Regarding the Runway selection bug:
The logic here is a bit convoluted, but I also had a dumb bug in normaliseBearing - I was clamping to the wrong range (0..360 instead of -180..180). This caused the scoring code to pick weird runways. I've added some extra cases to my local tests, and here's a fix.
Trivial patch, but an important milestone:
Convert FGAirport to inherit FGPositioned. This concludes the first phase of the FGPositioned changes, and hopefully the most intrusive ones - adding in the base class. There's lots (and lots) of further work to do on the indexing and querying side, as well as cleaning up the accessors, but that will happen in single source files, or a group of related files at a time.
As a trivial note, this patch does fix a bug where the very last airport in apt.dat would get an invalid type. So for all you people who just love to fly to EHYB (Ypenburg, The Hague), things may work a little more sanely.
I'll intentionally let the dust settle after this patch, so any weird behaviour I may potentially have introduced shows up. Just to re-iterate, so far there should be absolutely no user-visible change in the behaviour of anything - navaids, position init, the route manager, AI flight plans, etc. If there is, please let me know and I'll fix it ASAP.
By way of example, here's a patch to make the position init code (in fg_init.cxx) cleaner, partly thanks to the FGPositioned changes. It reduces the file size by 200 lines - virtually all of which was copy-and-paste. Once the remaining class (FGAirport) is converted to inherit FGPositioned, all the future patches should be like this - touching one or a few files at most.
This factors the start-offset logic out into a helper, which also does the final property setting (which has to happen on both the /preset and 'real' values). Using the accessors in FGPositioned, and the offset helper, a couple of cases become trivial (fix and nav) and others become much simpler.
Convert FGNavRecord to inherit FGPositioned. This is much more self-contained than the FGRunway change, since FGNavRecord already had good encapsulation of its state. However, it's a large diff due to moving around two nasty pieces of code - the 'align navaid with extended runway centerline' logic and the 'penalise transmitters at the opposite runway end' logic.
In general things are more readable because I've replaced the Navaid type enum, and the use of Robin's integer type codes, with switches on the FGPositioned::Type code - no more trying to recall that '6' is an outer marker in Robin's data. The creation code path is also pushed down from navdb into navrecord itself.
Convert FGRunway to be heap-based, and inherit FGPositioned. This is a large, ugly change, since FGRunway was essentially a plain struct, with no accessors or abstraction. This change adds various helpers and accessors to FGRunway, but doesn't change many places to use them - that will be a follow up series of patches. It's still a large patch, but outside of FGAirport and FGRunway, mostly mechanical search-and-replace.
An interesting part of this change is that reciprocal runways now exist as independent objects, rather than being created on the fly by the search methods. This simplifies some pieces of code that search for and iterate runways. For users who only want one 'end' of a runway, the new 'isReciprocal' predicate allows them to ignore the 'other' end. Current the only user of this is the 'ground-radar' ATC feature. If we had data on which runways are truly 'single-ended', it would now be trivial to use this in the airport loader to *not* create the reciprocal.
Here's part 2 - converting FGFix (the simplest one) to be both heap-based and inherit FGPositioned. One minor benefit from this is replacing some dangerous code in FGFixList which used to return the address of an iterator member ('&it->second'). To keep the diff a sensible size, I'm not updating the callers to use the richer FGPositioned types - i.e replacing separate lat/lon handling with SGGeod. I will make those cleanups, but in future patches.
If someone could kindly apply the attached patch, that'll keep this from crashing, I believe. The fix is easy since FGAirport can now always provide an active runway - there's no need to guess at random, or rely on the tower having set one up.
Small patch fixing bugs I've encountered while getting the current CVS to build in MSVC.
* std::lower_bound was used with the key-type of a map, but lower_bound expects the value-type of the collection it works on, with is std::pair. MSVC seems to be more strict about this.
* Added an missing include statement.
* Replaced an rint() call with floor() (MSVC does not offer rint).
Here is a little patch that changes the behaviour of the VOR CDI and OFF-flag
for indicators like the HSI when getting outside the range of the VOR
station.
Currently, when flying at a distance between the effective_range and twice the
effective_range of a VOR station, the in-range property is computed based on
a random value, causing the OFF Flag and the CDI bar to perform an ugly
jitter.
The attached patch introduces a new property signal-quality-norm which is
computed based on the distance to the station and the range. It is 1.0 when
the distance is less than the range and decreases by 1/x^2 for distances
greater than the range leading to a signal-quality-norm of 0.25 for distances
two times the range, 0.125 for three times the range and so on.
The in-range flag is tied to a signal-quality-norm greater than 0.2 (fixed
squelch).
The CDI and GS needle deflection is multiplied with the signal-quality-norm.
The benefit is:
- Ability to animate the OFF-Flag with a smooth transition.
- CDI and GS needle deflection shows correct values when in range
(signal-quality-norm=1.0) and show some wrong indication when the range is
exceeded
- CDI and GS needle start to move, even when the OFF flag is visible
- No more jitter for flag and needles
See the new SenecaII ki525a hsi as an example at
http://www.t3r.de/fg/navpatch.jpg
The numbers on the image are:
(1) the new property signal-quality-norm
(2) distance exceeds the effective-range by 30%
(3) NAV flag has a rotation animation bound to signal-quality-norm and is
partially visible
(4) CDI is partially deflected even with NAV flag shown
This implementation better matches reality - at least, how I observed it ;-)
Good news: I'm working on some automatic testing of the 'core' FG pieces, especially those I'm likely to break in my Navaids / airports / runways work
Bad news: I already broke something, in my runways refactoring. (But my tests caught it!)
Attached patch fixes it - it's (of course) the stupidest thing in the world. Incidentally, standardising this kind of code into some (inlined) header is becoming more and more of a priority for me - I've lost count of the number of times I've seen the 'clamp heading to 0..360.0' and 'reverse a heading and clamp it' idioms in the code. The KLN89 and MkVIII code have (of course) their own helpers for this.
This is a little intrusive on the KLN89 code, but avoids the wasteful cloning of the airports, runways and navaids which current happens, and also combines the ugly string ordering code.
- removes various members from FGRunway which no-one was using
- any of these can be trivially re-instated if and when someone
actually wants to use them - but right now they're simply bloating up
FGRunway, which we have lots of, because it currently includes all the
taxiways in Robin's data.
- that's it.
- Runways are now part of an airport, instead of a separate list
- Runways are no longer represented as a boring struct, but as a class
of their own.
-Improved runway access to unify various runway access methods.
use <enabled>false</enabled> flag for widgets that shouldn't be drawn
instead of <hide>true</hide>. This is consistent with other places
in fgfs, like menu entries, hud elements, subsystem switches, etc.
Attached patch + new file make FGNavRecord have a .cxx file, and a constructor w
hich allows all the parameters to be supplied. Along the way I also cleaned up t
he navrecord.hxx header, lots more header pollution has been killed.
Some long methods are no longer inline, but were all suspiciously long to meet c
ompiler inlining criteria (I'm not clear if the 'inline' keyword is advisory or
mandatory in this situation) - I don't expect this to affect performance in any
way whatsoever.
The constructor addition is to support some hacking I'm doing improving the star
tup performance of the navDB by lazily loading the data, and caching it in a mor
e efficient format than text. I'm submitting this change (and probably some othe
r small tweaks in the future) since they are worthwhile as cleanups regardless o
f how my current experiments work out.
CameraGroup supports cameras opened in different windows or in the
same window, with flexible view and perspective specification.
Clean up mouse click handling via osgViewer. Don't let any camera
"have focus;" this forces events to be reported in window coordinates
and simplifies life. Don't use the osgViewer::View::requestWarpPointer
method; instead use our own code which only deals with window
coordinates.
Make glut and sdl versions work with CameraGroup too.
I thought that this would fix the "black hole in the sky" problem,
which turned out to be caused by an OpenSceneGraph bug. Nevertheless
it is a simplification.
* experimental clean-up / reduction on two of the FG headers:
(I'm going to await feedback on the developers list before doing more of
these, to avoiding going over files multiple times, but in principle it
seems pretty straightforward.)
* final fixes for SG_USING_STD removal
- this exposed a bizarre issue on Mac where dragging in <AGL/agl.h> in
extensions.hxx was pulling in all of Carbon to the global namespace
- very scary. As a result, I now need to explicitly include CoreFoundation
in fg_init.cxx.
- change SG_USING_STD(x) to using std::x
SimGear change. It changes all the SG_xxxx to be the 'real' includes, and gets
rid of many #ifdef SG_HAVE_STD_INCLUDES. As an added bonus, rather than
replacing 'SG_USING_NAMESPACE(std)' with 'using namespace std', I just fixed
the small number of places to use std:: explicitly. So we're no longer polluting
the global namespace with the entire contents of std, in many cases.
There is one more 'mechanical' change to come - getting rid of SG_USING_STD(X),
but I want to keep that separate from everything else. (There's another
mechnical change, replacing <math.h> with <cmath> and so on *everywhere*, but
one step at a time)
PLETE_FUNCTIONAL from SimGear and FlightGear.
As a result, SG_HAVE_STD_INCLUDES is now *always* set, so I will get the boring
fixes for that done, but separately. I'm still auditing the other things in comp
ilers.h - there's a lot that can die now BORLAND is gone.
- remove the OSX_BUNDLE crap *I* introduced years ago - we're always a a bun
dle on Mac now.
- fix up the default fg-root on Mac to be FlightGear.app/Contents/Resources/
data - i.e the location used by the macflightgear.org distro, and indeed the obv
ious 'correct' location. Not sure why I didn't use that in the first place, back
in the day.
- remove the CPSForegroundEnable hack. For one thing, we're a bundle and don
't need it, and for another, osgViewer on Mac does the same logic using a newer,
public API rather than a hack into the OS.
- remove the strange logic for doing fgOSInit 'early' (in bootstrap rather t
han main) when running from the command line on Mac; again this is obsolete, and
no one seems to know why it was ever necessary. I guess it was an interaction w
ith SDL when running without a bundle.
- (not Mac related) remove obsolete code bracketed by ENABLE_PLIB_JOYSTICK a
nd USE_GLIDE (neither of which are ever set, even from config.h that I can see)
in main.cxx
Various other patches that have been lingering around for a while:
* Moved trafficcontrol.[ch]xx from the Airports directory to ATC, where
it really belongs.
* AI aircraft will request startup clearance, and ground control will
approve.
* Starting AI Aircraft will be pushed back to a predefined holding point
on the ground network, and wait a while before taxiing out to the runway
- drop unused parts (MouseQuat/GuiQuat)
- move "old-reinit-dialog" fgcommand to fg_command.cxx under new name
"reset" for now. (May later get merged with fgcommand "reinit".)
- move reInit() to fg_init.cxx: This was used by Shift-Esc and
Menu->File-Reset (via fgcommand "old-reinit-dialog"). We have already
a similar function fgReInitSubsystems() in fg_init.cxx, so these two
functions will probably get merged later.)
where chosen to match the SGPropertyNode::WRITE/READ symbol names and
should make it more consistent, but this isn't really something that
should be exposed to the Nasal developer.)
we know the stack trace), it's better to use a separate function for
it, as this can be used conditionally, as in:
setlistener("/the/property", func(n) {
if (n.getValue() < 0) # who's writing negative numbers to /the/property?!
abort(); # let's check the backtrace ...
});
and its function called. This is useful for debugging purposes, to get
a core dump which unveils which C++ code was writing to the property.
This isn't any more of a security problem as other Nasal code that
makes fgfs crash, such as: var x = func call(x); x(); The difference
is that the listener doesn't use up all memory before and is much quicker.
- don't allow to do that from any XML file. This is to prevent malign
code from writing a new fg-home in ~/.fgfs/autosave.xml or other files
in ~/.fgfs/.
After the changes that moved the GUI and HUD to a slave camera, the
texture-based fonts wouldn't display. The main fixes here are making
sure that the TXF textures are all loaded into the font cache early,
and explicitly setting the active texture unit in the GUI / HUD
drawImplementation.
- don't keep oodles of class member variables public (eek!)
- use ::hasValue() instead of ::getType() != SGPropertyNode::NONE
- consistency fixes and cosmetics
Switch to defining PU_USE_NONE and providing our own callback
functions to pui for "get window" and "get window size." A new
WindowSystemAdapter class assigns ID numbers to windows for the
purpose of identifying them to plib; the window size can be extracted
from the osg::GraphicsContext class in all the different
implementations (osgViewer, glut, sdl).
Implement a GraphicsContextOperation that runs code in a particular
graphics context, perhaps in another thread, and provides an
isFinished() method to test if the operation has finished. This allows
us to initialize plib PUI properly if there are multiple graphics
contexts without using fgMakeCurrent(). fgMakeCurrent() can't work in
multi-threaded OSG configurations.
Eliminate fgMakeCurrent() and all its uses, either by using
GraphicsContextOperation or by seeing that it is not necessary.
Attach the GUI camera as a slave camera.
Don't manipulate the OSG state in the drawImplementation() functions
for SGHUDAndPanelDrawable and SGPuDrawable; it's not needed.
Attached patch adds support for multiple FIXes with the same name.
Applies to both branches.
New functionality is in query_and_offset, which now returns the FIX
closest to the passed in location.
Updated route manager to take advantage of this.
Otherwise, query functions return an unspecified member from the set
of identically named FIXes. (This was previously the one occurring
last in the database file, but I don't think anybody counted on that.)
- it's unused
- it's hardcoded (we already got rid of all other hardcoded dialogs)
- we'll have to consider switching to osgWidget, which is planned to
get imported into osg. It's unlikely that we add *more* of PLIB
in the meantime.
"the [...] patch (for fg) increases the time before invisible geometry (i.e.
geometry not inside the view frustrum) gets deleted. default was 10 seconds
which i am increasing to 2 minutes."
on the '*'-key, but allow to cycle it by setting
/sim/rendering/on-screen-statistics to "true"
- move that function to the Debug menu (no more key assigned!)
- add "print-statistics" menu entry
under some circumstances. The history of this patch is somewhat unclear,
but was brought to my attention by Martin Spott, while preparing for the
Lelystad FSWeekend show. See also my posting on FlightGear devel, on
November 22, 2007 "(Multiplayer Local Echo Patch)", but wasn't committed
then because I/we assumed that had been superseded by other code
modifications. The local echo problem still persists, however, albeit
under specific circumstances. The current patch reportedly prevents this
from happening.
and the pushback code:
- Traffic manager starts modelload requests immediately upon program
loading
- Only create legs 1 (push back) or five (cruise) of AI traffic.
- AIAircraft's rather obsessive behavior to circle around a waypoint is
largely resolved
- More realistic ground steering for AI aircraft.
Thanks to Vivian Meazza for debugging this. The output deque for
FGDigitalFilter was not being kept long enough for the
doubleExponential filter. Reads from output[1] could cause a crash.
From Till:
i started the project at the end of february with a simple idea: move all
3d-model loading to the DatabasePager-thread. my first attempts looked
promising, though they were a little too optimistic (or naive?). the patch
has evolved a lot since.
currently it does the following things:
1. revive SGModelLib, move functions for xml-model-loading there
2. replace all calls to sgLoad3dModel with calls to either
SGModelLib::loadModel() or SGModelLib::loadPagedModel()
almost all models will be loaded by the DatabasePager. the few exceptions are:
your own plane, shared models in scenery, random objects, AIBallistic models.
3. simplify mode-loading functions (avoid passing around fg_root)
4. avoid supurious MatrixTransform nodes in loaded models
5. fix some memory leaks
Introduce a convenience member function in wxRadarBg for creating an
SGPropertyNode and initializing it in one step. Use this in
agRadar. This eliminates buggy behavior when the necessary radar
properties aren't defined.
$ fgfs --version
2.0pre-20080314
FG_ROOT=/usr/local/share/FlightGear
FG_HOME=/home/foo/.fgfs
... assuming that VERSION in config.h is "2.0pre-20080314". Which it isn't.
Instead it's "1.0.0" since yesterday, and was "0.9.10" before that. This
has yet to be discussed, and I'll add the option to options.xml later.
Reindent to Stroustrup style.
Make FGPrecipitationMgr an SGSubsystem and remove all references to it
in main.cxx and renderer.cxx.
Use SGGeod::makeZUpFrame instead of a private function in
tileentry.cxx. Rewrite that function, WorldCoordinate, to use
makeZUpFrame too.
Makes it possible to start at a parking location defined in the AI/Airports/*/parking.xml files, using the parkpos command line option.
Note that the name to pass is the concatenation of the "name" and "number" fields in the xml.
This allows Nasal managed views thanks to the proximity of the
event manager (which executes Nasal loops). The io manager must
come after the view manager to avoid jitter in network replay.
As discussed with Tim on irc: Here is a quick fix for the memory-hungry tile
manager. Due to bugs in FGNewCache, old tiles were never deleted.
I left the timestamp-updates in the cull-traversal. but imho things work
just as well when timestamps are updated in FGNewCache::insert_tile() and
FGNewCache::get_tile()
Lee has added two new filter types, "gain" and "reciprocal". These filters can
read their gain factor from a property. In the process we also added minimum
and maximum output clamps that are applicable to all filters.
I added the ability to configure adaptive controllers i.e. the controller gain
can be tied to a property, so that it can be changed at runtime. This
requires a change in the xml structure of the autopilot configuration file:
<Kp>
<prop>/autopilot/KAP140/settings/ROL/Kp</prop>
<value>0.10</value>
</Kp> <!-- proportional gain -->
The old method <Kp>0.10</Kp> still works so as to not break all existing
autopilots, but it will output a warning to use the new method.
Please find attatched a new version of xmlauto.cxx.
Remove the call to build() in reinit(). This prevents build() from being
called twice when Reload Autopilot is selected from the Debug menu.
I've also added the ability to define an enabled property for the filters.
It's used like the PID controllers. If there is no enabled tag then the
filter defaults to enabled so that nothing should get broken by this change.
This ability can be used to create a filter between the output of a PID
controller and the property that it controls (a control surface). By putting
a noise spike filter between the output of a controller and the control
surface that it controls, we can simulate the limited movement rate that is
inherent in autopilot servos.
This patch contains bugfixes from my previous patch
(i made mistakes when using "diff" utility last time):
1. Now ATC do not tells us to go around because of
traffic on RWy, when that traffic is WE.
2. Now ATC don't say that we must follow ourself
(like "GFS, number two, follow GFS on final").
And new Tower-ATC features:
1. Requesting departure clearance. When you are
on taxiway, you can request departure clearance.
Tower will answer with something like:
"Line up runway two eight right".
TODO: check if any AI-Planes on final and tell
something like: "After the landing AI-CALLSIGN
line up runway two eight right".
2. Requesting take-off clearance. If you are on
runway, you can tell to ATC that you are ready
for take-off. So, Tower will tell you that you
are cleared for take-off (or immediate take-off).
3. Reporting landing gear position on final by pilot.
If you are on aircraft, which landing gear can
be UP, when reporting final you will tell
"Gear down, ready to land".
4. Reporting landing gear position on final by ATC.
When i was newbie and was flying on Cessna 310, i
ALWAYS forgot landing gears on final. Now if you are on
final and forgot landing gear, Tower will tell you:
"LANDING GEAR APPEARS UP" or "Check wheels down and
locked". So you will have some time to push gear
down or execute missed approach ;)
5. ATIS phonetic ID in requesting landing message.
In FG pilots says: "for full stop with ATIS" everytime
(when airport doesn't have ATIS too). But in real life
pilot saying phonetic ID of ATIS message. Now we have:
"for full stop, information alpha ... / ... zulu".
If airport doesn't have ATIS, pilot don't say this.
TODO: add this feature to AI-Planes too.
6. Advising weather condition in airport by Tower.
If you are requesting take-off or landing in airport
that doesn't have ATIS service, Tower will tell you
about wind direction and speed, visibility and QFE.
7. Reporting downwind if missed approach. When some AI-Plane
is on final FG's ATC tell you to "continue approach".
And then you lost radio transmission. Now ATC says:
"continue approach and report left/right downwind".
And now we have "Report Downwind" entry (button)
in "ATC Communication" form.
This also fixes distortion weirdness in the osgviewer version. That
was caused by osg::GraphicsContext getting in the act on resize events
and adding its own scaling to the projectionOffset of slave cameras!
The LOD far range on the tile entry scenegraph node was initialized to
0. This meant that any traverals of active children that happened
before the tile manager updated the node would ignore the node
altogether. Among these is the groundcache traversal which was failing
at startup even though scenery was loaded.
Also added a function to dump scene graph nodes to files; very handy
in gdb.
Make an OSG file reader for .stg files.
New class flightgear::SceneryPager, which is a subclass
osg::DatabasePager to handle explicit delete requests.
Modify FGNewCache, FGTileEntry, and FGTileManager to use
SceneryPager. Mostly this involved removing the queues that talked to
FGTileLoader.
Calculate accurate tile timestamps from the time they are traversed in
the cull stage (which means that they are visible) instead of updating
them periodically.
Replace tile entry transform and range node with one LOD node
This fixes a race condition when teleporting somewhere outside of the
currently loaded scenery. Before, if there was no valid scenery, an
aircraft might end up at the center of the earth...
Instead of just reporting that the magic number, length, etc. of a
multiplayer packet is invalid, abort processing this packet. Also,
check if enough space remains to send a property string.
are not available out-of-the-box on all systems and keyboards, and should
therefore not be used in files committed to CVS. This makes them well
suited for local key bindings, as they aren't likely to get overwritten
with later releases. SDL supports Meta and Super, OSG supports only Meta,
and GLUT supports neither.
(Somehow this part escaped the previous commit. :-)
are not available out-of-the-box on all systems and keyboards, and should
therefore not be used in files committed to CVS. This makes them well
suited for local key bindings, as they aren't likely to get overwritten
with later releases. SDL supports Meta and Super, OSG supports only Meta,
and GLUT supports neither.
This uses the osgViewer infrastructure instead of setting up and osg::SceneView.
When the same change is made for glut, much of render.cxx can be deleted.
Some cleanup in tileentry.cxx, but mostly code to implement loading of tiles
and their models without reference to FGTileEntry or SGBucket structures.
Also, don't do deferred model loading; load them when the tile is loaded.
_material was uninitialized. This was causing a segfault with
as-yet-to-be-checked-in paging code; I don't know if it ever resulted
in a segfault with the old tile loader, but it certainly could have.
file. Possible uses of this functionality could include converting the
model to some other format or coordinate system for use in some other
visualization or simulation.
This is required to make sure the same letter key gets released,
which got the press event. (After Ctrl-press -> a-press -> Ctrl-release
a-release we want the Ctrl-a binding released, not the a binding.)
- add key listener interface for direct key access from Nasal space
This removes some jitter in cases where Nasal is used to set up view
parameters from FDM data, such as position and orientation. (The event
subsystem handles Nasal's settimer() calls.)
This was correct in the old repository and in revision 1.1 of the new,
but then broken in revision 1.2. After that, "lookat" and "lookfrom"
mode used different coordinate systems, and the "Adjust View Distance"
didn't work correctly in "lookat" mode.
"I have been investigating the Concorde IVSI problem. I came to the
conclusion that the trouble is that the environment altitude and thus
the pressure (which is calculated from that) is lagging by 1 frame.
Normally that wouldn't be a problem, but the IVSI calculates rate of
change and it will use the new dt with the old value difference,
thereby arriving at bad results if dt changes (and it does)."
invocations, but left it in place for miscellaneous allocation work.
Contexts cache allocated objects in a temps vector and only clear it
out when they are used.
Also, fix a type warning while I'm in there.
version
* Delete ai list objects in ~ATC/AIMgr.cxx:AIMgr::~AIMgr()
* Delete colors in GUI/new_gui.cxx: NewGui::~NewGui.cxx
* Delete memory allocated to the class member "route" in
Instrumentation/gps.cxx
* Delete all globals (except a few "unsafe" ones that still cause segfaults
and need further examination.
* Use an SGShared pointer for navaid memory allocation, so that pointers to
individual navaid objects can be included safely in multiple navaid lists
- simplify listener purging and fix removelistener() return value for
one-shot listeners
- listener: inherit virtualness
- s/handler/code/ in listener code to be consistent with NasalSys::call
argument:
setlistener(<property>, <func> [, <initial=0> [, <persistent=1>]])
This definition defaults to the old behavior. The fourth argument can
be set to zero, in which case the function is only called when the
node value has actually changed. It should really default to zero, but
then all callers would have to be reviewed, which is a bit too dangerous
"shortly" before a release.
0 -> trigger whenever property is written to (even if it's the old value; default)
1 -> like 0, but also trigger the function initially
2 -> trigger initially, but then only on changes
AMIBO 34.973889 021.658056 ARLOS 34.625278 023.000000 1 095 195 M871-MT E-MT W-N4
... where unexpected ids follow (here E-MT W-N4). The loader didn't skipeol.
The additional values are for a first quick&dirty fix just dropped.
remove all those 41 hacks if possible. Metrowerks is under category
"Defunct software companies" in Wikipedia, its compiler was known to
work on PowerPC (which we don't support, anway, right?) and on some
Unices. The hacks are in CVS since the first commit 2002/9/10 (old
repository!) and probably were in the code long before that. Any
platform where such a broken compiler is still used, will hardly
be suitable to run fgfs at all.
