This reverts commit fbc42b1930, which is
incorrect for the currently used version of Bison (in stretch).
Besides, even if the patch seems to work with recent versions of
Bison, it introduces hard to predict and probably very unpleasant
properties (this patch allows to use the `expectation` word anywhere,
this would be a problem if the user choose to name a variable or
parameter as `expectation`).
Backward incompatible change: the temporary terms for 3rd order are now stored
in "temporary_terms_third_derivative" (without the final "s"; same for external
functions), for consistency with the name of the slot for the derivatives
themselves ("third_derivative").
Ref dynare#217
Does not work for Julia mode, neither with block and bytecode.
Note: in DLL mode, the number of temporary terms in no longer given in the
‘ntt’ symbol; it must be computed using ‘M_.dynamic_tmp_nbr’.
Ref dynare#217
Before this commit, a dynare call such as
```
dynare <<mod file>> -Db=“A” -Da=@{b}
```
would not expand the value of `b` in `a` whereas
```
dynare <<mod file>> -Da=“A” -Db=@{a}
```
would expand the value of `a` into `b` because the arguments were stored in a map which printed the `@#define` statements in the .mod file in alphabetic order.
Here, emplace() and insert() are not interchangeable because this is a std::set of
std::vector<int>, and we really want the initializer-list constructor of
std::vector (and not the two int's constructor).
Fixes the test deterministic_simulations/lola_solve_one_boundary.mod
In particular, no need to put both upper and lower case letters in the
patterns, since the lexer is case-insensitive.
The only simplification that we do not perform is removing the quoting of
single quotes (i.e. replacing \' by '), since it confuses the C++ mode of
Emacs.
We must provide a return value for
VarExpectationNode::maxLagWithDiffsExpanded().
Return zero, which will lead to the right overall result even if it is not
strictly the right choice (see the comment in the code).
This feature is ill-designed and no longer needed by the GUI. And is not very
useful: it is not possible to interact with the preprocessor without using the
filesystem, since the preprocessor creates many files anyways.
If we really need to reimplement such a feature, it should rather be redesigned
by reading the modfile from standard input (cin). That could be triggered by
using "-" as the filename argument (as is already done by several standard Unix
utilities).
The M_.params structure was not correctly passed to
dynamic_set_auxiliary_series.m, leading to crashes if a parameter was needed to
compute an auxiliary variable (typically an auxiliayr lead variable
corresponding to a non-linear term in stochastic setup).
Since 38152c34a4, the M_.endo_histval variable is
generated via dynamic_set_auxiliary_dseries.m. But the latter file does not
handle auxiliary variables for Lagrange multiplier. We therefore now set them
by hand (at an initial value of zero).
Since 38152c34a4, the M_.endo_histval variable is
generated via dynamic_set_auxiliary_dseries.m. The problem is that, for
auxiliary variables corresponding to a lead, this will generate a NaN in
M_.endo_histval. This is conceptually correct, since such variables are never
used as initial conditions, but this is inconsistent with what we do with the
"initval" block, and leads to crashes in some routines where we fail when there
is a NaN. Therefore, replace these with a zero, as it used to be.
The idea is to make use of the dynamic_set_auxiliary_dseries.m file to generate
the initial conditions for all auxiliary variables, including the diffs.
Also remove the check done by the preprocessor for the lags in histval, since
it does not work correctly with the diff operator.
- ExprNode::maxLag() and ExprNode::maxLead() now take into account exogenous
deterministic variables, for consistency with M_.maximum_{lead,lag}
- ExprNode::maxLag() no longer behaves as if diff() operators were
expanded (i.e. it now returns 1 on diff(x(-1))), for consistency with
maxEndoLag() and maxExoLag()
- New ExprNode::maxLagWithDiffsExpanded() method, that behaves as maxLag() used
to behave (except that it also takes exogenous deterministic into account)
Previously, this function was counting the total number of diff() operators in
an expression. But this is not very useful, and is potentially misleading,
because in practice we use this function to compute the maximum lag on
variables in levels.
This function now returns the maximum number of nested diffs.
For example, on diff(x)+diff(diff(y)), this function was returning 3, and it
now returns 2.
The engine is now more robust and should reject any expression that does not
conform to the expected form. It is also able to deal with more cases, such as
terms appearing with a minus sign, or variables in the middle of a
three-factors product.
BTW, use a std::tuple for storing the result of the matching inside
PacExpectationNode, and change the order of components within the
structure (variable first, scalar last).
The preprocessor now writes all the symmetric elements in the "hp"
matrix (derivatives of the hessian w.r.t. parameters), for consistency with all
other derivatives output.
Previously it was only writing one of the two symmetric elements, when indices
of endogenous were different.
Also, no longer compute two times symmetric elements in derivation w.r.t.
parameters at order 2, for consistency with derivation w.r.t. endogenous.
It is therefore now necessary to duplicate them in the output to keep behavior
unchanged.
When creating the sparse matrix (in MATLAB or C mode), since storage is in
column-major order, output the first column, then the second, then the third.
This gives a significant performance boost in use_dll mode (at both compilation
and runtime), because it facilitates memory accesses and expression reusage.
On Windows/MinGW, use the -static linking flag. This enforces static linking of
Boost libraries, which is needed on MSYS2.
Also, use AM_CXXFLAGS and AM_LDFLAGS variables for changing flags, since
CXXFLAGS and LDFLAGS are user variables.
New options "mexext" and "matlabroot" are introduced, so that the preprocessor
knows where to find MATLAB and which architecture to compile for.
Only recent gcc is now supported. A set of optimization flags is used so that
compilation goes reasonably fast on large models.
Consequently, options "msvc", "mingw" and "cygwin" have been removed.
For correlations, there is one extra columns (since there are two symbol
indices). MATLAB automatically adds the missing column, while Octave is
stricter when verifying dimensions.
By the way, add a missing comma.
... when an initial value is given to a parameter that is not
estimated.
The generated driver file was crashing with cryptic error message
because we were searching in the first (and second) column of a
potentially empty array with 0 columns. The fix is to initialize the
fields of estimated_params_ with empty arrays with 10 columns (ie
zeros(0, 10)). Also print a message in the matlab command window if
parameter declared in estimated_params_init is not estimated.
Since, in this case, there are less equations than endogenous, the
variable_reordered structure was not filled with enough element, leading to an
invalid memory read when printing M_.state_var.
Ref #637
In particular, it is necessary to turn back DataTree::AddVariable() into a
non-virtual method, because it is called from DataTree's constructor. Enforcing
the absence of leads/lags is now done using a new boolean DataTree::is_static.
Take advantage of the new copy constructor for handling
PlannerObjectiveStatement more elegantly.
Unfortunately it is not possible to implement *move* constructor / assigment
operators, because the reference ExprNode::datatree is not mutable.
This facilitates switching variable types on the fly. In particular, this
allows removing the hack in DynamicModel::updateAfterVariableChange() that way
basically recreating all the nodes after the type change.
The check is done at the time the preprocessor is compiled, but it ought to be
done at the time when the C source file generated by the preprocessor is compiled…
Anyways, it is probably irrelevant now, since MSVC has supported
acosh/asinh/atanh for a long time now.
This was not conceptually an enum, but rather a collection of unrelated
constants:
- two constants for use as placeholder for symbol IDs at different places
- one constant for the default number of arguments
This mimicks the structure of M-functions (though the logic for filling the
temporary terms vector is a bit different).
This change implied a modification in the way we compute the checksum in case
of block decomposition (the temporary terms for the C output are not correctly
computed in that case).
Due to a limitation of the current implementation, this breaks syntaxes like
[ (i,j) ] (but not [ (2,j) ]; the problem only occurs when an array is
constructed by specifying as first element a tuple whose first element is a
variable name). Solving this problem requires an overhaul of the
macro-processor, with construction of ASTs at parsing time, and evaluation
later on (instead of doing on-the-fly evaluation).
Ref #5
- BTW, store them in a std::vector rather than std::list
- incidentally, fix issue in VariableNode::removeTrendLeadLag where expression
sharing was possibly violated when creating a new VariableNode
- == and != have now lower priority than <= < >= >, for consistency with the
Dynare modelling language (and incidentally C and C++, but not Julia).
- ^ has now higher priority and no associativity, for consistency with the
Dynare modelling language (and usual arithmetic notation).
- & has now higher priority than |, and both have lower priority than + and -,
but higher than inequality comparators. This is not the same as C and C++ (in
which & and | are just above && and ||), but this allows for expressions such
as "a|b == c" to have their most natural semantics (i.e. this will compare
the union of a and b with c; the C/C++ priority level would have resulted in
a type error).
Ref #5.