[sheet-] in execCommand(), fail if cmd does not exist

[midichef]

A command that does not exist, like qqquit-sheet, will be skipped during replay of a macro or command log, but the subsequent commands will continue to be replayed.

Some ways that could happen: a user makes a typo editing a command name in a macro or command log. Or in the past, the user defined a command to be used in a macro, for example, loggable-quit-sheet, and now they delete the command definition, breaking the macro that uses it.

It seems better to fail() instead of warning() in this case. Is there a case where warning() is better?

[midichef]

For this PR, I needed to make changes to get tests to pass.

One test was trying to run commands that couldn't be found. It was running commands on a test TsvSheet that the sheet didn't have, like commands for CanvasSheet or ColumnsSheet. I've made it skip those, there are around 150 of those commands.

The other test that started to fail was the macro test, test_macros.vd. It had been quietly not testing anything, because it had been trying to run a command macro. I cleaned it up and changed it to run a macro 1.vdj via a command exec-1.

But then I ran into another problem, which is that macros couldn't be run when --batch was used. Macros were only loaded when run() executes, which doesn't happen in batch mode. So now they get loaded, by vd.reloadMacros().

Note that there is still one case where macros aren't loaded, which is bare --batch with no --play. That should be fine, since there are no commands, so there's no way to run a macro.

[saulpw]

I have a vague recollection that fail was not desired in replay at one time. Maybe it was that some commands would fail, but you wanted them to continue anyway? Like select-regex, if nothing matches, you can still delete-selected as a no-op, but you don't want to abort the replay entirely.

That's almost entirely tangential to this, but it makes me wonder if something like this should be error instead of fail? There's almost no difference between them--error is red while fail is yellow. But we could use them more semantically, in case we want to catch fails and continue the replay, vs let errors abort the replay.

Either way, I agree that an unknown command should abort the replay.

[saulpw]

Thanks for fixing this, @midichef !