`exec` command does not work if the debugger is suspended on a `COPY` line

[rcjsuen]

docker buildx version
github.com/docker/buildx v0.29.1-desktop.1 28f6246ff24e2c05095e8741e48c48dcb2d3b4bc

FROM alpine:3.22
COPY . .

1. Create a Dockerfile.
2. Put a breakpoint on the second line with COPY . ..
3. Punch in exec in the Debug Console.
4. The terminal will open but it then just hangs.

Might be related to #3437.

[rcjsuen]

The same thing happens with ADD...but I assume it is the same code. :)

[rcjsuen]

And also if you use stopOnEntry or if your breakpoint is on a FROM line from what I can tell.

[jsternberg]

I just need to make this more friendly. What happens is during the first attach it waits peacefully for the first thing it can exec and doesn't say anything. Some steps don't have anything exec-able. The FROM line hasn't run yet so you can't exec anything and COPY is not currently something that can be exec'd. Only RUN lines are exec-able right now.

If you exec on a RUN and then continue it will attempt to reattach. If it can't reattach due to some reason, it'll tell you. I suppressed that message on the first attach because it was quite noisy and often gave nonsensical advice. I'll just have to retool that so the attach can know more information so it is not too verbose but a bit more verbose than it currently is.

I've done some research on #3437 and they're definitely unrelated.

[jsternberg]

Seems there might be a bug here. Even if you do stopOnEntry and do "step over" repeatedly, it skips over the copy. That indicates it's not finding a stopping point here. I'm investigating this as a bug now.

[jsternberg]

Expanding on this more as I work on it.

The good news is I think I've got a grasp of the problem. This bug cropped up when we switched the breakpoints to stop before the step and not after. What happens is the little program the debugger builds to map the digests to something that progresses iteratively also inserts some dummy locations in the progression. This is what controls step in, step next, step out, etc.

The problem is to facilitate stopping before the instruction is run, we didn't change the program. Instead, I just shifted it to check if the next instruction matched the digest and had it run the step right before. The problem is some steps don't actually have digests associated with them. So using continue here skipped to the step right before the digest which doesn't have any digest associated with it so it doesn't run anything. Since it doesn't run any solve, there's nothing to exec.

This is definitely a bug and will require a bit more of a change to the debugger than I anticipated. I can't promise this will be fixed in a patch release as the size may be too big to be worth it. I can promise it'll be fixed in at least the next minor version.