[Rd] [R logs] Help in develop a simply logs package

[Duncan Murdoch]

On 07/10/2014 11:11 AM, DataK - B. THIEURMEL wrote:
> Do you think there is a way to apply a "global" withCallingHandlers() on
> all environement / packages loaded, and then all R code will be executed
> with the defined handler ?

That question makes no sense.  withCallingHandlers is not applied to 
environments or packages, it is applied to expressions.

> With this way, we can define what we want for
> warning, message, error, ... one time, and apply it without needed of
> parse() or added withCallingHandlers in initial code...
>
> Something like this :
>
> withCallingHandlersGlobal <- function(...) {
>     handlers <- list(...)
>     classes <- names(handlers)
>     parentenv <- all.environment
>     if (length(classes) != length(handlers))
>       stop("bad handler specification")
>     .Internal(.addCondHands(classes, handlers, parentenv, NULL, TRUE))
> }
>
> withCallingHandlersGlobal(warning = function(w) {print("Global
> handler")})
>
> warning("A")
> > "Global handler"
> rnorm("A")
> > "Global handler"
>

You would need to write the read-eval-print loop.  Read text from the 
console until you have a complete expression, then evaluate it in 
withCallingHandlers.  Repeat.

Duncan Murdoch
>
> Le 2014-10-07 16:51, Duncan Murdoch a écrit :
> > On 07/10/2014 10:41 AM, DataK - B. THIEURMEL wrote:
> >> OK, thank you for your answers. We are thus going to continue by
> >> analyzing these features
> >
> > The general outline would be this:
> >
> > 1.  Call parse() on the whole file.  This will catch any syntax
> > errors.  If it parses okay, you'll get a vector of expressions to
> > evaluate.
> >
> > 2.  Evaluate each expression in sequence within withCallingHandlers().
> >  You need to decide what to do if you get an error(); source() would
> > quit the script at that point, so that's probably a good idea.
> >
> > Duncan Murdoch
> >>
> >> Le 2014-10-07 16:29, Duncan Murdoch a écrit :
> >> > On 07/10/2014 10:16 AM, DataK - B. THIEURMEL wrote:
> >> >> Thank. withCallingHandlers() and "pander::evals" seem to be very
> >> >> interesting, but little adapted to the analysis of one or several
> >> >> scripts R / of many lines of code. Our goal is one packages requiring
> >> >> no
> >> >> modifications of code R to be able to get back all the desired
> >> >> information.
> >> >>
> >> >> Is-there a hope in seeing R core team adding two options warn and
> >> >> message with the same features as options(error) ? Or if we try (and
> >> >> succeed) to code a patch for it, to see it integrating in R ?
> >> >
> >> > No, I don't think so.  withCallingHandlers is all you need for your
> >> > purpose.
> >> >
> >> > Duncan Murdoch
> >> >
> >> >>
> >> >> I think that it would be very useful.
> >> >>
> >> >> Benoit
> >> >>
> >> >> Le 2014-10-07 14:38, Gergely Daróczi a écrit :
> >> >> > On Tue, Oct 7, 2014 at 2:21 PM, Duncan Murdoch
> >> >> > <murdoch.duncan at gmail.com> wrote:
> >> >> >
> >> >> >> On 07/10/2014, 7:04 AM, DataK - B. THIEURMEL wrote:
> >> >> >>> Hi,
> >> >> >>>
> >> >> >>> With the use of R in production, it is necessary to have a system
> >> >> >> of
> >> >> >>> logs effective, and light.
> >> >> >>>
> >> >> >>> Package exist as to futile.logger, but it require the additional
> >> >> >> coding
> >> >> >>> of logs. So it is thus impossible / very difficult to use it with
> >> >> >> all
> >> >> >>> package them used in the calculation
> >> >> >>>
> >> >> >>> Our idea is to develop one packages global, simple, who would
> >> >> >> allow to
> >> >> >>> identify all the errors, warning, message generated by the
> >> >> >> functions
> >> >> >>> stop(), warning() and message() stop as well as by signals and
> >> >> >>> internally code, with log levels configurable later by package,
> >> >> >>> functions...
> >> >> >>>
> >> >> >>> One way is to overwrite temporarily the functions stop(),
> >> >> >> warning() and
> >> >> >>> message() of base package, but I think is not a good thing, and
> >> >> >>> furthermore, we lose all signals and internally "message"...
> >> >> >>>
> >> >> >>> A good use of options(error) seems to do the perfect job, but
> >> >> >> only for
> >> >> >>> error...
> >> >> >>>
> >> >> >>> Our problem / question :
> >> >> >>> - At present, how it is possible to have the same features for
> >> >> >> messages
> >> >> >>> and warnings? (like options(errors)) (I don't find...)
> >> >> >>> - Would new options be possible in a near future R ?
> >> >> >>> - Have there better / other possibilities to handle all the
> >> >> >> warnings,
> >> >> >>> message of the way which we wish?
> >> >> >>>
> >> >> >>
> >> >> >> ​​withCallingHandlers() lets you evaluate expressions with code
> >> >> >> to catch
> >> >> >> messages, warnings and errors.
> >> >> >
> >> >> > That's exactly what I'm using in "pander::evals" to capture all
> >> >> > error/warning/normal messages while evaluating an R command, and to
> >> >> > also capture the results (as R objects), stdout and the printed
> >> >> > version of the object -- which might be useful in a custom
> >> >> > environment. E.g. I use this function to evaluate all R chunks in
> >> >> > markdown document and also to store all R messages run at the
> >> >> > rapporter.net [2] API. Please let me know if anyone is interested, and
> >> >> > I will start cleaning up the related codebase and publish on GH --
> >> >> > although "pander" and "evals" is already
> >> >> > there: https://github.com/Rapporter/pander [3]
> >> >> >
> >> >> > Quick demo: http://pastebin.com/jCUkgKim [4]
> >> >> >
> >> >> >
> >> >> >> I don't know if there's a way to evaluate every expression entered
> >> >> >> at
> >> >> >> the console within withCallingHandlers() for an effect like
> >> >> >> options(error=), but you can certainly write code to read a file
> >> >> >> and
> >> >> >> evaluate every expression in it within a withCallingHandlers()
> >> >> >> call.
> >> >> >>
> >> >> >> Duncan Murdoch
> >> >> >>
> >> >> >>> Hope is clear. Open to any suggestions.
> >> >> >>>
> >> >> >>> Thank you in advance
> >> >> >>>
> >> >> >>
> >> >> >> ______________________________________________
> >> >> >> R-devel at r-project.org mailing list
> >> >> >> https://stat.ethz.ch/mailman/listinfo/r-devel [1]
> >> >> >
> >> >> >
> >> >> >
> >> >> > Links:
> >> >> > ------
> >> >> > [1] https://stat.ethz.ch/mailman/listinfo/r-devel
> >> >> > [2] http://rapporter.net
> >> >> > [3] https://github.com/Rapporter/pander
> >> >> > [4] http://pastebin.com/jCUkgKim
> >> >>
> >>
>