128-bit UUID byte swap problem

[kerem]

I'm implementing a custom service with one characteristic and 128-bit UUID.

I am able to set the custom base UUID using sd_ble_uuid_vs_add() but the resulting bytes look swapped in nRF MCP Android application. This is on Softdevice 7.0.0 and SDK 6.1.0.

#define BASE_UUID {0xAB, 0xCD, .., 0x12, 0x34}

the resulting UUID I see is first and last 2 octets swapped, also within themselves like this;

{0x34, 0x12, .., 0xCD, 0xAB}

All the octets in between these 4 are fine.

Any idea what could cause this behavior?

EDIT Attaching screenshot from Android MCP:

[RK]

There's nothing wrong with that, the bytes are little-endian and the number is displayed in the MCP as a big-endian number.

You say all the octets between those are fine ... well all the octets between those are all 0x00 so you wouldn't be able to tell which order they were in.

Why are they all zero by the way, it's probably a legal UUID but certainly not one you randomly generated.

[kerem]

I deliberately set them to 0x00 for brevity. At first sight i was able to observe the order problem in the first 2 and last 2 bytes since i manually entered them. So is this ok or should i encode the uuid first to change its endianness?

[RK]

I really don't understand your question. There is no problem. You have entered a series of bytes in an array to use for your UUID. Those bytes represent a 128 bit number. The first byte in the array is the least significant byte, the 16th is the most significant byte. When you look at the number formatted for humans to look at, which is most-significant byte first, you see the last byte in the array going all the way down to the first byte in the array.

This is not 'the first and last pairs of bytes are swapped and they are also swapped within the pair'. This is 'I see the number with the bytes displayed in the exact opposite order to the way they are in the array, with the 16th first, then the 15th, then the 14th all the way down to the first'.

[kerem]

I understand the reason is the reverse endianness between nrf51-dk and MCU Android app, fine. What i dont understand is why? I dont remember anything about this even though i used MCP quite a lot so far for displaying various diffrent UUIDs. It feels strange that i never noticed such behavior.

[RK]

It's just a corollory of the BTLE spec. Each parameter is transferred little-endian and the way it's specified in the code is in the order the data is going to be transmitted, so least significant byte first. That saves I suppose, the softdevice having to flip it all around before storing it/sending it. The conventional way for humans to look at UUIDs however is big-endian in groups with dashes.

It's quite easy to look at UUIDs in the MCP or other place they're displayed and not even think about the order they were specified in code originally, seems that this particular time you did look closely and saw it, but it will have been that way forever.