1) Using divU instead of fvc::absolute(phi,U) in TEqn as the latter uses latest time meshPhi which is inconsistent
2) Adding fvc::interpolate(U) when topo changes
3) in pEq for compressible dgdt is updated using the latest rho1 and rho2 after compressible effects are considered
- In the corner case with few faces or points, the normal List I/O
results in a compact list representation.
This is less than desirable for external programs with simple
line-based parsers.
- Write exactly the following
*Faces*
// Patch: <word-Region> <word-Patch>
<int-nFaces>
(
<int-faceSize>(<int> .. <int>)
...
)
*Points*
// Patch: <word-Region> <word-Patch>
<int-nPoints>
(
(<float-x> <float-y> <float-z>)
...
)
STYLE: only use serial form of createExternalCoupledPatchGeometry in tutorial
- less confusing for the user, who wonders why it is being done twice.
- Also fixed bug noted in issue #269
- Previous implementation had all faces together, which made
it difficult (impossible) for external applications to
figure out which geometry was being referred to.
- Provide separate region/patches as follows:
// Patch: <regionName> <patchName>
For example,
// Group: coupleGroup
// Patch: heater minY
8( ... )
The region-name is always present, even if there is only one region.
- This change is a partial reversion to the behaviour in 2.4.x, except
that we can now also handle multi-region geometries.
Changing the leading comment from "# " to "// " facilitates parsing
of the files with OpenFOAM itself if necessary.
- Cleanup/centralize handling of -decomposeParDict by relocating
common code into argList. Ensures that all processes receive
identical information about the -decomposeParDict opton.
- Only use alternative decomposeParDict for simpleFoam/motorBike
tutorial so that this will be included in the test loop for snappy.
- Added Mattijs' fix for surfaceRedistributePar.
- Can occur with some user names, or mounted paths.
Resolve by using '?' for the separation character.
Since '?' is a shell-glob, it is highly unlikely to occur appear in
filenames. Additionally, it is not a meta-character in standard sed,
nor in the GNU extension (which uses '\?').
changed flag which caused infinite while loop. Background info:
- findCellZoneTopo tries to find for all named surface intersections
which side of the face is in the faceZone
- i.e. it tries to make the cellZone consistent with the faceZone
(to fix small problems)
- this had some logic to assign the neighbour cellZone to the owner cellZone
- which didn't check for the neighbour being the same value as the owner
- but still set a 'changed' flag which caused the loop to never end.
All of the access methods for autoPtr include validity checks and will
fail if the underlying point is NULL. In some cases, however, we'd
like to retain the automatic deletion mechanism, but still address a
nullptr. This is mostly for cases in which a file-stream should be
allocated, but only on the master process. For these cases we'd still
like to pass through and reference the underlying pointer (eg, to
obtain the correct method call) without tripping the pointer check
mechanism. If we attempt to use the ptr() method, the autoPtr memory
management is bypassed and we risk memory leaks.
Instead provide an alternative mechanism to obtain the raw underlying
pointers/references. Use rawPtr() and rawRef() for these potentially
useful, but also potentially dangerous, operations.
- Normally use '()' to deference. This has extra safety and issues a
fatal error if the underlying pointer is not valid.
However, in some cases we are happy with getting a null reference.
The refOrNull() method returns the reference without any checking.
Usage example:
autoPtr<OFstream> osPtr;
if (Pstream::master())
{
osPtr.reset(new OFstream(...));
}
writeViaMaster(osPtr.refOrNull());
- The writeViaMaster() call takes an OFstream reference,
but this is only used directly on the master.
The slaves will pass things through to the master.
- Less looping when detecting lagrangian clouds and their fields.
- Avoid using Time::setTime() and IOobjectList in tight loops.
They both kill performance immensely.
ENH: provide a -noLagrangian option to foamToEnsight and foamToEnsightParts
for even more control.
- The new field needs initialization with a dimensioned<Type> not just
the dimensionSet.
- The new field was also incorrectly being registered, which could
cause issues later.
Old code:
Found 10990 time steps
Search for moving mesh ... no moving mesh detected.
Startup in 329.09 s
Updated:
Found 10990 time steps
Search for moving mesh ... no moving mesh detected.
Startup in 1.6 s
- Cause was checking "polyMesh/points" via an IOobject.
Short-circuit with a check for a polyMesh/ directory first.
Limit the check to the master-node as well to further reduce
load on the file-system.
------------------------------
ENH: improve per-step conversion times for foamToEnsight.
Old code:
Converting 11001 time steps
Time [0] = 0 Wrote in 1.53 s
Time [1] = 1 Wrote in 1.52 s
...
Time [100] = 100 Elapsed time 205.35 s
Updated:
Converting 11001 time steps
Time [0] = 0 Wrote in 1.4 s
Time [1] = 1 Wrote in 0.07 s
...
Time [100] = 100 Elapsed time 42.4 s
- Speedup by hashing test results from the first conversion step
instead of checking each time.
Check data on all nodes to avoid problems with incomplete writes.
------------------------------
BUG: moving mesh detection failed for foamToEnsightParts
- adjusted to agree with updated foamToEnsight
------------------------------
Note:
- foamToEnsightParts (serial) still has about twice the throughput of
foamToEnsight.
