without this, the restart writer will trigger a segmentation fault
error.
And also, some minor refactoring and function development so that the
simulator can check whether there are valid connections defined for
aquifers with certain id.
This commit switches to constructing Box instances from a GridDims
and two call-back functions instead of taking an EclipseGrid
directly. The two call-back functions are a predicate for active
cells and a translation from Cartesian to active cell indices
respectively.
This is intended to simplify working with nested boxes, such as
those that occur for local grid refinement.
This commit switches the Cartesian -> IJK decomposition in GridDims
from K->J->I into the I->J->K order. The latter is simpler.
While here, also consistently use 'std::size_t' instead of 'size_t'.
The constructor did implicitly assume that no cell in a numerical
aquifer would be declared inactive in ACTNUM, and this restriction
still applies. Moving numerical aquifer property and connection
processing into the constructor body enables using existing member
functions for non-neighbouring connections and removes a branch that
obscured the fact that we should have ACTNUM be as close to
finalised as possible before creating analytic aquifer connections.
Commit eca858335 introduced a more comprehensive way of forming the
initial ACTNUM array used when constructing the EclipseGrid from a
corner-point description. This commit applies that strategy to all
known grid types. As an additional benefit this will also cover one
special case that was not handled and would effectively discard the
work done in the generalised initial ACTNUM construction.
Add unit tests to exercise the ACTNUM handling for other grid types.
This commit introduces a new helper template class,
FlatTableWithCopy<RecordType>
which wraps a vector<RecordType>, provides constructors from
DeckKeyword and initializer_list<RecordType> and handles copying
tables for all-defaulted records.
We reimplement the PVTW, DENSITY, and GRAVITY tables in terms of
this helper class to reduce duplication. If the copy behaviour is
generally useful/needed we expect to replace the existing FlatTable
mechanism with the new helper.
Suggested by: Markus Blatt
A simulation model may choose to give PVTW data as
PVTW
1.0 1.0 1.0e-5 0.2 0.0 /
/ -- record 2 (copied from record 1)
if, for instance, the oil and/or gas tables are different in regions
1 and 2, but the water is the same. In this case we must properly
copy record 1 into record 2 and essentially recreate the table.
To this end, decouple the 'PvtwTable' from the 'FlatTable' machinery
and make the former into an independent type containing vector<>
instead of inheriting from vector<>. Implement the default->copy
behaviour in the new PvtwTable::PvtwTable(const DeckKeyword&)
constructor.
Faults are stored in an OrderedMap with the full string used during
construction even if that is longer than 8 characters. When later
these faults are e.g. modified whith MULTFLT using only the first 8
characters, the only message the user got was
```
An error occurred while creating the reservoir properties
Internal error: Key not found:
Unrecoverable errors while loading input: Key not found
```
With this patch we at least add some more context to the error
message:
```
Error:
An error occurred while creating the reservoir properties
Internal error: [/../EclipseState.cpp:362] Could not set fault transmissibility multiplier 5.23166e-10 for fault CFN2_05t: Key ABCD_05q not found. Similar entries are ABCD_05q_extended_width, UABCD_05q_extended_width.
```
