layout.read twice segfaults for a normal GDS

[thomaslima]

When I do layout.read twice from python in a particular gds file, the second time it segfaults for some reason.
I am attaching the gds file here. It contains polygons and a Basic.TEXT pcell.

inductor_1030pH.gds.zip

[thomaslima]

Luckily I realized my mistake:

https://www.klayout.de/forum/comments.php?DiscussionID=146

But it should still not segfault in my opinion.

[klayoutmatthias]

You're right, segfaults shouldn't happen. I just wonder what's the mistake in particular ... the discussion mentioned is a bit lengthy :-)

[thomaslima]

Hi Matthias. It was about not passing the layer map as a read option. That fixed this particular read.
But let me ask you this. Maybe it is minor, but when I read a second "almost identical (different names)" layout with a Basic.TEXT pcell, the first Basic.TEXT pcell gets replaced with the new one, instead of there being an extra one added.

[klayoutmatthias]

I see .. actually the layer map should not be required. Reading should be perfectly fine. But maybe there is an issue when the file is identical.

I need to look into the PCell replacement issue, but I'm afraid that the "read twice" approach is not safe against PCell merging. That's somewhat intrinsic - the "read twice" merge approach is based on a low-level feature that allows "reopening" of cells: when a cell is encountered a second time, the content of the new cell will be read into the first one. That's unsafe in general: if you accidentally used the same cell name in two layouts, reading the second into the first will combine those cells into one, whether that makes sense or not. In case of PCells the actual cell names are not carrying the PCell parameters to be compatible with GDS. So the cell may be called "Basic.TEXT" in both cases, regardless of the actual parameters and KLayout will decide for one parameter set. That's why the cells get messed up.

A safe approach is to load the the layouts into two Layout objects and use "copy_hier" to copy over one layout into the other. But I'm afraid this feature does not preserve the PCells as such - they are turned into normal cells.

[thomaslima]

Sorry just to clarify, do you mean 'copy_tree'? That works, I believe. Except that for my application case I wanted to save individual cells with shared instances into individual .gds files and upon loading them together, the instances would point to the same cell.

[klayoutmatthias]

Hi Thomas, I found the root cause of the segfault and I think I can provide a solution for the "read atop" approach also with PCells. At least for GDS2. But if you want to merge the "same cell" into one, the "read atop" approach is questionable as it would load the same cell twice (or many more times). So essentially the content would be duplicated. For a quick solution this is feasible, but I think in general, a smarter approach is required: load all layouts individually, identify the cells that are shared (i.e. by name or if a PCell by their parameters), prepare the shared cells in a destination directory and then copy over the top cells while mapping the child cells to the shared ones. Sounds a bit complicated, but I guess this is the safe way of doing it. Best regards, Matthias Zitat von Thomas Ferreira de Lima <notifications@github.com>:

…

Sorry just to clarify, do you mean 'copy_tree'? That works, I believe. Except that for my application case I wanted to save individual cells with shared instances into individual .gds files and upon loading them together, the instances would point to the same cell. -- You are receiving this because you commented. Reply to this email directly or view it on GitHub: #121 (comment)

[thomaslima]

Thank you for looking into this. =)