The above link gives details over the approach taken by the EPO, giving examples of pure software patents which were rejected whilst also illustrating the approach taken to granted patents in the area.
My understanding is that software in its own right cannot be patented but where software is incorporated into a specific product to provide functionality it can be patented.
/edit The following link also provides good information. Essentially, the barrier is very low to establishing that a software patent achieves a technical measure and falls outside the exception. For example in HTC v Apple, a lower court judge found a multi-touch related patent invalid as a computer programme, however a higher court overturned this ruling.
> where software is incorporated into a specific product to provide functionality it can be patented.
Actually it is more like when a process like the one that can usually be patented contains software, then the software part can be included in the patent covering the whole process.
The key part of Art. 52 §3 creating exemptions from §2 is "as such" [0].
Say you invent a robot able to weld faster and more precisely. You can request a patent covering the robot and, being a fundamental part of it, the control logic implemented in software. Yet you can not try and patent the control logic by itself, nor sue anyone using the exact same control logic anywhere else, including in robots (as long as they themselves don't infringe on the patent). Conversely, someone using the same robot design but a different software algorithm may not be infringing the original patent. This protects both the innovator from being flat out ripped of and everyone's right to use algorithms and mathematics freely, while incremental innovations are allowed either in software or in hardware.
The problem is that the meaning of "software as such" has been almost completely eroded over the past decades by the EPO and some national courts (including, in particular, the X. Civil Senate of the German Federal Court of Justice).
"Software as such" used to be excluded because inventions were required to be technical, which meant that a patent had to teach a methodical use of controllable forces of nature that attained a goal based on cause and effect. The criterion of "being technical" predated "software as such", but the latter term was indeed intended to distinguish between pure software and combined software/hardware solutions, where the software was integral to the overall solution, reflecting previous exclusions of business methods, mathematics, etc. under the same criterion.
What happened was that the "use of controllable forces of nature" got watered down incrementally until now practically anything that isn't defined in terms of a Turing Machine satisfies the criterion (sadly, I'm not really exaggerating much here). For example, anything that is designed to operate in finite (not even practically limited, just not infinite) memory can be argued to be related to managing resources and thus be technical. Any algorithm that is more efficient than another (uses less memory, is more cache-friendly, requires less I/O, etc.) is potentially technical.
As a result, tying your algorithm to a device as an integral part of a combined hardware/software system is not really a necessity any longer. That does not mean that anything goes, but there are way too many patents relating to what developers (but not the EPO) think of as pure software solutions that are patented and which can trip you up.
This is precisely what the German parliament is referring to when it complains that "in practice -- in particular by the European Patent Office (EPO) -- patents were granted for computer programs where the patent grant for teachings relating to pure data processing received only a formal disguise as a 'technical process' or 'technical device' and claims explicitly included also the computer programs implementing these processes or devices".
The above link gives details over the approach taken by the EPO, giving examples of pure software patents which were rejected whilst also illustrating the approach taken to granted patents in the area.
My understanding is that software in its own right cannot be patented but where software is incorporated into a specific product to provide functionality it can be patented.
/edit The following link also provides good information. Essentially, the barrier is very low to establishing that a software patent achieves a technical measure and falls outside the exception. For example in HTC v Apple, a lower court judge found a multi-touch related patent invalid as a computer programme, however a higher court overturned this ruling.
http://ipkitten.blogspot.co.uk/2013/05/from-alice-to-apple-p...