Better errors for `@includeRmd`

[lionel-]

I'm debugging errors that occur during #' @includeRmd. The NAMESPACE file is emptied at each error, and I need to reset it manually. Would it make sense to restore it to its original contents in case of error?

[gaborcsardi]

Interesting. I think it would be even better not to touch it when the Rd roclet is running. Or not to touch it at all, until the new contents of NAMESPACE is compiled.

Do you have a reprex procedure for this?

[lionel-]

#' Stop
#' @includeRmd "R/stop.Rmd"
#' @name stop
NULL

With stop.Rmd:

```{r}
stop("foo")
```

Another case where that comes up is when I interrupt document() after it has emptied the NAMESPACE.

[lionel-]

By the way it might be less confusing to output "Emptying NAMESPACE" the first time, rather than "Writing NAMESPACE" twice.

[hadley]

I think the @includeRmd code probably needs a tryCatch?

[lionel-]

We'd want document() to fail if the Rmd fails unexpectedly though?

[hadley]

I think it would emit a warning, like how other tags "fail"

[gaborcsardi]

IDK, this is a bit different, though, no? I think I usually want an error when the dynamic code fails.

[hadley]

Maaaaybe, but I think it's better to be consistent.

[hadley]

It's more clear now that this should fail with a warning since that's what child documents do.