A switch may lie on a concave vertical curve, and on a convex one no sharper than 2000 m (Dz.U. 2026 poz. 1225 § 58 ust. 16): the fit bends over the switches and their shared timbers where the ground does, a curve on a switch is a warning, and only a sharper convex one an error. A sharp crest of a hump keeps its curve off the switch zones. The zone of a crossing reaches only as far as the path crossed has rails, so the middle of a double slip keeps to itself.
A line runs through a switch along its main path, and through the quarters of a double slip straight on, so the long lines go through the slips and the connections between them are fitted after. Rails of the lines fitted already nearer than a sleeper is long are taken at their height, as both lie on the same timbers. A short line between two lines whose end grades can't be joined within the norms keeps only the heights of its ends, and the grade breaks at the joints. The norms speed of a line is the one most of its length is laid for, so a few tracks set much faster don't stiffen the curves of all of it.
The fault finder leaves out the places over the terrain model on a bridge or a viaduct, the ground falling away under them by the abutments. A formation modifier may be open at its start or end.
Laid to the norms, the readout says the design speed of the transition curves and of the cant: that of the track the laying goes on from or joins, or the one set for the laying. A new option takes the set speed also on from a track with its own, the new tracks get it; Ctrl+Shift+wheel steps it while laying and turns the option on. The panel shows the speed the laying under way is designed for.
A track laid on from a free end in the PKP PLK laying, by a drag out of the end, by clicks from it or laid back onto it, starts from the curvature and the cant the end has. An arc turning the same way as the end is a compound curve: a transition of the norms ramps the curvature and the cant from the one arc into the other. From a straight the arc gets its entry transition, the arc then straight its exit one. Where the way is too short the transitions are cut and the cant lowered to what they ramp, and the readout says so; with no room at all the track is laid as before and the note says why. The curves traced on the orthophoto and the recommended values of a compound curve get the transitions between their arcs too. The free laying stays as it was.
The free laying isn't the default any more. The first time a session opens the laying or the plan, a window asks how the
new tracks are laid, with the two cards of the panels (the norms marked as chosen); a click on one of them takes it and
closes the window, which comes before the offer of the terrain model of a new scenery. "Remember the choice, don't ask
again" keeps it in eu07_editor.ini (track_norms ask, free or plk), and a choice kept follows a later click on the cards of
the panels; the settings of the editor (General) choose between asking and either way. The self-test scripts start with the
free laying, as they were written for it.
M in any tab of the track mode turns the tape on: a click puts its start, the next one its end, at a free end, on the axis
of the track under the cursor or else on the ground. The readout at the cursor gives the distance in plan and in space,
the difference of height and the grade between them and the azimuth, while the second point follows the cursor and once
it's put; a further click starts a new measure, M or Esc puts the tape away. Self-test command tape.
In the top view at a metre a pixel or closer, each curved track of 5 m or more gets "R 600 L 120" by its middle and each
switch its name and its type from the kept switch record ("Rz 60E1-300-1:9 P") by its start, on a dark plate as on a plan.
A label which would cover another one is left out, the switches going first and then the sharper curves. View > Track labels
in the top view switches them off.
View > Track colours draws the rail tracks over the view in bands of the radius of their paths, of their speed or of the grade
along them, from above over the whole view and in the 3D view within 1.5 km of the camera, with the legend of the bands in
the bottom right corner. Plain keeps the lines of the top view as they were. Self-test command trackcolours.
After a change of the tracks, once the mouse is let go, the joints check looks at the joints of the tracks changed and of
the tracks they join, with the limits of the Joints panel: ends almost meeting, steps, kinks, jumps of the cant and of the
grade get a marker with the value in the view while the panel is closed, and go away once a later change or an undo puts
them right. The check of a track is shared with the panel (joints_check takes the list and the pairs seen), the pairs
already looked at are kept by the tool instead of a global. Self-test command jointslive.
The frog, movable nose and 3D sleeper and timber fields of the track properties and the frog chips of the switch panel stay
in place while track_features::crossings or track_features::sleepers are off, greyed, with a tooltip saying the feature is
switched off in this build; nothing changes them, copying and pasting the properties leaves them out.
Each end of a rail track joined to nothing, where a train would run off, gets a red dot next to the white dots of the joints,
in the 3D view near the camera and from above over the whole view. The ends are looked for again after an edit or once the
camera leaves the place they were found for, not every frame.
After the cursor rests on a track, its label gives the grade at the place pointed at and the cant of the path, besides the
name, the length, the radius and the speed; a curve whose radius isn't written in the scenery gets the one of its geometry
instead of being called straight.
Next to the length and the radius, the readout at the cursor shows the difference of height from the start to the end of the
track about to be laid, the grade along it (one value or the range it goes over) and the speed the sharpest arc lets the
trains through at within the cant and the unbalanced acceleration of the laying limits. The self-test layclicks logs it.
The points of a grade line read from the tracks keep the radius measured there; once a point between them was taken out, the
breaks of its neighbours changed while their radii stayed, so a break without a curve read as 0.4 per mille became one of
9 per mille with no curve, and the tracks took it as a kink. The neighbours of the point taken out follow the limits now,
with the button, Del and the self-test alike (profile_remove_point).
Applying a line with a break without a curve gave the path which ends at the break the grade of the element after it, so
its Hermite heights bowed 0.12 m off the line over 100 m before the break; each path takes the grade on its own side now.
Self-test commands vprofilerun, vprofiledel and vprofileput.
In place of a line bent every ten metres to keep within 0.1 m of every bump,
the grades and the vertical curves are fitted to the ground the way the
profile fits them. Where the formation is made in the terrain model the line
keeps to the radius of the norms and cuts and fills; over the ground left as
it is it bends as sharply as it has to. The joints keep the height and the
grade of the track joined, no grade is capped and no piece is under 5 m.
The panel gives the place on the map or typed, the area, one main button, a
progress bar with Stop and a banner over the view while it loads. The terrain
grown ahead of a track goes on at the height of the one beside it and skips
the ground made by hand. The formations follow the tracks moved, laid and
removed, in the same step of the history; a whole load is one step, the
ground grown ahead none.
An error put in the status bar also shows for 4 s by the cursor, and the
refusals the tools kept in their panels (a curve which can't be laid,
a switch which doesn't fit, a signal which can't be moved, a drive
without the variant) go there as well.
When the refitted curve needed more of the straight at its other end,
the ends of the line were both checked against the design, and the free
end that went along with the drag "missed" by as far as it was dragged.
Only an end whose straight was cut is checked now. The longer transition
curves of the PKP PLK norms ran into it more often.
Dragging an end of a straight refits the curve next to it with the cant
and the transition curves recommended for the speed, keeping the curve's
own only where it had some; a plain arc laid freely came out with
transitions. Break and S-curve took the recommended transitions too. The
recommendation of a design not checked against the norms is a plain arc
now, the tools take the norms in effect rather than the ones the design
was last computed with, and the Plan panel calls the automatic choice
"None" in the free design. Self-test: norms free|plk.
The shape pulled out of a free end took a typed radius for its curve,
but the typed keys were thrown away while the drag went on, so the value
never got there. Digits now type the radius whenever the shape has a
curve and a straight.
Esc in the middle of dragging a point of a track kept the moved point,
and the hint said so; dragging a straight was taken back but left an
empty step in the history; a vertex or the radius grip of a curve stayed
moved in the design. Now Esc puts back what was there before the drag in
all three, and the history doesn't keep a step for it.
Dragging a standing signal over the track takes it to the other side;
the side used to come from where the signal stood, so it never crossed.
The selected signal gets "To the other side" and "Turn around": the
first mirrors it over the axis facing the same trains, the second turns
it to the trains of the other direction and moves its event to the
other event list of the path. Either one swaps a head leaning to the
track for its counterpart, and an include keeps the template its
directive names, so undo brings the old file back too.
The name of a node or "(no name)" was written out in twenty places and
four local functions, a lower-case copy of a text in four files, the
test for a number, the file name of a template made comparable and the
heading of a direction in three and two, the green main button of a
panel in five and the key=value parameters of a self-test command in
five. Each of them is one helper now: name_of, tall_button_height and
main_button with the panel helpers, numeric with format, normalized
with the include helpers, heading_of with the plan geometry and
selftest_values with the tools; lower gives way to ToLower.
editor_services was growing into one interface every tool depended on
whole. It splits into the roles the editor plays for its tools: the
selection, the view, the undo history, the templates by the track, the
ground with the terrain model, the track design and the guide of the
first steps. Each tool is handed only the roles it uses, so what it
depends on can be read from its constructor; editor_services stays as
the sum of the roles the editor implements.
The seventh tool out of editor_mode. It reads the selected set of tracks
and the tab shown through the services, digitizes the tracks it lays and
refreshes the plan after it sets one parallel to another. The panel shown
for two selected tracks is a public call of the tool, and the self-test
command parallel moves with it and lays the same tracks as before.
The fifth and the sixth tool out of editor_mode, the two lineside ones
sharing their file. Placing, moving and taking out the templates which
stand by the track become services of the editor, with the template
types of their own, and so do the tab of the track tools shown, the
ground under a point and the first step done. The context menu, the
click on a fouling point in the view and the lineside tabs go through
the public calls of the tools. The self-test commands hekto and fouling
move with them and log the same as before.
The fourth tool out of editor_mode. The track under the cursor and the
first steps of the tutorial become types of their own, which the editor
gives to the tools, and a tool shown as a tab of the lineside panel keeps
its open and expanded flags in a tool_tab. The trainsets waiting for the
save are handed to the editor by the tool, which hears back when they
were written. The self-test command trainset moves with it.
The third tool out of editor_mode. It reaches the terrain model and the
places of the tracks on it through the services, the place of a track
for the terrain model is a type of its own now, and the laying and the
switches ask the tool for the kink it allows. A self-test command can
ask to be run again in the next frame, as the check against the terrain
model does while the model comes in.
The second tool out of editor_mode. Tools get two more hooks, the overlay
drawn over the view and the options in the tool window, and the editor
gives them the track it focuses on, the node it makes current, the norms
of the route design and the snapshot of the tracks for the undo.
The check gets a self-test command, speedcheck <scope>, which runs it
through and logs the paths found too fast for their geometry.
The first step of taking the tools out of editor_mode: a tool keeps its
own state and reaches the editor through editor_services, the few things
it needs from it, instead of through every member of the editor. The
editor holds its tools in a list for the hooks they share: a deleted
node is forgotten by all of them in one place, and a self-test command
the editor doesn't know goes to each of them.
The gauge moves with its state, its window and its self-test command;
the ImGui ids and the texts stay the same, and the self-test gives the
same hits on the test scenery as before.
The states of the objects which followed a track, kept in the history
for the undo, held the pointer of a model after it was deleted. Only the
address was checked before the model was moved back, so a node made
later at the same address, of another type too, was taken for it. The
states of a deleted model are dropped from the history with its other
pointers.
Found with an address and undefined behaviour sanitizer build, random
monkey sessions on the test sceneries and broken inputs:
- A deleted model left its pointer in the pick list of the renderer, the
bound node of the panels, the bindings of the infrastructure and the
selection: the next pick or panel frame read freed memory. The renderer
forgets it, and one erase_model does the cleaning for all the places
which deleted a model by hand.
- The plan band of a route read the vertex before the first one in the
frame a drag was let go.
- A scenery which can't be opened goes back to the launcher with the
reason in the log, a track of an unknown type or a tributary loads as
a normal track or a river with a message, and a spline with no finite
length or step gets a sane number of segments.
- Typed numbers which aren't (NaN, inf, ends 0 m or 1000 km apart, a
vector of 1e30, a diverging track part of 5 km, a curve threshold of 0,
a zero history size, radius norms below zero) are refused or clamped,
where they made the geometry run away or divide by zero.
- Folders which can't be read and file names which aren't valid in the
locale no longer throw out of the scans of sceneries, textures, sweeps,
roads and signal plates; the scans skip them.
- The launcher: Ctrl+Del with no trainset selected, a dropped vehicle
whose skin is gone and a text with a percent sign.
- A terrain mesh made for another resolution of its chunk isn't put into
the chunk, and a slot is reused only for a mesh of the same size.
- Vehicles on a track changed by the editor are moved over a copy of the
list they leave.
- The table of the keyboard shortcuts gives its stretched column a
weight, else ImGui divides 0 by 0 in the first frame of the window.
The python task queue was stopped after the windows were destroyed, and
its worker, leaving, released the context of its own window, which was
gone by then: a use after free on every exit. The queue is stopped
first.
The geometry of a track made anew, ten times a second while it's
dragged, only forgot the chunks it had: the bank kept them all, so the
buffer of the section grew by the whole track and its neighbours each
time and was sent to the gpu in full again. Ten rebuilds of a test
scenery took the bank from 2 to 29 MB; a long session would run the
card out of memory. The old chunks are erased from the bank now, but
for those of a moving nose, laid again in place, and a moved traction
wire drops its old one too.
A chunk swapped for one without the extra vertex data kept the flag of
having it and the upload read from an empty array; the buffer is made
again when that changes.
With the cursor out of the window ImGui has no place for it, and the ray
under it came out as NaN: the zoom put the camera there for good and the
next look at the ground under the cursor aborted. The wheel is left
alone then, and the ground isn't asked for a point which isn't one. The
self-test gets leave, the cursor taken out of the window.
The ends' budget of a curve too big for its leg lowered the scale of both
vertices beside it, a straight one too, which has no tangents to ask:
with no transitions the empty function was called and threw
bad_function_call, with transitions the vertex got a false warning of a
reduced R. Straight vertices are left out of the scaling.
The tiles are given to the gpu straight from memory; the garbage sweep
released them like the ones read from files and they came back as the
grey stub until made again. They're kept now, and a new texture name
clears the cache of the bound units.
The lift moved the models of the includes it remade a moment later, so
the cells kept the old ones after they were destroyed and the next look
at the ground there (a zoom) crashed. They stay as they are now, and the
models already waiting to be destroyed are skipped too. The self-test
gets danglers, which counts the models the cells hold past their end.
The terrain model was made once around the camera when a new scenery
took it; tracks laid on past it went over nothing. While a track is laid
or dragged out, the chunks missing within 400 m of the cursor are now
made from the model too, the nearest first, in a scenery with the
heightmap terrain placed on the map. The formations aren't made anew
after such chunks. A chip in the terrain panel turns it off.
A touchpad under Wayland sends a scroll as many fractions of a notch.
Shift+wheel took each of them for a step and went round the shapes of
the laid track, the joins and the connections several times in one
swipe; now the fractions are gathered into whole notches and a gesture
moves such a choice one step. Ctrl+wheel steps the radius once a notch.
The 3D zoom measured to the level 0 when nothing was under the cursor,
which in a scenery on the terrain model lies far below or kilometres
ahead, so a light swipe at the horizon flew the camera away. It
measures to the ground under the camera now, backs away by no more than
four heights of the camera over it, and takes at most two notches at a
time. Self-test laystate tells the shape, the radius and the camera.
Besides the steps of 15 degrees, Alt snaps the laid straight to the
ratios of the switches (1:4.8 to 1:26.5) of the straight track the
straight crosses, or of one within 30 m of where it starts; the label at
the cursor says which ratio to which track. The first click with Alt
within 2 m of a straight track puts the start on its axis, so a branch
or a crossing at 1:9 starts exactly on the track.
The reverse curves of a join start and end at the joined ends, with the
largest radius that fits, so a shift of the track spacing (4 -> 4.75 m
over 400 m) is one S instead of two curves with straights around them.
A straight or a single arc with a kink at its ends, though still under
the joints check, comes after the smooth curves among the variants; a
practically exact one stays first.
An element under 1 mm, like the straight left between the reverse
curves by the radius fit, no longer drops out of the pieces and leaves
a gap: the neighbouring piece takes it over.
A slip went to the crossing of the selected straight with another straight within 40 m of the click. When the track under the cursor wasn't taken as a straight (a slight curve, a switch), the slip silently went to another crossing up to 40 m away. A double slip within 25 m of the click was replaced, even when the click was on a free crossing.
- The crossing nearest the cursor, within 12 m, is found among all the rail tracks. A slip goes only there; when that track is a curve or a switch, the tool says so.
- A double slip is replaced only when it is nearer the cursor than any crossing.
- The cursor is taken at the height of the straight under it, so on a grade the 3D view doesn't put the click metres away from the track.
- Before the click, the view marks the crossing the slip goes to, with its two straights, or the reason it can't go there.
The joints check also compares the tracks with the terrain model (NMT), every 5 m. Bridges, tunnels and tracks over others are left out.
- Each path lying higher or lower than the model by more than the threshold (0.5 m by default) is listed once, at its farthest place off.
- The places are compared with the same offset the model has against the scenery; a scenery not fitted to the model as a whole is measured about its median.
- The squares of the model are loaded in the background, and the check waits for them with a progress bar.
- These entries can't be fixed from the list. In the top view, the overlay shows the issues over the whole screen.
- Self-test: jointsmodel <threshold>.
- In the top view of the tracks work area, the renderer's reach is the half diagonal of the screen. It used to be three extents, which cut wide windows short.
- Cells far from the camera no longer draw only their tracks: the terrain shapes are drawn too. Models keep their own ranges.
- In that view, the heightmap streamer loads chunks as far as the corners, up to 50 km. The detail comes from the size of a pixel, not from the distance.
- The walk over the chunks around the camera is skipped while it would ask for nothing, so a wide reach costs little.
- A setting, kept in the editor settings, leaves out the ballast of the railway in the top view, so the ground under the tracks shows. It is in the View menu and next to the extent slider.
- Self-test: ortho <extent> [noballast|ballast], orthophoto off.
