'indexed_assign_<%type%>(&<%exp%>, &<%cref%>, &<%ispec%>);'
  case PARALLEL_FUNCTION_CONTEXT(__) then
    'indexed_assign_<%type%>(&<%exp%>, &<%cref%>, &<%ispec%>);'
  else
    let tmp = tempDecl("real_array", &varDecls)
    <<
    alloc_<%type%>(&<%tmp%>, 1, size_of_dimension_<%type%>(<%exp%>, 1));
    indexed_assign_<%type%>(&<%exp%>, &<%tmp%>, &<%ispec%>);
    copy_<%type%>_data_mem(&<%tmp%>, &<%cref%>);
    >>
end indexedAssign;

looks like the else part is not correct here, but the other line is

[Lennart Ochel]

This blocks the Petri net library.

[Martin Sjölund]

Postponed until 1.9.1

[Mahder Alemseged Gebremedhin]

=> index starts with 3 and is thus too large for t2

Yes. that was the problem with the assertion failing. Dimensions of the temporary array should be equal to dimensions of the original array NOT the slice.

=> later t2 is copied verbatim to x which (a) does not use the index spec and (b) overwrites the content copied from t1 earlier.

This is also fixed now because we allocate the temporary array itself by copying from the original array. Thereby keeping any previous changes intact.

Fixed in r17741.