Hi Eli,

> Thanks, but could you please show the minimal change required to 
fix
> just the particular problem with this scenario

I've attached a new self-contained patch based on yours,
thank you for coming up with it.

In my opinion, one of the problems is really as you said
that len_mid (length in bytes)
is uses where length of characters is expected.
The patch contains an additional such case,
and replaces (start1 + len1) in your patch with the equivalent
but shorter end1.

The other problem is that the branch len1_byte == len2_byte
assumes len1 == len2 in several places:
undo records, positions after the transposition,
and the least obvious one is that even intervals between the 
region seem to need adjustment of positions.
(I don't know enough of intervals to understand it, but had failed 
tests with text properties at wrong places.)

Anyway, I've just added len1 == len2 as a condition to that branch
with comments where I think the assumption is used.
I've also added a new test for this case.

> The patch
> below passes both your test and the already-existing tests in
> test/src/editfns-tests.el.

For me after applying your patch, the tests crashed.
The crash message was hidden in the end of the output:

   passed  21/24  transpose-nonascii-regions-test-1 (0.000067 sec)
   passed  22/24  transpose-nonascii-regions-test-2 (0.000068 sec)
   passed  23/24  transpose-regions-text-properties (0.000074 sec)
Undo
Undo
make[1]: *** [Makefile:181: src/editfns-tests.log] segmentation 
fault
make[1]: Leaving directory "/home/gabor/src/build/emacs-29.3/test"
make: *** [Makefile:247: src/editfns-tests] Error 2


With the tests I intended to test all the branches in the code
where the regions don't touch each other, catching mistakes
where the wrong length is used.
I hoped that byte length is not system dependent,
it seems I have been mistaken, and the tests still need
improvement to be thorough on all systems.

So what differences exist or byte length?

Best wishes,

	Gábor