legacy codecs / mixing rates and versions

Today I checked the new AoIP core network using an old Nokia 1100 handset, which confronted me with codecs. The Nokia 1100 says it is capable of FR2, FR1, HR1: Channel Type - (Speech) Element ID: 0x0b Length: 5 0000 .... = Spare bit(s): 0x00 .... 0001 = Speech/Data Indicator: Speech (1) Channel Rate and Type: Full rate TCH channel Bm. Prefer full rate TCH (8) 1... .... = Extension: Additional Octet .001 0001 = Permitted speech version indication: GSM speech full rate version 2 (GSM EFR) (0x11) 1... .... = Extension: Additional Octet .000 0001 = Permitted speech version indication: GSM speech full rate version 1 (GSM FR) (0x01) 0... .... = Extension: Last Octet .000 0101 = Permitted speech version indication: GSM speech half rate version 1 (GSM HR) (0x05) The "stock" osmo-bsc.cfg I was using so far had only HR3 configured. So I naively thought, let's do: codec-list hr3 fr2 hr1 But OsmoBSC says: "Can not have full-rate and half-rate codec." Curious, is that a hard limitation? I'm not at all familiar with the reasons. Will we never be able to mix FR and HR? Why then does TCH/F_TCH/H_PDCH exist? Then I picked: codec-list hr3 hr1 That worked, but no audio: the Samsung Galaxy got HR3 and the Nokia HR1 (the BTS was configured for only TCH/F channels, so instead the Nokia actually got assigned an FR1 channel). The call was established but no audio worked. I believe in the OsmoNITB we used to complain in the log that we don't transcode in such a situation, and the call was aborted... When I pick codec-list hr1 then the call works between the handsets. Again both get assigned FR1 on TCH/F; if I configure the BTS to use TCH/H instead, both indeed get assigned HR1 as expected. All in all an interesting situation, which appears to mean that we practically can't allow the newer codecs alone because older handsets would not work, and we can't allow both because each side may pick a different one. Should we wait until we know both sides' codecs and consolidate with each other to pick matching ones before assigning? Is that a "late assignment" as in issue "avoid mismatching TCH/F vs TCH/H pchan types"? https://osmocom.org/issues/1778 But what if no match can be found? only MGW transcoding will solve all situations I guess. Are we planning to do that, also with IuUP in mind? https://osmocom.org/issues/1936 https://osmocom.org/issues/1937 ~N