Repository navigation
8.3.700
commit 8.3.700
"Tagging version 8.3.700"
See documentation at https://github.com/RTimothyEdwards/magic/blob/f790f2f1/appimage/7/README.md
See documentation at https://github.com/RTimothyEdwards/magic/blob/f790f2f1/appimage/8/README.md
See documentation at https://github.com/RTimothyEdwards/magic/blob/f790f2f1/appimage/9/README.md
See documentation at https://github.com/RTimothyEdwards/magic/blob/f790f2f1/appimage/10/README.md
Change Log (since previous tag):
f790f2f1 Claude Opus 5.5 found and fixed a handful of errors where split tiles were not handled in the "netmenu" window (which doesn't surprise me at all); the handling is straightforward and checks for split tiles before applying the type of the side indicated by the diagonal info.
0c46bb0e Fix missing dinfo parameter in five tile-search callbacks
4b919797 I got Claude Opus 5.5 to analyze PR #572 which solves a crash condition; after looking at the solution I decided that my full-plane searches in bloat-all were inefficient, and Claude replaced them with a tile reset stack. Claude also found several other places in the CIF generation code that had the same issue with scanning an entire plane to unmark tiles, and replaced the scan with a reset stack. This both solves the original segfault problem and also speeds up CIF generation.
4ba66a6d wheel: fix last set of compatibility concerns
900c9286 wheels: fix python 3.8 + build perf improvements
9508d0e6 fix linux builds
f810b60c wheel: build with Tcl 9.1 instead, dynamically pick graphics-free build
df4bcf50 Initial work on py3-none wheel
f15b6cf0 extresist: make triangle selection independent of pointer hashes
13a136c4 I had repeatedly run into issues with log-running connectivity searches, and working with full-chip layouts recently, this had become a serious issue. I had Claude Opus 5.5 analyze a connectivity search and identify issues. Two routines doing linear searches were changed to use hash table lookups. The main point of congestion, however, was due to searches being repeated over the same tiles numerous times. This was fixed by marking tiles as having been processed. For large nets like chip-wide power supplies, search times have been reduced from hours to minutes; the largest network I looked at now processes all connectivity in under two minutes. Further optimization would require significantly more effort.