From a speech by Australian politician David Leyonhjelm, but I've heard it elsewhere many times:
The senator borrowed other stories from history in the speech, opening with details of the old practice of using ducking stools to detect witchcraft. Ducking involved tying a woman to a chair and dunking her in water. If she floated, she was a witch. If she sank, she wasn’t, but she was probably dead.
TV Tropes is skeptical of this belief, and related versions. It explicitly states that some people are of the belief that testing for witchcraft involved the suspect either drowning and being declared innocent posthumously, or floating and being executed for witchcraft, and describes that belief as erroneous. It also says ducking chairs aren't involved. Unfortunately, TV Tropes isn't a sufficiently reliable source.
The "swimming" of witches, one of the most famous methods of interrogating a suspected witch, had the virtue of being both pointless and redundant. Popular belief makes it out as a Morton's Fork, saying that if the 'witch' floated, they'd pull her out and kill her. If the "witch" drowned, on the other hand... well, they were still dead, they just weren't a witch. Actually she would be tied to a rope: if she did float, they would pull her out, and the fact would be regarded as incriminating (of course sometimes they wouldn't do quick enough, and she'd still drown-"floating" could also be achieved by trickery with the ropes). If she sank, they would pull her out all the same, but cleared of charges. The ducking stool is an unrelated, non-lethal device of punishment where a woman was ducked in cold water for being a public nuisance of some sort.
I'm not really worried about whether ropes were used, or whether the ducking stool was used.
Did testing for witchcraft involve a scenario which was guaranteed (say >90% probability) to be fatal to the suspect, either because they drowned, or because they would get executed?