Oh, this discussion is getting more complicated, to answer to you both

I think many of the issues stem from different history in our countries. It seems that Czech (or Austro-Hungarian) and Swedish practices were in fact quite similar, while UK was in a bit different situation. The most important is, that all Austro-Hungarian railways were first built as single-track. The funding and revenues were not so big as in UK. Lines were upgraded to double track, only when the traffic was too high.
The first continental (horse-drawn) railway Budweis (CZ) - Linz (AT), was built in 1827–1832, long before Morse's Telegraph (patented 1847). The first operation was on strict adherence to timetable at passing loops, and drive by sight on the line. In case there were two trains in opposite direction too far to get back to passing loop, one of them had to be derailed... Well this is not possible to simulate in simutrans....
Later they used optical telegraph
http://www.laenderbahn-forum.de/journal/heldenzuege_1859/heldenzuege_1859.html, these baloon signals, are sometimes described as signals for the driver, sometimes as a means of communication between signalboxes. In either case, they informed about the current direction in which the traffic is allowed. Some sources say that these were used also as time-interval signals (clear, slowly, danger).
In 1855, electric bell signals were introduced to replace optical telegraph. They were clearly a means of communication between signalboxes. Optical telegraph was abondoned in 1877. To remember the agreed direction, perhaps pen & paper were used, later some sort of needle telegraph. Somewhere I read about some sort of departure signal that was interlocked with the agreed direction, but it was not very detailed, and now I cannot find it. I suppose that multiple trains in the same direction were allowed in time-interval, otherwise there would be no difference between time-interval with telegraph and token block (or absolute block with signalboxes only at stations/loops). Unfortunately, there is not much info I could find on the details how time-interval signalling worked. In essence the dialog between stations was probably close to absolute block signalling. The stations had to count or somehow identify the trains and get confirmation that it arrived, in order to be sure that the line is clear. But the exact method would be just a speculation.
To the incentive to upgrade from time-interval to token block. Safety is not simulated. If multiple trains are allowed between stations, then you need a few signalboxes on the line. These will be removed in token block - as a result you cut costs, but reduce throughput. Maybe that's why this transition did not happen in Austria-Hungary, and absolute block was used instead - safety improved, while costs and throughput remained roghly the same. However, there was a big cost saving with the transition from optical telegraph to electric bells (and other electric devices). Signalboxes for optical telegraph had to be in sighting distance to each other. Introducing electric communications allowed big reduction in signalboxes. But this change was independent of time-interval vs. absolute block.
When absolute block was introduced in 1880's, the electromechanical devices that were used, had a provision for setting the direction of traffic - so they were a bit more complex than their british counterparts. In absolute block, the departure signal was locked at danger, if the traffic was agreed in opposite direction, but I have no evidence if that was so in time-interval as well. If only telephone/telegraph communication was used, the direction probably had to be put on paper, and doublechecked by the other side when negotiating next train.
As to the described dialog, and Ves's comment. Most of the sigle tracked lines have signalboxes only at stations (passing loops), with a few stops in between, but busier lines (or where the stations are too far) have an intermediate signalbox. Perhaps even more intermediate signalboxes were possible, but according to this map, the current practice is to have only one (on single track line):
http://provoz.szdc.cz/PORTAL/Show.aspx?path=/Data/Mapy/TZZ.pdfIf there are no intermediate signalboxes, then the whole dialog is much simpler, especially in regard to direction of traffic - there can be only one train between stations, and it does not matter in which direction.
I have not encountered any notice about using token machines in Austria-Hungary or Czechosolvakia. Perhaps staff and ticket+staff were in use somewhere, either to cut costs (no signals at all), or as a fallback if the communication between signalboxes is broken (that's even the current practice). However if you imagine a token machine as an upgrade for one-train staff - where the token can "magically" reappear at the other end of line, then you can think about the needle telegraph (or improved absolute block machine) as an extension to ticket and staff, where the staff (permission to dispatch trains) can be handed over to the other signalbox instantly. I acknowledge that UK practice is different, but the game should be country independent. You can then limit the available signals at pakset level.
On czechoslovak railways, there is almost no difference between absolute block a and circuit block. It is just improved safety, faster negotiation, reduced manpower - fully automatic signalboxes on the line, remotely controlled stations, and the integration with cab-signalling. But the procedures, e.g. to switch direction on single tracked line is the same as with electro-mechanical absolute block:
Back to the idea of special signals for directional reservations: Even in circuit block, you can think of the line as being split into small blocks - signal to signal, where only one train is permitted, and longer blocks - station to station, where multiple trains in the same direction are permitted. Making these blocks explicit (e.g. by redefining the use of long-signals), could help to solve also the problem of junctions without passing loops. No need to have two levels of signalboxes, just two levels of signals: block, and long-block. Is that description clear?
Can either of you think whether there are any situations in which you would want a bidirectional signal that is not one that creates a directional reservation in the track circuit block method, or will having either bidirectional signals or "longblock" signals creating directional reservations suffice?
I have no problem with bidirectional signal creating directional reservation. It is the other way around - I need a one-directional signal (usually ahead of platform - departure signal) to create a directional rerservation up to the next passing loop. As the passing loop is also not easily detected by code, the end could be marked by the same signal (and aditionally by choose signal to aviod the choose platform logic in directional reservation). Moreover I find too little use for bidirectional signals anyway. Even in places where they could be used I would like them spaced further apart (so that a stop with platform fits in between), or to put them around road crossing so that they are back to back, rather than face to face as they are now. So if you implement the proposed "new long block" in addition to keeping directional reservations from bidirectional signals, it is ok for me. Just that the bidirectional signals would not be necessary for me any more.
Ves: as to the spoken token block (D3 in CZ) - I think there is no need for different behavior to physical token block. I remember that at one such line, the driver had to get out of the engine, go to the station building and get the permission to continue by phone. Nowadays they can get it over radio or mobile phone, or a special radio device (RETB in UK, D4 in CZ), but that usually happens on places where those trains stop anyway. At least it would be an incentive to upgrade to proper signalling.
What you describe as the "2 block signal" is, I believe, the same as the pre-signal from Standard (which, of course, is not a pre-signal at all in real world terms). Does this method of signalling actually reflect any real life use case, or is it intended as a workaround for some Simutrans specific issues? I am still a little unclear on how this type of signal would help in stations.
As I undertand Ves, yes 2-block signal is the pre-signal from standard. Read further on real use:
2-block signals (hand signals). We have those in CZ as well. Some smaller stations have one departure (exit) signal common for all tracks, and the signalman has to give the departure signal to particular train. Perhaps this was even more common before WWII. German occupation introduced the 2-block signal - visually similar to UK banner signal - that could be put at every track and allow shunting as well as departure (then it had to be in sync with the main signal). There was also a color light version, but they were not very widespread and later replaced by full-featured main signals at every track. Nowadays there are only a few stations in CZ/SK with those signals. I do not consider them as essential (but I dont know how common they are in Sweden). If directional reservations would be resolved by explicit (long) signals, then these may not be needed (but nice for extra reality). Otherwise these 2-block signals may be useful to get single tracked lines working.
In real use - if the "banner" signal was pulled off it allowed either shunting or departure from the particular track, but for departure you had to obey the main signal as well. The color light version has 3-aspects: danger, shunting allowed, departure allowed. In simutrans it would solve the need for bidirectional signals. Look at the graph I posted yesterday. To get directional reservations in current circuit block, I would have to make the entry signal bidirectional. But then a train could depart and get stuck in switches. So a 2-block signal at the platform would be handy. But I think that a long-signal with explicit directional reservation would be a better solution to that.
Uff sorry if it is not consistent, It took me too long to write this (lots of interruptions like family lunch, etc...)
I hope to get the server working tonight - will be back at 64-bit machine...
And for the test game, it does not work. Here is what I did:
git clone --branch devel-new git@github.com:jamespetts/simutrans-experimental.git
git clone --branch half-heights git@github.com:jamespetts/simutrans-pak128.britain.git
cd simutrans-experimental; make;
cd makeobj; make;
cp build/...../makeobj-experimental ~/simutrans.pak128-britain
cd ~/simutrans.pak128-britain; make
etc, etc...
In the pak directory I have:
roadsign.lq-semaphore-token-narrow.pak
roadsign.lq-semaphore-token.pak
roadsign.lq-semaphore-token-white.pak
roadsign.two-aspect-colour-light-token-narrow.pak
roadsign.two-aspect-colour-light-token.pak
roadsign.uq-semaphore-token.pak
In game I cannot build any token signal, only plain signal, permissive (with disc), distant, combined, choose signal and one-staff cabinet.