R devel - Apr 2025 - on.exit() handler being interrupted by time limit

This StackOverflow question:  https://stackoverflow.com/q/79594532 and 
this discussion on Github:  https://github.com/r-lib/evaluate/issues/242 
are talking about an interesting situation.

Under a fairly simple setup shown in both, setTimeLimit() can interrupt 
the on.exit() handler that was installed by evaluate::evaluate(), 
leaving the system in a weird state.

Here's the setup:

If the document

````
```{r}
ans <- 6 * 7
Sys.sleep(2)
```

Answer: `r ans`
````
is saved in timeout.Rmd and then the code below is run you get weird 
results, because plot hooks have been left in place:

```
setTimeLimit(elapsed = 1)       # set a timeout
knitr::knit("timeout.Rmd")      # knit the document which triggers it
print(1)                        # this works, but the hooks are still there
plot(1)                         # this triggers them
print(1)                        # Now this doesn't appear
```

What appears to be happening is that the timing error happens in the 
middle of an on.exit() handler execution.

I'm not sure if this is being caused by the complicated way that 
evaluate::evaluate() sets up the on.exit() handlers, or is being caused 
by a problem in R.

BTW, the help for setTimeLimit() says that the time limit check happens 
frequently during `Sys.sleep()`, but that doesn't appear to be true. 
I've tried a variation on the code above which sleeps for 20 seconds, 
even with a time limit of 1 second.

Duncan Murdoch

On 4/27/25 22:19, Duncan Murdoch wrote:
>
> BTW, the help for setTimeLimit() says that the time limit check 
> happens frequently during `Sys.sleep()`, but that doesn't appear to be 
> true. I've tried a variation on the code above which sleeps for 20 
> seconds, even with a time limit of 1 second.
Regarding base R, this works for me in Linux:

 > setTimeLimit(elapsed=10)
 > for(i in 1:1000) Sys.sleep(0.1)
Error in Sys.sleep(0.1) : reached elapsed time limit

That is,? the loop gets interrupted after about 10s.

Reading the code of R, there is a check in Sys.sleep(), but only after 
the first wait (in a loop, iteratively waiting for the required 
duration). When polled events are not active, the first wait is the 
whole duration.

I would read the documentation of ?setTimeLimit as that the check 
happens during Sys.sleep(), but not necessarily frequently, that 
"frequently" only refers to R code. And then the documentation would 
seem correct (or at least not contradicted). On the other hand, right, 
we could change Rsleep() to use a shorter sleep when elapsed time limit 
is in place, and that wouldn't contradict the documentation, either.

Best
Tomas

>
> Duncan Murdoch
>
> ______________________________________________
> R-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel