[ESS-bugs] OOps! ESS partly fails in non-English locales!

[Martin Maechler]

>>>>> "StEgl" == Stephen Eglen <S.J.Eglen at damtp.cam.ac.uk>
>>>>>     on Wed, 3 Aug 2005 16:39:09 +0100 writes:

    StEgl> To follow up on parsing the output in *ESS-errors*

    StEgl> I can fix the regexp for this particular error:
    >> 
    StEgl> Error in parse(file, n = -1, NULL, "?") : syntax error on line 3
    >> 
    StEgl> perhaps by a regexp searching for
    >> 
    StEgl> syntax error on line [line number]
    >> 
    StEgl> but is that likely to be good enough for catching all R errors?
    >> 

    StEgl> Taking a look at the function ess-parse-errors, it would need not just
    StEgl> the line number, but of course, the name of the file!  (One hack could
    StEgl> be to write the name of the source file as a comment at the top of *ESS-errors*
    StEgl> within ess-load-file, e.g. 
    StEgl> ### Source file: foo.R
    StEgl> so that could then be grabbed by ess-parse-errors.)

    StEgl> However, a longer term thought might be to see if e.g R could output
    StEgl> its errors in a way more typical of  unix commands, e.g. grep:

    StEgl> M-x grep RET defun *.el RET

    StEgl> The output comes in a format that Emacs finds easy to parse:

    StEgl> ess-comp.el:43:(defun ess-message (msg)
    StEgl> ess-comp.el:60:;;; Components:  defuns, defvars, defconsts

    StEgl> ditto for outptu from gcc which can be parsed easily by Emacs.
    StEgl> Is that likely to be seen favourably by R core?

I'm not sure I understand.
Do you mean R's error message should be as terse as Unix' ones?
e.g. mentioning the word 'error' at all??

R currently has basically only one of two

1) Error: ............

2) Error in foobar(.........) : ...........

where the 'foobar(.......)' can be quite quite a general
expression and doesn't need to look closely like a simple
function call.  
The whole thread started because  'Error' and 'Error in' will be
translated into different locales.

I think one main difference between R and typical unix programs
is that many functions in R print information messages, text, ...
whereas in Unix successful command completion almost always
means "no output to the console at all".

In a very general, yes, I agree that R-core should think about
how error messages are to be formatted.  Just because one cannot
use  
     r <- try(......) 
     if(inherits(r, "try-error"))

everywhere --- as it is usually suggested; I think we (the R
community) have recently seen examples where this lead to
wrongly running code when the "......" call also used try()
somewhere.   Even more an R-core issue??

Martin