Repository navigation
v0.1.30
Twenty-one Pulsar wireless mice get a sleep-timer control they never had, and stop having a register written that nobody has identified. Measured by @bluesceada capturing Fusion with a video beside it and @f-o-o-g-s reading the packets off.
The sleep timer
The mice on the Nordic chipset — the Xlite and Xlite V3 families, X2, X2A, X2H, X2 V2 and their variants — had no sleep-timeout control at all: the accessors the CLI and GUI look for sat in the X2 CrazyLight driver, two models away, while the register they write belongs to the whole family. They are now where every one of those mice can reach them.
Its bounds come from Fusion's own slider: 10 seconds at one end, 10 minutes at the other, in steps of 10 seconds. Those stops are not evenly spaced — three of them sit under a minute and then it goes by whole minutes — so a value between them is untested rather than known good. The timer is stored per profile.
A register that was being written on trust
Setting that timeout also wrote 0xAD, because a capture of a different application version driving a mouse that application does not even list had Fusion writing both. On these mice 0xAD is not the timer: it reads the same value in every dump and capture taken of them, including a profile whose timer reads ten seconds. It is now left alone, and what it actually does is still unknown.
Two answers that were being taken on trust
Byte 2 of a reply is a status — 00 in all 218 replies of the capture this was measured from — and two requests were ignoring it:
- A refused identity read came back as model 0, which was then remembered for the rest of the session, so the mouse was never asked again. It is asked again now.
- A profile switch the mouse refused went unnoticed, which leaves every read and write after it addressing the profile the mouse is still on while calling it the new one.
Also
max_dpi_stages=4 for this family was inherited from another project and had never been checked. All 21 models in Fusion's own config carry DM=4, with four values in each DPI= line and four colours in each DC=. That file has nothing to say about profile counts, so the one number in these capabilities still known to be wrong — num_profiles, which reports 1 for mice that have more — is waiting on a measurement rather than on a guess.
Full attribution is in the README.