Strange difference between x and get("x")?

Question

Somehow, sometimes, I'm ending up in a state like this :

> x
[1] 1 2 3
> get("x")
Error in get("x") : object 'x' not found
> x
[1] 1 2 3

I can't reproduce it reliably. What sort of things might I have done wrong in my C code? Why would typing x at the prompt find it, but get("x") not? What's the difference internally between x and get("x")?

Any hints much appreciated. I've started seeing this since R 2.14.0 but my C code has also been changing too.

EDIT : reproducible example

// test.c
#include <R.h>
#include <Rdefines.h> 

SEXP test(SEXP df)
{
    SEXP levels, s;
    int j;

    levels = getAttrib(VECTOR_ELT(df,0), R_LevelsSymbol);
    Rprintf("levels %u, type %d, length %d, truelength %d\n",
             levels,TYPEOF(levels),LENGTH(levels),TRUELENGTH(levels));

    for (j=0; j<length(levels); j++) {
        s = STRING_ELT(levels,j);
        Rprintf("%d %d %s %u %d %d\n", length(levels), TYPEOF(s),
                        CHAR(s), s, LENGTH(s), TRUELENGTH(s));
        SET_TRUELENGTH(s,1);  // clobbers the 65, but why 65 ("A") there?
        Rprintf("%d %d %s %u %d %d\n", length(levels), TYPEOF(s),
                        CHAR(s), s, LENGTH(s), TRUELENGTH(s));
    }
    return(R_NilValue);
}

and to run it :

R --vanilla

system("R CMD SHLIB -otest.so test.c")
dyn.load("test.so")

if (FALSE) A     # needed for error to occur (!)

DF <- data.frame(a = c("A", "Z"), b = 1:4)
print(DF)
.Call("test",DF)
print(DF)

A = data.frame()
for (i in 1:100) {
    cat(i,"")
    assign(paste("v",i,sep=""),i)
    get("A")
}

The output I get :

$ R --vanilla    
R version 2.14.0 (2011-10-31)
# [snip header]
> system("R CMD SHLIB -otest.so test.c")
gcc -std=gnu99 -I/usr/share/R/include      -fpic  -std=c99 -O6 -Wall -Wno-unused -pedantic -c test.c -o test.o
gcc -std=gnu99 -shared -o test.so test.o -otest.so -L/usr/lib/R/lib -lR
> dyn.load("test.so")
> 
> if (FALSE) A     # needed for error to occur (!)
> 
> DF <- data.frame(a = c("A", "Z"), b = 1:4)
> print(DF)
  a b
1 A 1
2 Z 2
3 A 3
4 Z 4
> .Call("test",DF)
levels 151395176, type 16, length 2, truelength 0
2 9 A 149596512 1 65   # why this 65 here?
2 9 A 149596512 1 1
2 9 Z 149596320 1 0
2 9 Z 149596320 1 1
NULL
> print(DF)
  a b
1 A 1
2 Z 2
3 A 3
4 Z 4
> 
> A = data.frame()
> for (i in 1:100) {
+     cat(i,"")
+     assign(paste("v",i,sep=""),i)
+     get("A")
+ }
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 Error in get("A") : object 'A' not found
> 
> sessionInfo()
R version 2.14.0 (2011-10-31)
Platform: i686-pc-linux-gnu (32-bit)

locale:
 [1] LC_CTYPE=en_GB.UTF-8       LC_NUMERIC=C              
 [3] LC_TIME=en_GB.UTF-8        LC_COLLATE=en_GB.UTF-8    
 [5] LC_MONETARY=en_GB.UTF-8    LC_MESSAGES=en_GB.UTF-8   
 [7] LC_PAPER=C                 LC_NAME=C                 
 [9] LC_ADDRESS=C               LC_TELEPHONE=C            
[11] LC_MEASUREMENT=en_GB.UTF-8 LC_IDENTIFICATION=C       

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     
> 

Any ideas? If the if (FALSE) A line is commented out then it works fine. For repeated tests, R must be started fresh each time.

Answer 1

This indeed turned out to be my C code. I knew TRUELENGTH is sometimes used by R but I didn't think on CHARSXP. When a variable name is the same as some character value, the CHARSXP's TRUELENGTH is used by R to hold an internal hash value, see main/envir.c. My SET_TRUELENGTH on the CHARSXP was clobbering the hash. Thanks to Simon Urbanek for explaining this, and thanks for all the tips and ideas in comments.

To demonstrate :

$ R --vanilla
R version 2.14.0 (2011-10-31)

> system("R CMD SHLIB -otest.so test.c")
> dyn.load("test.so")
> truelength = function(x)invisible(.Call("truelength",x))
> 
> truelength("A")
'A' has length 1 and truelength 0
> truelength("ABC")
'ABC' has length 3 and truelength 0
> A=123
> truelength("A")
'A' has length 1 and truelength 65    # 65 is the HASHPRI, for bound variable A
> truelength("ABC")
'ABC' has length 3 and truelength 0    # no variable ABC so truelength unused
> ABC=456
> truelength("ABC")
'ABC' has length 3 and truelength 17763   # now ABC symbol is bound
> 
> foo = 7
> truelength("foo")               
'foo' has length 3 and truelength 27999   # bound
> truelength("bar")               
'bar' has length 3 and truelength 0       # not bound
> .Internal(inspect("foo"))
@876eb08 16 STRSXP g0c1 [NAM(2)] (len=1, tl=0)   # tl=0 of STRSXP vector
  @81759e8 09 CHARSXP g0c1 [gp=0x21] "foo"       # tl of CHARSXP not shown by inspect

where the C code to see the TRUELENGTH of the CHARSXP is :

// test.c
#include <R.h>
#include <Rdefines.h> 

SEXP truelength(SEXP v)
{
    SEXP s = STRING_ELT(v,0);
    Rprintf("'%s' has length %d and truelength %d\n",
                  CHAR(s), LENGTH(s), TRUELENGTH(s));
    return(R_NilValue);
}