I recently posed this question and thankfully was pointed to withRestarts()
which seems pretty awesome and powerful to me :-) Now I'm eager to understand R's error handling capabilities in a bit more detail.
Actual questions
- What is the recommended usage of
simpleCondition()
? Never used it before, but I thought it might be useful for designing custom errors and warnings that are in fact "true" conditions. Could it be used to build a database of specific conditions for which specific handlers are available? - Is there a way to "freeze" a certain state of the entire R workspace and return to it to restart a computation at a certain point? I'm aware of
save.image()
, but AFAIU, this doesn't store the "state" of the search path (search()
orsearchpaths()
).
For those interested
Two code examples
- illustration of my current use of
withRestarts
in dependence on this blog post - attempt to define a "custom condition"
I'd appreciate any comments/suggestion on what to do better ;-)
Example 1
require("forecast")
autoArimaFailsafe <- function(
x,
warning=function(w, ...) {
message("autoArimaFailsafe> warning:")
message(w)
invokeRestart("donothing")},
error=function(e, ...) {
message("autoArimaFailsafe> error:")
message(e)
invokeRestart("abort")}
) {
withRestarts(
out <- tryCatch(
{
expr <- expression(auto.arima(x=x))
return(eval(expr))
},
warning=warning,
error=error
),
donothing=function(...) {
return(eval(expr))
},
abort=function(...) {
message("aborting")
return(NULL)
}
)
}
data(AirPassengers)
autoArimaFailsafe(x=AirPassengers)
autoArimaFailsafe(x="a")
Example 2
require("forecast")
autoArimaFailsafe <- function(
x,
warning=function(w, ...) {
message("autoArimaFailsafe> warning")
invokeRestart("donothing")},
error=function(e, ...) {
message("autoArimaFailsafe> error")
invokeRestart("abort")},
condition=function(cond, ...) {
out <- NULL
message(cond)
condmsg <- conditionMessage(c=cond)
condclass <- class(cond)
if (any(class(cond) == "simpleWarning")) {
out <- warning(w=cond)
} else if (any(class(cond) == "simpleError")) {
out <- error(e=cond)
} else if (any(class(cond) == "simpleCondition")) {
if (condmsg == "invalid class: character") {
out <- invokeRestart("forcedefault")
}
}
return(out)
}
) {
withRestarts(
out <- tryCatch(
{
expr <- expression(auto.arima(x=x))
if (class(x) == "character") {
expr <- signalCondition(
simpleCondition("invalid class: character",
call=as.call(expr))
)
}
return(eval(expr))
},
condition=condition
),
donothing=function(...) {return(eval(expr))},
abort=function(...) {
message("aborting")
return(NULL)
},
forcedefault=function(...) {
data(AirPassengers)
expr <- expression(auto.arima(x=AirPassengers))
return(eval(expr))
}
)
}
autoArimaFailsafe(x=AirPassengers)
autoArimaFailsafe(x=NULL)
autoArimaFailsafe(x="a")
This post references the inspiration for R's condition handling.
For 1., I think of simpleCondition
as illustrating how one can construct custom conditions, e.g,.
myCondition <-
function(message, call=NULL, type=c("overflow", "underflow", "zero"))
{
type <- match.arg(type) # only allowed types past here
class <- c(type, "my", "condition")
structure(list(message = as.character(message), call = call),
class = class)
}
is a constructor for making custom conditions
> myCondition("oops")
<overflow: oops>
> myCondition("oops", type="underflow")
<underflow: oops>
These conditions can be used in tryCatch
or withCallingHandlers
xx <- tryCatch({
signalCondition(myCondition("oops", type="underflow"))
}, underflow=function(e) {
message("underflow: ", conditionMessage(e))
NA # return value, assigned to xx
})
These are S3 classes so can have a linear hierarchy -- bad
and worse
are both subclasses of error
.
myError <-
function(message, call=NULL, type=c("bad", "worse"))
{
type <- match.arg(type)
class <- c(type, "error", "condition")
structure(list(message=as.character(message), call=call),
class=class)
}
One might also create an error that extends the 'simpleError' S3 class as cond <- simpleError("oops"); class(cond) = c("myerr", class(cond)
With tryCatch
we just get access to a single handler, the first (in the sense described on ?tryCatch) to match the class of condition
tryCatch({
stop(myError("oops", type="worse"))
}, bad = function(e) {
message("bad error: ", conditionMessage(e))
}, worse = function(e) {
message("worse error: ", conditionMessage(e)) # here's where we end up
}, error=function(e) {
message("error: ", conditionMessage(e))
})
With withCallingHandlers
we have the opportunity to hit multiple handlers, provided we don't invoke a restart
withCallingHandlers({
stop(myError("oops", type="bad"))
}, bad = function(e) { # here...
message("bad error: ", conditionMessage(e))
}, worse = function(e) {
message("worse error: ", conditionMessage(e))
}, error=function(e) { # ...and here...
message("error: ", conditionMessage(e))
}) # ...and top-level 'error'
withCallingHandlers({
x <- 1
warning(myError("oops", type="bad"))
"OK"
}, bad = function(e) { # here, but continue at the restart
message("bad warning: ", conditionMessage(e))
invokeRestart("muffleWarning")
}, worse = function(e) {
message("worse warning: ", conditionMessage(e))
})
I'm not so sure about your question 2; I think this is the situation that calling handlers are designed to address -- the entire frame where the condition was invoked is poised waiting to continue, once you invoke the restart.
来源:https://stackoverflow.com/questions/15363004/advanced-error-handling