trezor-firmware
trezor-firmware copied to clipboard
trezor
core UI: single global layout
Fixes #2299, also should fix a race condition in tests that has caused problems in the past, also reduces the need to pass Context to layout functions.
I'm not very happy about BufferLock-related changes in 6000b74 as they add a level of indent to already complex code. Changing this part was needed because test_show_multisig_xpubs causes both WebUSB and DebugLink to process a protobuf message at the same time, so we just allocate new buffer instead of aborting. Let me know if you have better idea how to solve this.
BufferLock-related changes
how about keeping most of the original, but make the BufferLock async, as in:
async def __aenter__(self):
while self.in_use:
yield
(and then async with buffer_lock)
async def __aenter__(self): while self.in_use: yield
I've tried this and it resulted in a deadlock - firmware writing to WebUSB buffer, trezorlib reading from DebugLink interface (or vice-versa, don't remember).
I've tried this and it resulted in a deadlock
right so what if we just put back debuglink's own buffer? it doesn't need to be 8 kB, by a rough calculation 512 bytes might be enough, or 1 kB to be on the safe side. we should be able to afford that now.
so what if we just put back debuglink's own buffer?
Done in 3a976973965971dbf650ad7bbbff6199ee52169c
hackathon results are in hackathon-global-layout branch
further idea: debuglink should send decisions not via the global confirm_chan, but directly to the running layout object -- presumably what is now self._cancel_chan can be just self.chan or something, and both the do_cancel() method and debuglink can insert results into it directly.
this also allows us to get rid of confirm_chan integration in a lot of places (but that becomes less relevant when we switch over to RustLayout)
Pushed:
- 585e525 cleaned-up hackathon changes
- 8397f7e ugly hack that prevents layout desynchronization in
test_show_multisig_xpubs
Unfortunately this breaks click tests because layout notifications now seem to have problem working across session restarts. @matejcik can you please take a look at this? I'm out of ideas what to try. It's reproducible with PYTEST_TIMEOUT=5 pytest -x ../tests/click_tests -k 1of1.
In mmilata/global-layout-notify-after-br I tried sending the layout notification after ButtonRequest but before the layout is actually painted. Sounds wrong but works surprisingly well except that it breaks test_signtx_data_pagination[input_flow_go_back].