Normally, if I promise you something, you can release me from my promise. But suppose that I promise to fire you if you are once again inexcusably late to a meeting with upper management. And you are inexcusably late. It would have been a pointless promise if you could release me now!
So, either that wasn't a promise or not all promises allow for release by the promisee. I want to explore the second option. So it was a promise but I can't be released.
Consider another case. I promise you that if it rains during a full moon, I will fire you. My intuition is that in the lateness case, the promise takes—I am bound. But the full moon case is not a case of a successful promise (at least under normal circumstances where there is no benefit to you of being fired; we can imagine circumstances in which you expectantly await the full moon and pray for rain)—I have no obligation to fire you, even if your contract allows me to fire you for no cause.
Why the difference? My best explanation is that in the lateness case, your punishment is a benefit to you. It is good for one to suffer a just punishment (I am assuming it's just—if not, then I think the promise doesn't take). And only benefits can be promised.
But that only heightens the first puzzle. If I promise you a good, can't you sacrifice that good, thereby releasing me?
There is another case. You promise me that if I ever become cynical and lose my ideals, you will try to convince me to return to them. I have become cynical and lost my ideals. It would be silly to think I could release you from your promise so as to avoid bothering with your arguments.
This and the punishment cases are cases where the good that is promised is one that the promisee doesn't want when the good is bestowed. In the punishment case, perhaps the promisee never wanted it, while in the ideals case the promisee wanted it but lost the desire.
Maybe there is no such thing as a general release condition on promises? Maybe it's just typically an implicit part of the promise, unless either the content or the context or explicit speech cancels it. But when we promise punishment or convincing, the content of the promise removes the implicit "unless you don't want me to"?
Here's another idea. Catholic canon law says that a private vow to God can be commuted to a better vow. If I vowed to give a dollar to the soup kitchen, I can commute it to volunteering for a weekend. Normally, even if there is no "unless you don't want me to" qualifier in a promise, the promise becomes better when the qualifier is added. So normally the promise can be released from, since I can just add the qualifier, thereby improving the promise, and then get released by your triggering the exception clause. But adding the "unless you don't want me to" clause to the punishment or returning-to-ideals promises doesn't make the promise better. It makes it worse.
Note that it is essential to this solution that being punished is good for the punishee.