The main bottleneck was in EquationTags::erase(), which had quadratic
complexity in the number of equations. This commit optimizes it to n·log(n)
complexity.

Ref. dynare#1960

Commit 8c021084 had removed the possibility of
using “residual” as a variable name. In particular, this broke the
lmmcp/rbc.mod test.

By the way, also improve the check for when the pac_expectation is used two
times on the same model.

Implement the fix that is already present in assignment operator (in particular
for dynare#1782).

Ref. dynare#1949

Use simulation of the model, based on a first order approximation, as
an initial guess for the solver in each period.

However do not set that option permanently (and also disable RemoveSemicolon by
default), since the clang-format manual states that these options can lead to
incorrect formatting and thus their result should be carefully reviewed.

The regression was introduced in commit
84d792bc, when leads/lags were allowed on
model-local variables. The treatment of model-local variables without
leads/lags was negatively impacted. This commit fixes the regression by
restoring the former behaviour for model-local variables without leads/lags.

Reported on https://forum.dynare.org/t/pre-processing-takes-very-long/26865

Ref. dynare#1929

NB: the “overwrite” option only works for the same type of “learnt_in” value,
i.e. an integer-valued “learnt_in” cannot replace a date-valued “learnt_in” and
vice versa.

New options {first,last}_simulation_period to perfect_foresight{,with_expectation_errors}_setup commands

Allow non_zero option in steady command

See merge request !109

Automatically detected by clang-tidy using modernize-use-equals-default check.

Automatically detected using clang-tidy with modernize-use-nodiscard check.

– enable new options InsertNewlineAtEOF and RemoveSemicolon
– reformat the code accordingly

[skip ci]

Dates handling inside Dynare statements was done in the Flex scanner using a
hack that would introduce some extra characters in the character stream (in
order to return a call to the “dates('…')” constructor to Bison). But this
would break the column location counter, because the latter would count
characters that do not exist in the user file.

This commit fixes it by implementing a more natural handling of dates, adding
the “dates('…')” constructor directly at the level of the semantic value.

– Harmonize location output for warnings so that it is the same as for
  errors (i.e. more explicit about what is the line number and the column number)
– Remove obsolete stuff for delayed error messages that should have been
  removed in db1f6c29.
– Factorize some printing logic with ParsingDriver::undeclared_model_variable_error().
– Simplify WarningConsolidation class

Also do some coding style manual enforcement.

The call to model_error() would return and lead to an incorrect dereferencing
of the optional<int> rv.

The call to model_error() should have been replaced by a call to error() in
db1f6c29. Then the move to an optional<int> in
46b2c735 introduced the bug.

See https://forum.dynare.org/t/preprocessing-failed-for-specific-mod-file/26705