"Error: attempt to apply non-function" - tryCatchLog package must be attached to work

[chrsigg]

Using tryCatchLog without attaching the package produces an error:

tryCatchLog::tryCatchLog({
 stop("stop")
})
#> Error in h(simpleError(msg, call)): attempt to apply non-function

This is not ideal when using tryCatchLog in a script, and a real problem when using it inside a package. It forces the package author to specify the dependency using Depends: in the DESCRIPTION file, which should be avoided as it modifies the search path (see http://r-pkgs.had.co.nz/namespace.html#namespace for example).

Please consider rewriting the .onAttach function such that your package can be used without having to attach it.

And thank you for your work. Having access to the full stack trace inside tryCatch blocks is much appreciated.

[aryoda]

Thanks for reporting this issue, this must definitely work correctly and looks like a bug!

I am trying to find out what happens internally ;-)

BTW: I am using the tryCatchLog package in my own packages by specifying it in the
Imports section and this should work (at least I have never observed a problem like above).
Edit: I have used the prev. CRAN version (without .onAttach) so this does not prove anything ;-)

[aryoda]

I can reproduce the wrong behaviour and the reason is (as you have already found out) the .onAttach function I am using. Thanks for your help!

I will do some more testing before pushing an fixed version.

Unluckily I have published this new version at CRAN today and I guess others may be affected as well.

At all readers: Please vote here if you are in need for a fixed version at CRAN soon (since I don't want to "flood" the CRAN volunteers with CRAN submissions until it is really necessary).

[AndreMikulec]

One Vote is for: "to have the fixed CRAN version soon"

aryoda,
Thanks for the excellent package.

I have not tried the function" error." But the same happens on simple "message"

Attachment or "not attached" does not matter.

I have not tried "stop".
The error (for me) happens on the simple "message" function.

I wrap 100+ functions with tryCatchLog::tryCatchLog
in my in-development tradeModel package:

https://github.com/AndreMikulec/tradeModel/tree/feature/in_development

I will try to help a little. . . .
Simple.

foo1 <- function(Symbol = NULL) {
  message(stringr::str_c("Begin - ", Symbol))
}
# foo1(Symbol = "IBM")

> foo1(Symbol = "IBM")
Begin - IBM

Simple OK.
Wrapper.

foo2 <- function(Symbol = NULL) {
  tryCatchLog::tryCatchLog({
  message(stringr::str_c("Begin - ", Symbol))
})}
# foo2(Symbol = "IBM")

> foo2(Symbol = "IBM")
Error in .tryCatchLog.env$info.log.func(log.msg) : 
  attempt to apply non-function

So the wrapper is not O.K.
Goes into here ...

withRestarts({
    signalCondition(cond)
    defaultHandler(cond)
  }, muffleMessage = function() NULL)

With condition

cond
message(stringr::str_c("Begin - ", Symbol))

Try without any wrapper and no message function.

foo3 <- function(Symbol = NULL) {
  message("Begin - ")
}
# foo3(Symbol = "IBM")

Begin -

so that works.
Test if tidyverse is the problem.

foo4 <- function(Symbol = NULL) {
  tryCatchLog::tryCatchLog({
  message(base::paste0("Begin - ", Symbol))
})}
# foo4(Symbol = "IBM")

> foo4(Symbol = "IBM")
Error in .tryCatchLog.env$info.log.func(log.msg) : 
  attempt to apply non-function

So, the tidyverse is not the problem.
Try if the :: function is the problem.

foo5 <- function(Symbol = NULL) {
  tryCatchLog::tryCatchLog({
  message(paste0("Begin - ", Symbol))
})}
# foo5(Symbol = "IBM")

The :: function is not the problem.

In the meanwhile, I will "try" to revert to the older version of package tryCatchLog.
Thanks,
Andre

[AndreMikulec]

I tested many github versions all the way back to Jun 2, 2018.
No old github install worked.
6626ffa

The previous cran release does work.:
https://cran.r-project.org/bin/windows/contrib/3.4/tryCatchLog_1.0.2.zip

[aryoda]

@AndreMikulec Thanks for your support :-)

I have just pushed a new release candidate that should fix the problems.

As @chrsigg has indicated using .onAttach was a stupid idea of mine :-(
Now I am initializing everything in the .onLoad hook and using .onAttach only to spit out a message depending on the initialized internal state.

Can you please test the new github version by installing it via

# install.packages("devtools")
library(devtools)
install_github("aryoda/tryCatchLog")

and give me a feed back here?

THX a lot!

Edit 1: I guess I have introduce the problem at github with 03c6293 (2018-06-01)...

[AndreMikulec]

@aryoda ,

Thanks, it seems to work again:

> devtools::install_github("aryoda/tryCatchLog")

> futile.logger::flog.threshold("ERROR")
NULL
> (function() { tryCatchLog::tryCatchLog({ message(paste0("Hello")) }) })()
Hello

> devtools::session_info()
 tryCatchLog      1.1.3   2019-03-22 [1] Github (aryoda/tryCatchLog@3effccb)
 
> (function() { tryCatchLog::tryCatchLog({ stop(paste0("Hello")) }) })()
ERROR [2019-03-22 14:09:26] [ERROR] Hello

Compact call stack:
. . . etc . . .

[chrsigg]

I tested 3effccb with the built in logger (that is, without futile) from a script and from inside a package. It works in both cases wiiiithout having to attach the package. Thank you for the quick response.

[aryoda]

The fixed version is now published at CRAN: https://cran.r-project.org/package=tryCatchLog