upvote
With the information you're giving, my decision would probably be to take the non-clip TC2030 or TC2050 (I've never needed 14 pins) footprint and overlay the footprint for a regular 1.27mm SMD connector on top of it. Cortex debug connector should be a good fit but I haven't checked.

That seems to be the "get your cake and eat it" (though it does mean you're spending the space and drilling the holes for TC.) But still -

> They're fine for one guy using them on the bench

reply
If you're insistent on Tag-Connect, that's a pretty good way to go. Those legs are a big part of the problem and so the No-Leg version helps a lot. But then it also falls out of the board....

Seriously, I tried to like Tag-Connect. I did like it before supporting a CM and a hardware team trying to use it (and lose the cables...). Now I just plain don't think it adds value over the alternatives. The header is three cents. Three. Cents. The cable is $39 (with legs, $34 without). That buys you over 1,000 headers and then you can use the free cables that come in the box with all the debug probes and live in the pile over there in the shop.

reply
You can get retaining clips to hold it on. While these can be a bit fiddly, I’ve found them good enough for bench testing with no legs tag connect

https://www.tag-connect.com/product/tc2030-retaining-clip-bo...

reply
Bending the little pins also works (they will tend to bend themselves after a little while of use anyway), but at the cost of making insertion a little harder as well. I found that to be the best compromise for me, but YMMV.
reply
To be clear, I'm not criticizing the CM and didn't get hung up on the OP's tone, though I can see how folks would. Sorry about that.

I can understand how folks have issues with TC connectors.

As a counterpoint, I run a small CM and engineering firm and we love Tag Connect.

If you're dealing with truly high volume runs, get the ICs programmed before they're on the boards.

For smaller runs, the clipless TC connectors work great IME.

reply
> If you're dealing with truly high volume runs, get the ICs programmed before they're on the boards.

That is absolutely 100% the way to go.

However, this week I got tremendous pushback on this for a current project with a build volume in the mid six figures... exactly where you'd want to be using pre-programmed parts. It took me a while to figure out what was really happening was the firmware lead (who thinks himself an EE) was spewing out nonsense to cover for the fact that he'll never be ready to deliver in time to preprogram anything. I wish he'd have just said that and saved us all the nasty argument....

reply
That kind of thing definitely gets under my skin more than it should.

I'm sorry you had to deal with that!

reply