ChimeraX distribution is too big, over 1 Gbyte uncompressed

[Tom Goddard]

We need to look into trimming ChimeraX bloat. It is absolutely enormous with compressed size 300-400 Mbytes and uncompressed 1.15 Gbytes on macOS. I never saw such a large application. The large download size over slower network connections, slow install, and slow startup are all reasons to take a closer look at the size and start trimming it down.

I think without PySide2 the macOS distribution uncompressed was still 600 Mbytes, so this can't all be blamed on massive Qt libraries. Another clue, site-packages/chimerax is 50 Mbytes in ChimeraX 1.1.1 on Mac. So it looks like a lot of prereq packages are causing the huge size.

[Eric Pettersen]

Things in site-packages taking at least 10MB:

PySide2 459MB
scipy 118MB
numpy 82MB
pydicom 63MB
chimerax 62MB
simtk 42MB
babel 29MB
matplotlib 21MB
imagecodecs 17MB
OpenGL 14MB
tables 14MB
sphinx 14MB

[Eric Pettersen]

numpy and scipy both have a .dylibs subdirectory with the exact same 4 libraries in them, which is a duplication of 62MB.

[Greg Couch]

We could remove the parts of PySide2 that we don't use.  We're probably 
including the GPL parts of Qt as well as the LGPL parts, even though we 
don't use them.

     -- Greg

[goddard@…]

There are developer tools in PySide2 for instance to build UIs from xml descriptions that we could nuke.  But it should be considered whether ChimeraX is intended to include that stuff for third-party developers to make use of.

[Tom Goddard]

The current Mac ARM dmg distribution is 365 Mbytes, and 1 Gbyte uncompressed (ticket #7222), with the Mac universal build 569 Mbyte dmg, and 1.6 Gbytes uncompessed. The 10 largest packages in site-packages for the ARM Mac build are

$ cd site-packages/
$ du -hs * | sort -rh | head
du -hs * | sort -rh | head
341M	PyQt6
 81M	scipy
 79M	chimerax
 46M	numpy
 44M	openmm
 31M	babel
 21M	debugpy
 17M	matplotlib
 14M	OpenGL
 12M	tables

and QtWebEngine is 189 Mbytes.

189M	PyQt6/Qt6/lib/QtWebEngineCore.framework