klayout throws an ERROR reading a magic file with "tri ... sw"

[stefanottili]

It seems that something goes wrong reading "tri ... sw".

https://github.com/RTimothyEdwards/gf180mcu_ocd_sram_test/blob/main/magic/wafer_space/gf180mcu_ws_ip__id.mag

klayout magic/wafer_space/gf180mcu_ws_ip__id.mag
ERROR: Expected end of text here: w (line=8, file=magic/wafer_space/gf180mcu_ws_ip__id.mag)

   1 magic
   2 tech gf180mcuD
   3 magscale 1 5
   4 timestamp 1765407047
   5 << metal1 >>
   6 tri 0 14144 136 14280 se
   7 rect 136 14144 544 14280
   8 tri 544 14144 680 14280 sw
   9 tri 680 14144 816 14280 se
  10 rect 816 14144 1224 14280
....

src/plugins/streamers/magic/db_plugin/dbMAGReader.cc

483 void  
484 MAGReader::read_tri (tl::Extractor &ex, Layout &layout, cell_index_type cell_index, unsigned int layer)
485 {     
486   double l, b, r, t;
487   ex.read (l);
488   ex.read (b);
489   ex.read (r);
490   ex.read (t);
491       
492   bool s = false, e = false; 
493   while (! ex.at_end ()) {
494     if (ex.test ("s")) { 
495       s = true;
496     } else if (ex.test ("e")) {
497       e = true;
498     } else {
499       break;
500     }
501   } 
502   ex.expect_end ();

[stefanottili]

The following code seem to read the .mag file correctly, assuming se|sw|ne|nw.

klayout writes this layout as a .mag file with s and se ... that seems to be a separate issue.
And compared to the full chips gds, the units are off too.
How would I map the mag layer names to gds/oas layer, datatype integer ? lyp ?

klayout gds/gf180mcu_ocd_sram_top.gds.gz -s magic/gf180mcu_ocd_sram_top.mag.gz
is missing some cells referenced by the mag, but shows that units and layers are not aligned.

492   bool s = false, e = false;
493   while (! ex.at_end ()) {
494     if (ex.test ("se")) {
495       s = true;
496       e = true;
497     } else if (ex.test ("sw")) {
498       s = true;
499     } else if (ex.test ("ne")) {
500       e = true;
501     } else if (ex.test ("nw")) {
502     } else {
503       break;
504     }
505   }

[klayoutmatthias]

I honestly don't know what "sw" etc. means. There is some documentation (http://opencircuitdesign.com/magic/manpages/mag_manpage.html), but it doesn't even mention "tri".

Scanning the code suggests that unknown characters simply should be ignored and only "s" and "e" are needed (https://github.com/RTimothyEdwards/magic/blob/0cbed6078a6e28f52c8457c4c32e2505ae5baa74/utils/getrect.c#L118). So the writer should be fine and the reader should simply ignore other characters.

[stefanottili]

Tims mag files show that magic writes nw,ne,se,sw.
I would think its meaning is north-west, north-east, south-west, south-east.
With my code change klayout reads all files and the triangles look correct.
The id cell contains qcode made of octagons that show this.

The scaling will be different on a per file basis.

[klayoutmatthias]

The "magscale" command isn't documented either in the man pages. From the code it looks like the two numbers are nominator and denominator.

[klayoutmatthias]

When I look at the files from the sample project, there is this:

magic
tech gf180mcuD
timestamp 1765407464
use sealring$1  sealring$1_0
timestamp 1765317384
transform 1 0 0 0 1 0
box 0 0 78640 50620
<< properties >>
string GDS_END 28412362
string GDS_FILE ../../gds/sealring.gds.gz
string GDS_START 28412324
<< end >>

looks like the Magic format has much evolved beyond the documentation ...

[klayoutmatthias]

I created a patch, fixing a number of other issues, but I feel that it's a somewhat wasted effort.

[stefanottili]

I feel that it's a somewhat wasted effort

I disagree, I feel it's very nice to be able to "have a look" at .mag files, after all it's layout data (rect, tri, text, hierarchy).
Considering that magic was created in 1983, it shows remarkable longevity ;-)
I do agree that the gds/cif <-> mag conversion seems to be just some hard to explain tech file magic.

export __TESTSRC_ABSPATH=~/klayout
klayout gf180mcu_ocd_sram_top.mag.gz -s ../gf180mcu_ocd_sram_test.cif.gz`

shows that the units are different between the two layouts and I don't know how to map the layers.