compiling files in ghci on MacOS eventually results in malformed mach-o: load commands size (32800) > 32768

compiling files in ghci on MacOS eventually results in a panic with error:

malformed mach-o: load commands size (32800) > 32768

I can reproduce this problem by repeatedly compiling (and loading) a file and then deleting the .o file. I understand that this an OS limitation but why does repeatedly doing the same thing results in this? i.e. why don't I get it on the first compile (and load). It seems that successive loads are being appended to something eventually resulting in the error.

To duplicate:

ghci -ignore-dot-ghci -fobject-code < ghci.sh >& ghci.out &

If you look for the first "panic" in the .out file you'll see

Prelude Main> ghc: panic! (the 'impossible' happened)
  (GHC version 8.6.0.20180714 for x86_64-apple-darwin):
	Loading temp shared object failed: dlopen(/var/folders/9b/rh4y2gy92hgdb6ktv4df1jv00000gn/T/ghc55569_0/libghc_2084.dylib, 5): no suitable image found.  Did find:
	/var/folders/9b/rh4y2gy92hgdb6ktv4df1jv00000gn/T/ghc55569_0/libghc_2084.dylib: malformed mach-o: load commands size (32800) > 32768
	/var/folders/9b/rh4y2gy92hgdb6ktv4df1jv00000gn/T/ghc55569_0/libghc_2084.dylib: stat() failed with errno=25

The critical part being

malformed mach-o: load commands size (32800) > 32768

My understanding from https://stackoverflow.com/questions/44902695/too-many-commands-dyld-message-malformed-mach-o-load-commands-size is that this has to do with sizeofcmds but if I understand the output of otool -l this is only 512 so I don't understand what is causing the errror.

 otool -l c.o|head
c.o:
Mach header
      magic cputype cpusubtype  caps    filetype ncmds sizeofcmds      flags
 0xfeedfacf 16777223          3  0x00           1     4        512 0x00002000

This is irritating but perhaps the priority should be low rather than normal.