Fixes an error case where a manual TerraSync server is specified; we
would attempt to use an empty string as the OSM2City server, with
hilarious consequences.
Sentry-Id: FLIGHTGEAR-NCZ
Avoid hard-coding the archive extension, and ensure the extracted
archive directory is not orphaned on update. Finally, use the
literal filename in the .dirindex when computing the hash, rather
than adding a .tgz extension.
Ensure all data members are correctly re-initialzied when doing a reset.
This shoed up as negative ‘bytes downloaded’ counts after a TerraSync
abandon and retry.
Extend our own particle manager to replace the OSG particle system
updater. This fixes thread-safety and also timing (better match to
simulation dt values). We also use weak pointers (observer_ptr in
OSG terminology) to ensure particle systems are released once their
frame is gone.
The 'spin' animation has a strange behaviour:
it pushes a temporary rotation matrix on the transformation stack during
cull traversal, and removes it thereafter.
Thus, the rotation matrix is missing outside of cull traversal, e.g.
when the position of Compositor lights is computed.
The same issue was fixed for the 'rotate' animation by commit
e202d4e4a5
(this mentions broken 'picking' animations as a different manifestation
of the same issue).
Fix this by setting the angle of the persistent SGRotateTransform,
instead of creating a temporary rotation matrix.
Previously attempting to start on an ocean tile with static
models present cause FG to spin on "Loading Scenery".
This was caused by not creating a BVH for an ocean tile in
the specific case where there was STG models, due to a PagedLOD
node being inserted in the scene graph. So
FG never thought the scenery was loaded sufficient to place
the aircraft.
By explicitly creating a BVH for ocean tiles the problem is
fixed.
Candidate for 2020.X
Traversing a container which is modified causes crashes, take a copy
during traversal for firePackageStatus.
Sentry-Id: FLIGHTGEAR-CJF
Sentry-Id: FLIGHTGEAR-CJ5
Metar sky condition without base layer height (e.g. FEW///) is the cause of this issue.
Randomize the base layer height when the actual height is unknown.
- scoped locks using lock_guard are better
- because of this the exception handler can also be removed;
although this was originally intended to manage the locks
it never did work properly with C++ exceptions.
Change logic so we create an empty file for such cases, i.e exactly
matching the repository. This simplifies logic in downstream code,
compared with not creating a local file.
Add a test-case to cover this
Modify TerraSync to detect a failure of Airports_archive downloading,
and fall back to file-by-file updating.
Fixes crashes where a request times-out, but then is completed by
UDN sometime afterwards, with a free-d object. Have the DNS::Client own
requests, and be able to retrieve the udns_query to cancel them, in
the timeout case.
Fixes a couple of Sentry reports.
As of 2020-08-01, OpenBSD's system zlib is 1.2.3 which doesn't
have gzoffset(). However we can get away with this by making
gzfilebuf::approxOffset() always return zero.
