toke.c: Return to pre-existing gcb rules

[khwilliamson]

As stated in the C comments added in this commit, it has always worked to say s/mark1/mark2/ where the marks are literal characters. This is incorrect because the marks should be considered modifiers of the '/' delimiters. (The substitute should be changed to do something like s/\x{foo/\x{bar}/.) But that may not happen in time for 5.36. So this commit reverts to the behavior prior to 32b8779 for the ASCII delimiters that have always worked this way. Non-ASCII delimiters will continue to work the new way, which could break code expecting the old. But use of non-ASCII delimiters is pretty rare, and the users of that could be expected to have more knowledge of Unicode and would be less likely to do something contrary to the Unicode Standard than someone using the typical ASCII delimiters.

[jkeenan]

Was there a specific commit at which we moved from the "pre-existing rules" to the existing rules?

[khwilliamson]

I revised the commit message to include the breaking commit number

[rjbs]

As I commented on #19554, I think we should leave blead as is and view this as as correction that does not need a long warning period.

[demerphq]

Just to be clear, @rjbs you are saying we should close this PR and not apply?

[rjbs]

That's my personal opinion. I could certainly be outvoted, and possibly swayed.

[demerphq]

Its been a long time with no activity on this, so I am going to close the ticket. @khwilliamson if you object please reopen.