Check that a chromote connection is alive

[hadley]

When your computer goes to sleep, the websocket and all debugging sessions are closed. This PR ensures that Chromote and ChromoteSession report this accurately.

I distinguished between active and alive for Chromote but not ChromoteSession since the browser process continues to run in the background, even though we have lost our connection to it. There's no way to restore a closed debugging session. However, I'm not sure about the names; maybe I have alive and active around the wrong way?

Part of #94

---

I tested this code with:

library(chromote)

sess <- ChromoteSession$new()
sess$is_active()
#> [1] TRUE
sess$parent$is_alive()
#> [1] TRUE

# SLEEP

sess$is_active()
#> [1] FALSE
sess$parent$is_alive()
#> [1] FALSE

[hadley]

Now Chromote will reconnect if the websocket is dead but the underlying process is alive:

library(chromote)

chrome <- Chromote$new()
chrome$is_alive()
#> [1] TRUE
chrome$is_active()
#> [1] TRUE
chrome$check_active()

# SLEEP -----------------------

chrome$is_alive()
#> [1] TRUE
chrome$is_active()
#> [1] FALSE
chrome$check_active()
#> ! Reconnecting to chrome process.
#> ℹ All active sessions will be need to be respawned.
chrome$is_active()
#> [1] TRUE

chrome$close()
#> [1] FALSE
chrome$is_alive()
#> [1] FALSE
chrome$is_active()
#> [1] FALSE
chrome$check_active()
#> Error in `chrome$check_active()`:
#> ! Chromote has been closed.

[wch]

Overall I'm OK with this change, but we will want to make it very clear to the user what the difference is between is_alive and is_active.

Perhaps the print method for Chromote could also print out that information? Like Process is [not] alive and Websocket connection to Chrome is [not] active.

And we'll also want to mark it as a breaking change in the NEWS file.

[hadley]

I've addressed your concerns, apart from the print method, which I agree is a good idea and I'll do as part of #140, once this PR is merged.

Since your last review, I've added some additional logic to ChromoteSession to distinguish between the session and target being closed. This means that you get a reasonable error message if you $close the session (and hence the underlying target) and then try to $respawn() it. This is just a static value I cache in R, so it'll be incorrect if you close the target some other way, but I think that'll be uncommon and thus not worth handling. Let me know if you disagree.

library(chromote)

sess <- ChromoteSession$new()
sess$is_active()
#> [1] TRUE

# SLEEP -----------------------

sess$is_active()
#> [1] FALSE

sess$check_active()
#> Error in `sess$check_active()`:
#> ! Session has been closed.
#> ℹ Call session$respawn() to create a new session that connects to the same target.

sess$close()
#> ! Reconnecting to chrome process.
#> ℹ All active sessions will be need to be respawned.

sess$respawn()
#> Error in `sess$respawn()`:
#> ! Can't respawn session; target has been closed.
#> [1] TRUE

[wch]

It looks like target_closed is unused?

[hadley]

Doh. I realised that I forgot to call spawn() before closing in my script test. Here's the new output:

library(chromote)

sess <- ChromoteSession$new()
sess$check_active()
#> NULL

# SLEEP -----------------------

sess$check_active()
#> Error in `sess$check_active()`:
#> ! Session has been closed.
#> ℹ Call session$respawn() to create a new session that connects to the same target.

sess2 <- sess$respawn()
#> ! Reconnecting to chrome process.
#> ℹ All active sessions will be need to be respawned.

sess$close()

sess$respawn()
#> Error in `sess$respawn()`:
#> ! Can't respawn session; target has been closed.

[wch]

What do you think about target_is_alive? (I'm asking honestly, not suggesting that it should or should not be that name.) I ask because it might help conceptually separate the two alive names. Also the target exists independently of any connection to it, sort of like how the Chrome process exists independently of a connection to it, and in that case we call it alive.

[hadley]

I don't mind changing the name, but calling it alive does introduce another inconsistency — Chromote$is_alive() is always accurate because we have a connection to the process that we can check; but there's currently no analog for the target. I could add an is_alive() method that checks if the target ID is found in the chromote instance, but I'm not sure if that's worth it (plus calling that repeatedly might be expensive).

So while I agree that "active" isn't the best name, and has some potential for confusion, I think changing it to "alive" is a net neutral because it just introduces a different source confusion.

[wch]

OK, let's keep it as "active" then.

[wch]

I think the overall approach is good -- just have a couple questions.