This is the trick to retransmitting all the received satellite signals. Their is no requirement that the re-transmitter know where it is at all. Since you receive everything it knows about its GPS signals, you can calculate where the re-transmitter was when it sent the new signal. (Interestingly, if you are a military user, and have the P-code keys, the re-transmitter can send you both the C/A and P channels without itself needing to know how to decode the P channel, and you can then work out where the re-transmitter was to P channel accuracy.) The neat thing is that since the re-transmitted signals contain all the GPS satellite signals, they implicitly include all the timing information, which means the re-transmitted signal can participate in location calculations in your receiver.
You can actually buy re-transmitters that provide that part of the functionality, but I don’t know if there are any receivers that are able to make use of the idea. These re-transmitters are used in some radiosondes. I suspect the prime reason they are designed this way is to enable military precision uses, but also to unload the location computation to the receiver for the sonde. Sondes are less worried about precise location and much more about velocity.
It strikes me that you could use the idea with fixed re-transmitters to ameliorate the canyon effects in large cities. It would need a new GPS receiver design, and allocation of bandwidth to the system, but if someone like Garmin, or Qualcomm did it, they would be able to market an end to end solution. Multipath problems will remain, so it isn’t a universal panacea.