I've written a "window server" for the Apple II (7-pixel aligned, window stack you could only add and remove rectangles from the top) with about 1K of code. It was not that difficult to reason about, but I had a much younger brain at the time.
But on top of that, a lot of software on the C64 at least (I'd like to think this didn't apply to any "professional" software, but I suspect you'd find a lot of hairy stuff there too) was not written with a macro assembler but directly into a machine code monitor with no ability to store labels or comments.
In many cases you'd be lucky to have a sheet of handwritten notes about which addresses contained which code (and yes, that meant re-writing bits and pieces of code if you wanted to insert something new).
For my part I used a machine code monitor without labels etc. for several years before I got a proper macro assembler (Turbo Assembler) with a full screen editor.
[1] For those who don't have any exposure to 6502 assembler, here's a basic subroutine to copy blocks of memory - after ~25 years of not touching it so bear with me (and corrections welcome) especially with respect to syntax.
Lets say we have the high byte of a 16bit address we want to copy from in the A register, and the high byte of a 16bit address we want to copy to in the X register (often it wouldn't be unusual for the calling code to directly modify the src/dest addresses instead to set up the loop...), and for simplicity the low byte of both is 0, and the number of 256 byte chunks we want to copy in Y.
I'm sure I've forgotten some more efficient way of doing this - it was all cycle-counting all the time, and lots of ugly (uglier) tricks, but the basic mechanism of directly modifying the high bites of addresses all over the place to deal with loops beyond 256 iterations was pretty common.