← Back to community index
# contributor·Active

0hmx - Problem: <trace path> is unordered today...

Problem: <trace path is unordered today. It's just a set of pins to connect. Making the order meaningful would change the behavior of every trace that uses path. The workaround (what core 5384 does) is to treat the order as meaningful only when a <differentialpair references the

Started by 0hmxNov 11, 202611 messages

Discussion

Last active 10 days ago · plain text
0hmxoriginal post
**Problem:** `<trace path>` is unordered today. It's just a set of pins to connect. Making the order meaningful would change the behavior of every trace that uses `path`. The workaround (what core #5384 does) is to treat the order as meaningful only when a `<differentialpair>` references the trace. I don't like that: it's hidden, undocumented behavior that you can't see by looking at the trace. **Proposal:** give `<differentialpair>` two flat props that list each lane's pins in signal order, from the first pin to the last: ```tsx <differentialpair name="USB" positiveOrderedPath={["J1.DP", "ESD.IO1", "R1.pin1", "R1.pin2", "U1.DP"]} negativeOrderedPath={["J1.DM", "ESD.IO2", "R2.pin1", "R2.pin2", "U1.DM"]} maxLengthSkew="0.1mm" /> ``` This keeps `<trace path>` exactly as it is, makes the order explicit where it matters, and gives the router what it needs: an ordered list of pins per lane, which maps directly onto the SRJ "path of point ids". The props are flat rather than nested, following the handbook (§1.2, "avoid nesting"). `positiveConnection` and `negativeConnection` are **deprecated** but keep working, so existing designs don't break. <@757706909351411845>
Seve
i think path is ordered
Seve
path should be ordered i think
Seve
<@400684445054795796> you REALLY need to be careful, because even though we've discussed this a lot, you've dropped the VERY IMPORTANT `connectionPath` naming
Seve
you've omitted `connection`, which is so important
Seve
<@809856037376491570> note the comment regarding core not always treating the path is ordered, that is a bug you've introduced here. The path should always be ordered, not conditionally
Rishabh
Ohh
0hmx
the auto router uses the mst and dose not respect core; this is more on the auto router side; not because of <@809856037376491570> core respects the order internally but auto router dose not care
Seve
the netpointpairssolver should use the path given by the user
Rishabh
https://github.com/tscircuit/length-matching-solver/pull/87 https://github.com/tscircuit/tscircuit-autorouter/pull/3643 cc <@400684445054795796> <@757706909351411845> This was the approach I had
Seve
i think your autorouter isn't using the path data properly- i could be wrong

Want to add to the conversation?

Reply in Discord so your notes stay connected to the source.
Continue in Discord ↗