Indeed, appending an empty string to -i doesn't modify it in any way, AFAIK.
It's that the GNU version doesn't require an argument, but, if an argument is to be provided, it must be done as part of the flag, not as a separate element of argv. The MacOS version allows either way of providing an argument, so -i.orig tends to be portable (assuming the -i flag is supported in the first place).
Differences in how "traditional" (be they sysv or bsd) versus gnu utilities handle flags and arguments [1] is very well rooted in history, and is hardly unique to sed.
I suspect the main reason this has been forgotten is that Linux, which ships with gnu utilities, has been so dominant for so long, though, even before that, it was difficult to find a then-current unix on which gnu utilities couldn't be installed.
[1] As the sibling comment points out, the source of difference is getopt, of which there were more than just two versions.. including not even using a library
It's that the GNU version doesn't require an argument, but, if an argument is to be provided, it must be done as part of the flag, not as a separate element of argv. The MacOS version allows either way of providing an argument, so -i.orig tends to be portable (assuming the -i flag is supported in the first place).
Differences in how "traditional" (be they sysv or bsd) versus gnu utilities handle flags and arguments [1] is very well rooted in history, and is hardly unique to sed.
I suspect the main reason this has been forgotten is that Linux, which ships with gnu utilities, has been so dominant for so long, though, even before that, it was difficult to find a then-current unix on which gnu utilities couldn't be installed.
[1] As the sibling comment points out, the source of difference is getopt, of which there were more than just two versions.. including not even using a library