TGM 2019-07-15

notes.Rmd

---
title: "Some new features in processx and callr"
output: html_notebook
---

# Better errors in processx, callr, gh

## Better errors from C

```{r error = TRUE}
processx::run("foobar")
processx::run("/tmp")
```

Two portable files:
<https://github.com/r-lib/processx/blob/master/src/errors.c>
<https://github.com/r-lib/processx/blob/master/src/errors.h>

API: 
```
R_THROW_ERROR("Failed to open file `%s`, path);
R_THROW_SYSTEM_ERROR("Failed to set permission on file '%s', path);
R_THROW_SYSTEM_ERROR_CODE(errorcode, ...);
```

## `errors.R` standalone file from processx

Supports `.Last.error` and `.Last.error.trace`:

```{r error = TRUE}
processx::run("foobar")
.Last.error
.Last.error.trace
```

### in gh

```{r error = TRUE}
gh::gh("/foobar")
.Last.error
.Last.error.trace
```

### in callr

```{r error = TRUE}
callr::r(function() library("xxx"))
.Last.error
.Last.error.trace
```

# `callr::r_session` debugger

```{r}
rs <- callr::r_session$new()
rs$run(function() Sys.getpid())
Sys.getpid()
```

```{r error = TRUE}
rs$run(function() knitr::knit("foo-bar"))
.Last.error
.Last.error.trace
```

Now call `rs$debug()` and see the help.

# Pseudo-terminal with processx

Useful for
- driving interactive programs / R code
- testing interactive R code

```{r}
r2 <- processx::process$new("R", pty = TRUE)
r2$poll_io(1000)
```

```{r}
cat(r2$read_output())
cat(r2$read_output())
```

```{r}
r2$write_input("yesno::yesno()\n")
r2$poll_io(-1)
```

```{r}
cat(r2$read_output())
cat(r2$read_output())
```

```{r}
r2$write_input("1\n")
r2$poll_io(1000)
```

```{r}
cat(r2$read_output())
cat(r2$read_output())
cat(r2$read_output())
```

# Memory mapped data transfer in callr

This needs the `feature/mmap` branch of callr.

```{r}
data <- list(a = 1:100, b = runif(1e7), c = charToRaw("foobar"))
```

```{r}
fun <- function(a, b, c) {
  list(mean(a), median(b), rawToChar(c))
}
```

```{r}
system.time({
  cl <- parallel::makePSOCKcluster(1)
  parallel::clusterExport(cl, c("data", "fun"))
  res <- parallel::clusterEvalQ(cl, do.call(fun, data))
})
```

```{r}
system.time(
  res <- callr::r(fun, args = data)
)
```

```{r}
system.time(
  res <- callr::r(fun, args = data, transfer = "mmap")
)
```