@User: I'll take a look. Thanks for reporting back.

I updated the message with how to reproduce the error after you responded.

Even better!

@Tim Edwards I logged this on github with some more information. https://github.com/RTimothyEdwards/magic/issues/70

@Mitch Bailey: I basically gave up on doing this the preferred way. I had previously removed a restriction from magic about using brackets in instance names. Possibly if I stopped worrying about this in net names and just escaped slashes in instance names, it would work. For now, I am just prohibiting slashes in instance names and converting them to underscores when reading from GDS or DEF. That should take care of the problem for now.

@Tim Edwards Your fix 8.3.132 appears to work. I logged the issue to github because on my free slack account, I can't access messages older than a month or so. The current issue may be unrelated. The ext file contains merge statements for nets that are internal to subckts.

merge "PHY_1162/VPWR" "soc.soc_mem.mem.SRAM/FILLER_173_42/VPWR"
merge "soc.soc_mem.mem.SRAM/FILLER_173_42/VPWR" "PHY_463/VPWR"

soc.soc_mem.mem.SRAM

is an instance defined in a

use

statement. I think it wants to merge with the net

FILLER_173_42/VPWR

of

soc.soc_mem.mem.SRAM

but that's not a port of the subckt. How does magic extract connections to non-ports in subcells?

@Mitch Bailey: It's not a valid port, but it's a valid hierarchical name. This was the problem I was having with the slashes in the instance names. Magic uses the "/" in the .ext file to represent the hieararchy, so when the "/" character shows up in instance names, the extraction gets totally screwed up. In the above example, it's only necessary that the last node in the chain of "/" splits should be a port.

@Tim Edwards With magic 8.3.125, I was able to extract caravel/storage. With magic 8.3.132, I can not. A bunch of these errors.

connect(1): no such node pk_pdriver_3_0\/pk_pinv_16_0/vdd

I checked the ext files, and though there are no / in the use line instance names, there are multiple slashes in the merge line nets. As you suggest, changing / -> _ when gds is read may work. If you let me know when you have a new stable version, and I'll do a regression test on the caravel blocks that have passed so far.

@Mitch Bailey: Yes, revision 132 is effectively hosed. Use 133.