R devel - Nov 2005 - (PR#8337) formatC adds leading space -- on some Windoze

>>>>> "KevinW" == Kevin Wright <kwright68 at
gmail.com>
>>>>>     on Mon, 21 Nov 2005 18:13:36 +0100 (CET) writes:
    KevinW> Full_Name: Kevin Wright
    KevinW> Version: 2.2.0
    KevinW> OS: Windows 2000
	        ^^^^^^^
this must be part of the problem


    KevinW> Submission from: (NULL) (170.54.58.4)



    KevinW> Apologies if my expectations (or reading of the man page) are
incorrect.

    KevinW> I seem unable to left-justify exponential format
    KevinW> numbers.  There appears to always be an extra space
    KevinW> inserted to the left.

    KevinW> Using the example from the formatC help page:

    R> xx  <- pi * 10^(-5:4)

    R> cbind(formatC(xx, wid = 9, flag = "-"))
    KevinW>      [,1]        
    KevinW> [1,] " 3.142e-05"
    KevinW> [2,] "0.0003142" 
    KevinW> [3,] "0.003142 " 
    KevinW> [4,] "0.03142  " 
    KevinW> [5,] "0.3142   " 
    KevinW> [6,] "3.142    " 
    KevinW> [7,] "31.42    " 
    KevinW> [8,] "314.2    " 
    KevinW> [9,] "3142     " 
    KevinW> [10,] " 3.142e+04"

which is also not obeying the 'wid' argument.

I get something much more reasonable:

      [,1]       
 [1,] "3.142e-05"
 [2,] "0.0003142"
 [3,] "0.003142 "
 [4,] "0.03142  "
 [5,] "0.3142   "
 [6,] "3.142    "
 [7,] "31.42    "
 [8,] "314.2    "
 [9,] "3142     "
[10,] "3.142e+04"

formatC uses your system's C library printf {that's where the
"C" comes from in 'formatC'} which seems to be
broken or at least not performing as we think it should.

On a "Windows 2003 Server" I have access to, I see the same
wrong behavior as above.

Martin Maechler, ETH Zurich

On Tue, 22 Nov 2005 maechler at stat.math.ethz.ch wrote:
>>>>>> "KevinW" == Kevin Wright <kwright68 at
gmail.com>
>>>>>>     on Mon, 21 Nov 2005 18:13:36 +0100 (CET) writes:
>
>    KevinW> Full_Name: Kevin Wright
>    KevinW> Version: 2.2.0
>    KevinW> OS: Windows 2000
> 	        ^^^^^^^
> this must be part of the problem
It is, and it is a known inconsistency with Linux (but I do not consider 
it to be a bug or `wrong behavior' or not `reasonable').

Windows always uses three digits for the exponent, e.g. E+001. This 
results from adjusting the returned result to be more consistent with 
other platforms.  (BTW, since width (sic) is a lower bound, it _is_ 
respected.)  Even if the layout is not ideal, the results are at least 
diff-able against those from other platforms.

If Kevin (or anyone else) wants it done even more consistently, he could 
contribute a patch.  Now, we _have_ done that for print(), but it did not 
seem worth it for formatC (especially as sprintf() is now widely used and 
would also need to be made consistent). (It also did not seem worth it 
given how little credit is given for such work.)

>    KevinW> Submission from: (NULL) (170.54.58.4)
>
>
>
>    KevinW> Apologies if my expectations (or reading of the man page) are
incorrect.
>
>    KevinW> I seem unable to left-justify exponential format
>    KevinW> numbers.  There appears to always be an extra space
>    KevinW> inserted to the left.
>
>    KevinW> Using the example from the formatC help page:
>
>    R> xx  <- pi * 10^(-5:4)
>
>    R> cbind(formatC(xx, wid = 9, flag = "-"))
>    KevinW>      [,1]
>    KevinW> [1,] " 3.142e-05"
>    KevinW> [2,] "0.0003142"
>    KevinW> [3,] "0.003142 "
>    KevinW> [4,] "0.03142  "
>    KevinW> [5,] "0.3142   "
>    KevinW> [6,] "3.142    "
>    KevinW> [7,] "31.42    "
>    KevinW> [8,] "314.2    "
>    KevinW> [9,] "3142     "
>    KevinW> [10,] " 3.142e+04"
>
> which is also not obeying the 'wid' argument.
>
> I get something much more reasonable:
>
>      [,1]
> [1,] "3.142e-05"
> [2,] "0.0003142"
> [3,] "0.003142 "
> [4,] "0.03142  "
> [5,] "0.3142   "
> [6,] "3.142    "
> [7,] "31.42    "
> [8,] "314.2    "
> [9,] "3142     "
> [10,] "3.142e+04"
>
> formatC uses your system's C library printf {that's where the
> "C" comes from in 'formatC'} which seems to be
> broken or at least not performing as we think it should.
>
> On a "Windows 2003 Server" I have access to, I see the same
> wrong behavior as above.
>
> Martin Maechler, ETH Zurich
>
> ______________________________________________
> R-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel
>
>
-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595

>>>>> "BDR" == Brian Ripley <ripley at
stats.ox.ac.uk>
>>>>>     on Tue, 22 Nov 2005 09:35:19 +0100 (CET) writes:
    BDR> On Tue, 22 Nov 2005 maechler at stat.math.ethz.ch wrote:
    >>>>>>> "KevinW" == Kevin Wright
<kwright68 at gmail.com>
    >>>>>>> on Mon, 21 Nov 2005 18:13:36 +0100 (CET)
writes:
    >> 
    KevinW> Full_Name: Kevin Wright
    KevinW> Version: 2.2.0
    KevinW> OS: Windows 2000
    >>          ^^^^^^^
    >> this must be part of the problem

    BDR> It is, and it is a known inconsistency with Linux (but I do not
consider
    BDR> it to be a bug or `wrong behavior' or not `reasonable').

    BDR> Windows always uses three digits for the exponent, e.g. E+001. 

    BDR> This results from adjusting the returned result to be more
consistent with
    BDR> other platforms.  (BTW, since width (sic) is a lower bound, it _is_ 
    BDR> respected.)  Even if the layout is not ideal, the results are at
least
    BDR> diff-able against those from other platforms.

yes, and that's good ("diff-able").

Now, after looking in the source (src/appl/strsignif.c),
I understand what you mean above: Because Windows libc's sprintf()
produces 3-digits exponents *and* because you amended the code
to change these back to 2 digits, the extraneous blank is a
consequence that one had to live with.. 

OTOH, the oldest R on Windows I have, "2.0.1 patched
(2004-11-27)" does *not* prepend the extra " " and gives the
same as on non-windows in some cases, i.e.,
> formatC(pi * 10^(-5:4))
 [1] "3.142e-05" "0.0003142" "0.003142" 
"0.03142"   "0.3142"    "3.142"
 [7] "31.42"     "314.2"     "3142"     
"3.142e+04"

Insofar, the  svn r35148 (2005-08-04 18:17:00) change
did produce a backward incompatibility.  
Which is why I confirmed Kevin that this was wrong behavior.

    BDR> If Kevin (or anyone else) wants it done even more
    BDR> consistently, he could contribute a patch.  

yes, that would be useful.
Particularly, since for some cases formatC() *was* more
consistent (see above).

    BDR> Now, we _have_ done that for print(), but it did not seem worth
    BDR> it for formatC (especially as sprintf() is now widely
    BDR> used and would also need to be made consistent).

I agree that sprintf should also be consistent, and I also agree
that for "real" programmers sprintf() is probably more useful
than formatC() which OTOH is easier to use for simple useRs.

    BDR> (It also did not seem worth it 
    BDR>  given how little credit is given for such work.)

    KevinW> Submission from: (NULL) (170.54.58.4)
    >> 

    KevinW> Apologies if my expectations (or reading of the man page) are
incorrect.
    >> 
    KevinW> I seem unable to left-justify exponential format
    KevinW> numbers.  There appears to always be an extra space
    KevinW> inserted to the left.
    >> 
    KevinW> Using the example from the formatC help page:
    >> 
    R> xx  <- pi * 10^(-5:4)
    >> 
    R> cbind(formatC(xx, wid = 9, flag = "-"))
    KevinW> [,1]
    KevinW> [1,] " 3.142e-05"
    KevinW> [2,] "0.0003142"
    KevinW> [3,] "0.003142 "
    KevinW> [4,] "0.03142  "
    KevinW> [5,] "0.3142   "
    KevinW> [6,] "3.142    "
    KevinW> [7,] "31.42    "
    KevinW> [8,] "314.2    "
    KevinW> [9,] "3142     "
    KevinW> [10,] " 3.142e+04"
    >> 
    >> which is also not obeying the 'wid' argument.
    >> 
    >> I get something much more reasonable:
    >> 
    >> [,1]
    >> [1,] "3.142e-05"
    >> [2,] "0.0003142"
    >> [3,] "0.003142 "
    >> [4,] "0.03142  "
    >> [5,] "0.3142   "
    >> [6,] "3.142    "
    >> [7,] "31.42    "
    >> [8,] "314.2    "
    >> [9,] "3142     "
    >> [10,] "3.142e+04"
    >> 
    >> formatC uses your system's C library printf {that's where
the
    >> "C" comes from in 'formatC'} which seems to be
    >> broken or at least not performing as we think it should.
    >> 
    >> On a "Windows 2003 Server" I have access to, I see the
same
    >> wrong behavior as above.
    >> 
    >> Martin Maechler, ETH Zurich
    >> 
    >> ______________________________________________
    >> R-devel at r-project.org mailing list
    >> https://stat.ethz.ch/mailman/listinfo/r-devel
    >> 
    >> 

    BDR> -- 
    BDR> Brian D. Ripley,                  ripley at stats.ox.ac.uk
    BDR> Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
    BDR> University of Oxford,             Tel:  +44 1865 272861 (self)
    BDR> 1 South Parks Road,                     +44 1865 272866 (PA)
    BDR> Oxford OX1 3TG, UK                Fax:  +44 1865 272595

    BDR> ______________________________________________
    BDR> R-devel at r-project.org mailing list
    BDR> https://stat.ethz.ch/mailman/listinfo/r-devel