Fix bug with hanged pseudoterminal

[solomatov]

This PR fixes #84174
More time to initialize terminal as well as exponential backoff for wait periods.

[Tyriar]

You can run these tests with ./scripts/test-integration.sh (or .bat). To skip the other integration tests you can change this:

vscode/extensions/vscode-api-tests/src/singlefolder-tests/terminal.test.ts

Line 9 in 434662c

suite('window namespace tests', () => {

To use suite.only (but we don't want to commit that).

[solomatov]

@Tyriar It seems that finally I made the tests pass. Could you take a look? :-) Glad that you have the tests. When I made initial change, I didn't find them.

Also, what do you do if a build breaks. Is there a good way to clean everything? I have it all the time when I have yarn watch running, and change branches.

[solomatov]

@Tyriar I also have a question for getting rid of delays in other place. There's this code:

	// TODO: This could be improved by using a single promise and resolve it when the terminal is ready
	private _getTerminalByIdEventually(id: number, retries: number = 5): Promise<ExtHostTerminal | undefined> {
		if (!this._getTerminalPromises[id]) {
			this._getTerminalPromises[id] = this._createGetTerminalPromise(id, retries);
		} else {
			this._getTerminalPromises[id].then(c => {
				return this._createGetTerminalPromise(id, retries);
			});
		}
		return this._getTerminalPromises[id];
	}

Do I understand correctly that the else branch does nothing useful? The then block is activated only if the terminal is found, and the value returned by then has no side effects.

[Tyriar]

Will try review soon

Is there a good way to clean everything? I have it all the time when I have yarn watch running, and change branches.

Typically we don't need to do anything here as TypeScript is pretty good at handling this. If I do get into this state I'll either delete out or nuke everything and start from scratch (git stash or commit then git clean -xfd).

Do I understand correctly that the else branch does nothing useful?

I think that was to restart retrying maybe? That TODO is actually talking about how I proposed we fix this; create a bunch of promises and store their resolves instead of using polling.

[solomatov]

@Tyriar Are there any updates on this? This fix is very important for us.

[solomatov]

@Tyriar Any updates on this?

[Tyriar]

@solomatov I made some changes to simplify the PR. Let me know if this still looks good and if so I'll merge in.

[solomatov]

@Tyriar You know this code definitely better than me, and your changes looks good to me, so feel free to merge.

[solomatov]

@Tyriar Do you need help resolving this merge?

[solomatov]

@Tyriar Remerged again, and hopefully, it will work well.

[solomatov]

@Tyriar Seems merge worked this time.