The big technical issue with swim watches is the way they register a new length, but there are also simple design issues.
As far as the technical issue is concerned, Garmin seems to be without any serious rival as far as I can see. Tom Tom entered the market but seems to have an larger problem than even v.2.x of the Garmin formware. With v.2.x, registering lengths was hit and miss especially (and as acknowedged by their techies on their forum) you are a TI swimmer. Poolmate doesn't work at all for SPLs lower than about 16 (from memory - could be 14).
Happily v.3.0 of the Garmin 910 firmware fixed the major part of the problem - with it the watch(es) now count lengths pretty flawlessly. But that isn't the same as saying the problem is fixed. And the problem ripples out into every other part of the data gathering.
For whatever reasons, real or imagined, the algorithm used to interpret the accelerometer output(s) requires a period of time in which something different occurs e.g a turn PLUS a glide. For the watch then, a new length is "begun" about 6m after the turn. Its algorithm then makes a guess in order to compensate for this, with predictable results. There may be a real problem or there may be inadequate pressure from the market. As v.3.0 solved a substantial part of the problem with no impact on the system overall my opinion leans to the latter.
This view is reinforced by the design issues. The arithmetic for instance is incorrect, something which must be inexcusable. If I swim 25m in 30.00 secs then my 100m pace is 2:00.00 mins if the watch shows otherwise it is absurd, yet it does so and at all interval lengths. The only reliable data is the duration of an interval and (almost always anyway) the total number of lengths.
Recently I've realised that the watch display truncates SPL for intervals, rather than rounding. This leads to inaccurate feedback during a session. I am unable to imagine any excuse for this as there is room on the display. The stroke count generally seems pretty reliable excpet for this but obviously it needs interpreting as it's only for one arm.
Stoke identification is another matter as it relies on quite detailed assumptions about stroke patterns. For me it's a frill that could be dispensed with and perhaps that would help get a better algorithm for the turn.
A psychological disorder is: "Any personal construction which is used repeatedly in spite of consistent invalidation."
~ George Kelly
"The water is your friend.....you don't have to fight with water, just share the same spirit as the water, and it will help you move."
~ Aleksandr Popov