Rose/Fell still not being DomainRenamed correctly? (SyncSerialILA not working reliably in other domains)

[miek]

I've been trying to use the SyncSerialILA in non-sync domains and have been having issues with it either returning all zeroes, or random garbage. I can reliably reproduce that by modifying the basic_ila_fast_domain example to run in the usb domain.

I tried setting up the SyncSerialILA's SPI interface to output a fixed word to exclude any issues with the ILA itself and I was still getting zeros out. I eventually narrowed it down to no edges occuring on the SPIDeviceInterface's output_edge (by routing that out to one of the user IOs and watching it on the scope while apollo did the download).

If I modify the Rose/Fell calls in spi_edge_detectors to use domain="usb" then everything seems to work correctly. This was supposed to be fixed by #30 but maybe there's some other corner case happening here? I don't know how to inspect the nMigen elaboration process well enough to debug that further.

(Also here's the IRC discussion that led to the PR above: https://freenode.irclog.whitequark.org/nmigen/2020-07-12#27492949;)

[miek]

Uses of Rose/Fell were removed in #231