... continued ...
Description:
When a PKGBUILD has source_i686 and source_x86_64 updpkgsums tends to remove sums and put the wrong ones back.
* md5 is reasonably well supported. sha1 is somewhat supported. sha256 is not supported and gets replaced with md5. sha512 is wider than 80 columns so I don't take much interest in them.
* updpkgsums removes all sums and puts only the sums in for the current $CARCH.

makepkg seems to be to blame too since it won't output both sums

Additional info:
* package version(s)
* config and/or log files etc.
updpkgsums 4.2.1

Steps to reproduce:
See basis-pro5 on AUR for an example of what I had to do to keep updpkgsums greasy mitts off my sums. I had sha1sums working but they don't work now so I switched them to md5sums.

This works just fine for me, I do it all of the time with sha256sums. It appears to me that the problem is in the extremely overcomplicated PKGBUILD.

Attach an example PKGBUILD before and after running updpkgsums

Software makers just assume that you'll "sudo make install" and let the makefile throw slop all over the file system. That's fine for the systems that are reformatted every upgrade which are the ones they prefer to support. The over complication let me improve on the original install and make upgrades and crossgrades between i686-multilib and x86_64 seamless. Arch Linux users expect no less.

Here's a test package that shows the all->md5 bug and the annoyance where updpgksums writes all sums together. I was able to fix the PKGBUILD so makepkg -g outputs sums for both architectures.

This is a small patch that retains the hashes used in the PKGBUILD. If you have mixed hashes, then it will generate all types for each architecture, e.g.

md5sums_i686
sha256sums_x86_64

results in

md5sums_i686
md5sums_x86_64
sha256sums_i686
sha256sums_x86_64

I did not try to keep the hashes at their original locations.