Before SimGear commit a962c90b30f36575d01162b64471fa77473237a0,
SGPath::pathListSep was a char in static memory that was not necessarily
followed by '\0'. As a consequence, using &SGPath::pathListSep as a
C-style string could result in a string containing the correct separator
*plus* whatever followed in memory until the first null byte...
SimGear commit a962c90b30 changes this situation by making
SGPath::pathListSep an array of two const chars: the path list separator
followed by a '\0'.
This commit simply adapts FlightGear to this change, which fixes a
couple of bugs where the separator was used, mainly unneeded NavCache
rebuilds due to the "apt.dat", "fix.dat" and "nav.dat" properties in the
SQLite database containing the correct paths separated by a possibly
incorrect separator string (there was no alteration of the cache
contents as far as I can tell, since the db property is only used to
check if the lists of apt.dat, fix.dat and nav.dat files have changed).
Some buggy *.groundnet.xml files (as KSEA currently on TS) define the
pushback hold point for some parking positions as a node on a runway.
In this case, this the pushback hold point for parking
'North_Cargo_Ramp', defined as node 5344 in
Airports/K/S/E/KSEA.groundnet.xml, which is defined twice (second error),
first as:
<node index="5344" lat="N47 27.774559" lon="W122 18.465257" isOnRunway="1" holdPointType="PushBack" />
and then as:
<node index="5344" lat="N47 27.725747" lon="W122 18.159649" isOnRunway="1" holdPointType="PushBack" />
(due to code in flightgear/src/Airports/dynamicloader.cxx, it should be
the second one that wins, which is not on a runway but on apron in the
north cargo area)
As a consequence, when this gate is selected for an AI aircraft, the
pushback route has only one node (since the pushback hold point is then
the closest point to itself supposedly on runway!), and the
corresponding FGTaxiRoute instance has an empty 'routes' member
variable, which FGTaxiRoute::next() doesn't handle gracefully
(segfault).
It may be that an additional check/change could be desirable in
FGTaxiRoute::next() in such a case (one node and obviously no route in
the FGTaxiRoute instance), however I'm not sure how Durk wants this case
to be handled, since FGTaxiRoute::next() seems to iterate on nodes.
This fixes the bug reported at:
https://forum.flightgear.org/viewtopic.php?p=308397#p308397 and
https://sourceforge.net/p/flightgear/mailman/message/35776552/
Thanks to yanfiz and wkitty42 for the report, and to gooneybird for
inspecting the groundnet file.
This is because FGTaxiNode::ident() is generally (always?) an empty
string for FGTaxiNode instances. This concerns the:
unreferenced groundnet node: ...
warning. Also remove one tiny use of boost.
We now show paths in ‘view command line’ and set them through the
standard mechanism. Re-ordering the paths also notifies the rest of
the system correctly.
Code and tests to demonstrate migrating of older auto-save files, with
blacklisting support to exclude properties. Disabled pending agreement
on the required blacklisting values.
Some pieces of code such as fgMainInit() and, by cascading effect,
fgInitHome(), were careful to return a meaningful value indicating
success or error, however the main() function in src/Main/bootstrap.cxx
ignored it royally so far.
main() now returns:
- EXIT_FAILURE if fgMainInit() or fgviewerMain() throws an exception;
- whatever said function returns otherwise.
- Rename fatalMessageBox() to fatalMessageBoxWithoutExit(). This should
prevent the kind of bug that prompted this set of changes: someone
calling fatalMessageBox(), assuming the program would stop at that
point, whereas in reality it did not.
- Add new function fatalMessageBoxThenExit(). This is not vital of
course, but allows one to spare one line here and there and to apply
the DRY principle for such fatal exits.
- Replace every existing call to fatalMessageBox() with one or the other
of the two new functions. Improve formatting along the way. This
fixes a few bugs of the kind explained above.
This reverts commit 9e6a3ebc6b ("Make
fatalMessageBox() end with std::abort() and declare it [[noreturn]]").
After reflexion, it seems better to let fatalMessageBox() return,
because there is existing code that appears to be relying on this aspect
to do some work after having called fatalMessageBox() (cf. main() in
bootstrap.cxx). Also, the way of exiting from fatalMessageBox() after
commit 9e6a3ebc6b (std::abort()) was probably too brutal for a
controlled exit---as opposed to a terminate handler.
Basically, this is because fatalMessageBox() is only safe to call from
the GUI thread, however it seems fg_terminate() can be called from any
thread (according to C++11 semantics). Additionally, fatalMessageBox()
typically requires some work to happen in the GUI thread (event loop) in
order to display something, but we can't realistically expect this while
running a terminate handler just before the program dies.
See messages around
<https://sourceforge.net/p/flightgear/mailman/message/35775803/> for a
discussion of this subject.
+ Minor header cleanup (<locale.h> replaced with <clocale>, etc.)
