[Dustin L. Howett (MSFT) (DHowett-MSFT)]

Fixes #4107.

[Dustin L. Howett (MSFT) (DHowett-MSFT)]

(thanks James Holderness (@j4james))

I chose to flip the sense here instead of rewriting the function. Options include...

bool success = SUCCEEDED(x) && SUCCEEDED(y);

if (success)
{
    success = SUCCEEDED(a) && ...
}

if (success)

and

THROW_IF_FAILED();
THROW_IF_FAILED();
THROW_IF_FAILED();

(this works because we have a CATCH_RETURN_FALSE)

[James Holderness (j4james)]

I was going to say you actually shouldn't even need those SizeTToShort checks, because a VT parameter can never be greater than a short (it should be clamped by the state machine), but I see now that PR #3956 changed that behaviour. So instead of being pointless, these checks are now also wrong.

If a parameter is greater than a short, it should be clamped rather failing, i.e. you should be able to use a sequence like \e[1;999999H to move to the top right corner of the screen. That used to work, but doesn't anymore.

And note that CUP won't be the only operation broken by that change (some may even crash the conhost). Either they all need to add their own parameter clamping code, or we should add the clamping back in the state machine (which would be my preference). That said, it's probably best to get this particular fix merged ASAP and worry about the clamping in a followup issue.

[Leonard Hecker (lhecker)]

Ah dammit I was about to submit a PR myself after debugging this issue today. 😣
(In fact PowerShell was broken entirely due to this change. It returns cursor position sequences on each input!)

I agree with James Holderness (@j4james) that values should be capped because such a lenient behavior is probably what you'd want anyways. From a cursory glance SetCursorPosition seems to be robust against negative values too. For instance this is what I'd personally imagine for this PR:

bool TerminalDispatch::CursorPosition(const size_t line,
                                      const size_t column) noexcept
{
    const auto clamp = [](const size_t val) noexcept -> short {
        constexpr size_t min = 1;
        constexpr size_t max = std::numeric_limits<short>::max();
        return static_cast<short>(std::clamp(val, min, max) - min);
    };

    return _terminalApi.SetCursorPosition(clamp(column), clamp(line));
}

(It's certainly not beautiful code, so please feel free to use it. 😄 A wrapper class with saturating arithmetic would surely be helpful here in the future.)

[James Holderness (j4james)]

Leonard Hecker (@lhecker) Note that the CursorPosition method is just one example of the clamping problem. We're going to have the same issue in many other operations.

A wrapper class with saturating arithmetic would surely be helpful here in the future.)

Yeah, that's being discussed in #4013. In the sort term, though, adding the clamping back in the state machine would at least solve a lot of the problems.

[Mike Griese (zadjii-msft)]

James Holderness (@j4james) Leonard Hecker (@lhecker) those are both good suggestions, though I think I'd probably want to get this fix merged in ASAP, since the Terminal's pretty horribly broken without it. I'd probably just file a follow up with those suggestions personally, though I'd be curious to hear Dustin L. Howett (MSFT) (@DHowett-MSFT)'s thoughts on the matter.

[Michael Niksa (miniksa)]

We'll talk about prioritizing #4013 for this week since it's becoming a problem and will support other work.

[Dustin L. Howett (MSFT) (DHowett-MSFT)]

We'll likely end up doing 4013 this week.

[Dustin L. Howett (MSFT) (DHowett-MSFT)]

This is an unblocker.