`RRuntime` `execute()` requests don't populate the Console history correctly

[DavisVaughan]

Discovered while working on https://github.com/posit-dev/positron/pull/2007

It seems like the code sent through execute() isn't populated immediately in the console history, but is there after a restart of positron

https://github.com/posit-dev/positron/assets/19150088/aaa53e6e-3637-4ebe-8dde-fd8c1f0c73b9

[DavisVaughan]

Ah I wonder if the history issue is due to the fact that when you call a runtime's execute() method directly, this _onDidExecuteCodeEmitter event is not fired (it is called in the other "blessed" paths where runtime.execute() is called) https://github.com/posit-dev/positron/blob/9ae3137d234058543f9f075e76cbada1b3289498/src/vs/workbench/services/positronConsole/browser/positronConsoleService.ts#L1830

It seems like it is highly related, and was a part of https://github.com/posit-dev/positron/pull/1663

---

My gut tells me that we need to move some of this behavior down into the kernel execute() method. https://github.com/posit-dev/positron/blob/74f52bab1a5103425f5ddba6b462d0fbbb58a3f3/extensions/jupyter-adapter/src/JupyterKernel.ts#L897

Like, if store_history is TRUE, then we do some kind of signal to store the result in the execution history.

If the signal came from there and was propagated upward, that would be low enough that an RRuntime execute() call would trigger it, making that method safer to use.

---

I think ideally I'm looking for some solution that ensures a runtime's execute() method exactly matches what happens when a user types code interactively at the console (at least if the Interactive mode is set)

[DavisVaughan]

#2007 is now merged, so a simple reprex to reproduce this is:

Run

cli::cli_text("{.run rlang::abort('oh no')}")

Click the code hyperlink to automatically run the code

Press Up to go to the previous command in the History. It should be rlang::abort('oh no'), but is not

[juliasilge]

We talked at one point about whether it is a good idea at all to be able to execute with an RRuntime (for exactly the reasons you outline here) but we use it quite a bit now. 😬

[testlabauto]

Verified Fixed

Positron Version(s) : 2025.07.0-200
OS Version          : OSX

Test scenario(s)

Code sent via execute is not immediately available in console history.

Link(s) to TestRail test cases run or created: