10:37 <pepijndevos_> daveshah, fyi Python 3.9 gives some deprecation warnings

11:31 <pepijndevos_> hmmm... so Gowin has their own IBUF/OBUF primitives, which synth_gowin also generates with iopadmap. But nexpnr *on top of that* creates $nextprn_ibuf etc

11:32 <pepijndevos_> Wondering if I should write a packer that deletes the $nextpnr ones and packs the custom ones into a bel, or tell yosys to just not generate them.

11:35 <pepijndevos_> Packing is some tricky business... very easy to crash all the things with dangling pointers.

11:36 <daveshah> remove the $nextpnr ones

11:37 <daveshah> eventually I want to get rid of IO buffer insertion in nextpnr, it's more error prone than it needs to be when it comes to complex inout patterns

12:41 <pepijndevos_> ooops

12:41 <pepijndevos_> ERROR: net 'led[5]' driver port 'O' missing on cell 'cpu.alu_out[0]_LUT4_F_I2_LUT4_F_1_LC'

12:45 <pepijndevos_> huuuhhh, I'm not sure I understand what that means. uh... so the LC is driving net led[5] yea? LC doesn't have a port called 'O', so that's... weird.

14:27 <pepijndevos_> so led[5] is the top level output, which should be driven by an IOB, not by a slice... still no clue how tf that is happening.

15:01 <daveshah> make sure you are removing the nextpnr_iob s properly

15:12 <pepijndevos_> daveshah, pretty sure I'm *not* removing them properly...

15:16 <pepijndevos_> what does disconnect_port(ctx, ci, p.first); do compared to ctx->nets.erase(dnet);? Because the generic code does... both in different cases

15:16 <pepijndevos_> https://github.com/YosysHQ/nextpnr/blob/master/generic/pack.cc#L219-L240

15:17 <pepijndevos_> that just seems to delete the net and remove it from the port, and then delete the whole device

15:18 <pepijndevos_> but then a bit later in another case it just calls disconnect_port https://github.com/YosysHQ/nextpnr/blob/master/generic/pack.cc#L242-L244

15:29 <daveshah> they are different things

15:29 <daveshah> disconnect_port doesn't delete the net

15:29 <daveshah> ctx->nets.erase deletes the net, but doesn't disconnect any ports first, you have to do that with disconnect_port first

15:33 <pepijndevos_> right... but the first part just seems to do iob->ports.at(ctx->id("PAD")).net = nullptr; rather than calling disconnect.

15:33 <pepijndevos_> but okay, so the correct procedure is calling disconnect and then deleting the net?

15:33 <pepijndevos_> Because that's what I was doing.

15:34 <daveshah> yes that's correct

15:38 <pepijndevos_> this is my currently incredibly broken IO packing code: https://github.com/pepijndevos/nextpnr/blob/gowin2/gowin/pack.cc#L212

15:41 <pepijndevos_> *somehow* the driver of the top-level net... which should have been deleted... becomes a nonexistant slice output.

15:43 <pepijndevos_> oh... maybe?? hmmmm

15:46 <pepijndevos_> waaaat...

15:46 <pepijndevos_> problem magically went away after shuffling some things around...

15:47 <pepijndevos_> it's now happily running the placer

15:51 <pepijndevos_> hurray!!

15:51 <pepijndevos_> routing doesn't work yet, but that's a challenge for tomorrow.

15:52 <pepijndevos_> from the error seems like some dumb name mismatch, and still need to properly set up timing info.