When an interrupt fires, the interrupthandler is called in the context of the currently executing task. Shouldn't the first thing that is done in the interrupthandler be a contextsave?
Now each interrupttype has to determine by itself if a contextsave is needed.
Or am I missing something?

The interrupt handler prvLowInterrupt() in port.c first checks to find the source of the interrupt. No context has been saved yet so it is *VERY* important that no code is added to this function that could change the context. If you look at the assembler of the code that already exists you will see that nothing gets modified.

If the interrupt was the RTOS tick then the ISR does first save the context before switching, and restoring the new context.

If the interrupt was not the RTOS then what happens is application dependent. If the interrupt wants to cause a task switch it can do by calling taskYIELD - but this MUST only be done after the interrupt source has been cleared. See vSerialRxISR() for an example. This then saves the context and performs the switch as per required. This means the saved context is as it is within the ISR - the ISR is not completed. When the task runs again it will start from the point it left in the ISR and complete the ISR. The ISR cause was already cleared so this is ok - it is as if the ISR completed in one go.

The save and restore macros handle the interrupt flag settings that make this possible. See the comments above the portSAVE_CONTEXT and portRESTORE_CONTEXT macros.