R devel - Mar 2009 - incoherent conversions from/to raw

i wonder about the following examples showing incoherence in how type
conversions are done in r:

    x = TRUE

    x[2] = as.raw(1)
    # Error in x[2] = as.raw(1) :
    #  incompatible types (from raw to logical) in subassignment type fix

it seems that there is an attempt to coerce the raw value to logical
here, which fails, even though

    as.logical(as.raw(1))
    # TRUE

likewise,

    x[2] = 1L
    # the vector is silently coerced upwards to integer
    x[2] = as.raw(1)
    # Error in x[2] = as.raw(1) :
    #  incompatible types (from raw to integer) in subassignment type fix

even though

    as.integer(as.raw(1))
    # 1

and likewise for double and complex.

there's another incoherence:

    x = 1
    x[2] = 1i
    x
    # 1i 1i

    x = 1i
    x[2] = 1
    x
    # 1i 1i

in both cases, the higher type is used for the result;  in the former
case, the vector is coerced upwards, in the latter, the assigned value
is coerced upwards.  however:

    x = 1
    x[2] = as.raw(1)
    # error: incompatible types (from raw to double)

    x = as.raw(1)
    x[2] = 1
    # error: incompatible types (from double to raw)

leaving aside that

    as.double(as.raw(1))
    # 1
    as.raw(as.double(1))
    # 1

work just fine, in both cases there is an attempt to coerce the assigned
value to the vector type, and not to the higher type (which would
presumably qbe double, as in ?c), as in the previous example.  

interestingly,

    c(1, as.raw(1))
    # error: type 'raw' is unimplemented in 'RealAnswer'

(note the 'real', not 'double'), whereas

    1 == as.raw(1)
    # TRUE

works just fine.  furthermore,

    c('1', as.raw(1))
    # "1" "01"

whereas

    x = '1'
    x[2] = as.raw(1)
    # error: incompatible types (from raw to character)

yet another issue is that of indexing a raw vector with an out-of-bounds
index.  the r language definition, sec. 3.4.1 [1] says:

"
We shall discuss indexing of simple vectors first. For simplicity,
assume that the expression is x[i]. (...)
If i is positive and exceeds length(x) then the corresponding selection
is NA.
"

it's probably correct to assume that 'simple vector' means
'atomic
vector', though not all r core members seem to be quite sure [2]:

"
So what is a simple vector? That is not explicitly defined, and it
probably should be. I think it is "atomic vectors, except those with a
class that has a method for [".
"

it appears that raw vectors are atomic vectors:

    is.atomic(as.raw(1))
    # TRUE

so an index out of bounds (at least, a positive integer index exceeding
the length of the vector) should (?) produce an NA;  however,

    as.raw(1)[2]
    # 00

this is presumably because there is no raw NA, and an NA of whatever
type is converted to 0 by as.raw:

    as.raw(NA)
    # 00
    # warning: out-of-range values treated as 0 in coercion to raw

but in this case there's a warning.  why does not out-of-bounds indexing
of a raw vector not produce a warning?  following the language
definition, the fact that a raw vector is atomic, and the above informal
statement on simple and atomic vectors, the selection should first
produce an NA, which only subsequently is coerced to the raw 0 -- with a
warning.  there's an analogous issue with out-of-bounds assignment:

    x = as.raw(1)
    x[3] = as.raw(3)
    x
    # 01 00 03

but

    x = 1
    x[3] = 3
    x
    # 1 NA 3
    as.raw(x)
    # warning: out-of-range values treated as 0 in coercion to raw


is all this an intended feature?

vQ

[1] http://cran.r-project.org/doc/manuals/R-lang.html#Indexing-by-vectors
[2] http://tolstoy.newcastle.edu.au/R/e6/devel/09/03/0954.html

Wacek Kusnierczyk wrote:
> interestingly,
>
>     c(1, as.raw(1))
>     # error: type 'raw' is unimplemented in 'RealAnswer'
>
>   
three more comments.


(1)
the above is interesting in the light of what ?c says:

"
The output type is determined from the highest type of the
     components in the hierarchy NULL < raw < logical < integer <
real
     < complex < character < list < expression.
"

which seems to suggest that raw components should be coerced to whatever
the highest type among all arguments to c, which clearly doesn't happen:

    test = function(type)
        c(as.raw(1), get(sprintf('as.%s',type))(1))

    for (type in c('null', 'logical', 'integer',
'real', 'complex',
'character', 'list', 'expression'))
       tryCatch(test(type), error = function(e) cat(sprintf("raw won't
coerce to %s type\n", type)))

which shows that raw won't coerce to the four first types in the
'hierarchy' (excluding NULL), but it will to character, list, and
expression.

suggestion:   improve the documentation, or adapt the implementation to
a more coherent design.



(2)
incidentally, there's a bug somewhere there related to the condition
system and printing:

    tryCatch(stop(), error=function(e) print(e))
    # works just fine

    tryCatch(stop(), error=function(e) sprintf('%s', e))
    # *** caught segfault ***
    # address (nil), cause 'memory not mapped'

    # Traceback:
    # 1: sprintf("%s", e)
    # 2: value[[3]](cond)
    # 3: tryCatchOne(expr, names, parentenv, handlers[[1]])
    # 4: tryCatchList(expr, classes, parentenv, handlers)
    # 5: tryCatch(stop(), error = function(e) sprintf("%s", e))

    # Possible actions:
    # 1: abort (with core dump, if enabled)
    # 2: normal R exit
    # 3: exit R without saving workspace
    # 4: exit R saving workspace
    # Selection:
 
interestingly, it is possible to stay in the session by typing ^C.  the
session seems to work, but if the tryCatch above is tried once again, a
segfault causes r to crash immediately:

    # ^C
    tryCatch(stop(), error=function(e) sprintf('%s', e))
    # [whoever at wherever] $

however, this doesn't happen if some other code is evaluated first:

    # ^C
    x = 1:10^8
    tryCatch(stop(), error=function(e) sprintf('%s', e))
    # Error in sprintf("%s", e) : 'getEncChar' must be called
on a CHARSXP
  
this can't be a feature.  (tried in both 2.8.0 and r-devel;  version
info at the bottom.)

suggestion:  trace down and fix the bug.



(3)
the error argument to tryCatch is used in two examples in ?tryCatch, but
it is not explained anywhere in the help page.  one can guess that the
argument name corresponds to the class of conditions the handler will
handle, but it would be helpful to have this stated explicitly.  the
help page simply says:

"
   If a condition is signaled while evaluating 'expr' then
     established handlers are checked, starting with the most recently
     established ones, for one matching the class of the condition.
     When several handlers are supplied in a single 'tryCatch' then the
     first one is considered more recent than the second. 
"

which is uninformative in this respect -- what does 'one matching the
class' mean?

suggestion:  improve the documentation.

vQ

> version
               _                          
platform       i686-pc-linux-gnu          
arch           i686                       
os             linux-gnu                  
system         i686, linux-gnu            
status                                    
major          2                          
minor          8.0                        
year           2008                       
month          10                         
day            20                         
svn rev        46754                      
language       R                          
version.string R version 2.8.0 (2008-10-20)


> version
              
_                                                              
platform      
i686-pc-linux-gnu                                              
arch          
i686                                                           
os            
linux-gnu                                                      
system         i686,
linux-gnu                                                
status         Under development
(unstable)                                   
major         
2                                                              
minor         
9.0                                                            
year          
2009                                                           
month         
03                                                             
day           
19                                                             
svn rev       
48152                                                          
language      
R                                                              
version.string R version 2.9.0 Under development (unstable) (2009-03-19
r48152)

>>>>> "WK" == Wacek Kusnierczyk
<Waclaw.Marcin.Kusnierczyk at idi.ntnu.no>
>>>>>     on Thu, 19 Mar 2009 10:17:20 +0100 writes:
    WK> Wacek Kusnierczyk wrote:
    >> interestingly,
    >> 
    >> c(1, as.raw(1))
    >> # error: type 'raw' is unimplemented in
'RealAnswer'
    >> 
    >> 

    WK> three more comments.


    WK> (1)
    WK> the above is interesting in the light of what ?c says:

    WK> "
    WK> The output type is determined from the highest type of the
    WK> components in the hierarchy NULL < raw < logical < integer
< real
    WK> < complex < character < list < expression.
    WK> "

    WK> which seems to suggest that raw components should be coerced to
whatever
    WK> the highest type among all arguments to c, which clearly doesn't
happen:

    WK> test = function(type)
    WK> c(as.raw(1), get(sprintf('as.%s',type))(1))

    WK> for (type in c('null', 'logical', 'integer',
'real', 'complex',
    WK> 'character', 'list', 'expression'))
    WK> tryCatch(test(type), error = function(e) cat(sprintf("raw
won't
    WK> coerce to %s type\n", type)))

    WK> which shows that raw won't coerce to the four first types in the
    WK> 'hierarchy' (excluding NULL), but it will to character, list,
and
    WK> expression.

    WK> suggestion:   improve the documentation, or adapt the implementation
to
    WK> a more coherent design.

Thank you, Wacek.

I've decided to adapt the implementation
such that all the above  c(<raw> , <type>)  calls' implicit
coercions will work.


    WK> (2)
    WK> incidentally, there's a bug somewhere there related to the
condition
    WK> system and printing:

    WK> tryCatch(stop(), error=function(e) print(e))
    WK> # works just fine

    WK> tryCatch(stop(), error=function(e) sprintf('%s', e))
    WK> # *** caught segfault ***
    WK> # address (nil), cause 'memory not mapped'

    WK> # Traceback:
    WK> # 1: sprintf("%s", e)
    WK> # 2: value[[3]](cond)
    WK> # 3: tryCatchOne(expr, names, parentenv, handlers[[1]])
    WK> # 4: tryCatchList(expr, classes, parentenv, handlers)
    WK> # 5: tryCatch(stop(), error = function(e) sprintf("%s", e))

    WK> # Possible actions:
    WK> # 1: abort (with core dump, if enabled)
    WK> # 2: normal R exit
    WK> # 3: exit R without saving workspace
    WK> # 4: exit R saving workspace
    WK> # Selection:
 
    WK> interestingly, it is possible to stay in the session by typing ^C. 
the
    WK> session seems to work, but if the tryCatch above is tried once again,
a
    WK> segfault causes r to crash immediately:

    WK> # ^C
    WK> tryCatch(stop(), error=function(e) sprintf('%s', e))
    WK> # [whoever at wherever] $

    WK> however, this doesn't happen if some other code is evaluated
first:

    WK> # ^C
    WK> x = 1:10^8
    WK> tryCatch(stop(), error=function(e) sprintf('%s', e))
    WK> # Error in sprintf("%s", e) : 'getEncChar' must be
called on a CHARSXP
  
    WK> this can't be a feature.  (tried in both 2.8.0 and r-devel; 
version
    WK> info at the bottom.)

    WK> suggestion:  trace down and fix the bug.

[not me, at least not now.]


    WK> (3)
    WK> the error argument to tryCatch is used in two examples in ?tryCatch,
but
    WK> it is not explained anywhere in the help page.  one can guess that
the
    WK> argument name corresponds to the class of conditions the handler will
    WK> handle, but it would be helpful to have this stated explicitly.  the
    WK> help page simply says:

    WK> "
    WK> If a condition is signaled while evaluating 'expr' then
    WK> established handlers are checked, starting with the most recently
    WK> established ones, for one matching the class of the condition.
    WK> When several handlers are supplied in a single 'tryCatch'
then the
    WK> first one is considered more recent than the second. 
    WK> "

    WK> which is uninformative in this respect -- what does 'one matching
the
    WK> class' mean?

    WK> suggestion:  improve the documentation.

Patches to  tryCatch.Rd  are gladly accepted
and quite possibly applied to the sources without much changes.
Thanks in advance!

Martin Maechler, ETH Zurich