R devel - Oct 1999 - read.table problem on Linux/Alpha (seg faults caused by isspace(R_EOF)) (PR#303)

Full_Name: Naoki Takebayashi
Version: 0.65.1
OS: Linux/Alpha
Submission from: (NULL) (129.79.224.171)


When I was reading a file with read.table("junk.data"), it
seg-faulted.
I found out that it seg-faulted when the last line of data file didn't
have the newline char.  For example, file like this:

23	3
31	2
24	1<EOF>


Here is a fix.

--- R-0.65.1/src/main/scan.c.orig       Wed Sep 29 11:11:45 1999
+++ R-0.65.1/src/main/scan.c    Fri Oct 29 17:34:22 1999
@@ -124,7 +124,7 @@
                if (bufp >= &buffer[MAXELTSIZE - 2])
                    continue;
                *bufp++ = c;
-           } while (!isspace(c = scanchar()) && c != R_EOF);
+           } while ((c = scanchar()) != R_EOF && !isspace(c));
            while (c==' ' || c=='\t')
                c=scanchar();
            if (c=='\n' || c=='\r' || c==R_EOF)
@@ -647,7 +647,7 @@
                }
            }
            else {
-               while (!isspace(c=scanchar()) && c != R_EOF)
+               while ( (c=scanchar()) != R_EOF && !isspace(c))
                    ;
                if (c==R_EOF) c='\n';
                unscanchar(c);



When the scanchar() returns R_EOF (=65535 #defined in src/include/Defn.h),
isspace(R_EOF) seq-faults.  By switching the order of two conditions, I could
get around with it.  I just made a test to see the behavior of isspace() by
giving
isspace(65535).  It seg-faults in both Linux/alpha and i386 (is it supposed
to??).
However, R on i386 doesn't seg-faults when the last line does not have
newline.

Naoki


-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To:
r-devel-request@stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._

ntakebay@bio.indiana.edu writes:
> Here is a fix.
> 
> --- R-0.65.1/src/main/scan.c.orig       Wed Sep 29 11:11:45 1999
> +++ R-0.65.1/src/main/scan.c    Fri Oct 29 17:34:22 1999
> @@ -124,7 +124,7 @@
>                 if (bufp >= &buffer[MAXELTSIZE - 2])
>                     continue;
>                 *bufp++ = c;
> -           } while (!isspace(c = scanchar()) && c != R_EOF);
> +           } while ((c = scanchar()) != R_EOF && !isspace(c));
....
> When the scanchar() returns R_EOF (=65535 #defined in src/include/Defn.h),
> isspace(R_EOF) seq-faults.  By switching the order of two conditions, I
could
> get around with it.  I just made a test to see the behavior of isspace() by
> giving
> isspace(65535).  It seg-faults in both Linux/alpha and i386 (is it supposed
> to??).
> However, R on i386 doesn't seg-faults when the last line does not have
newline.
Thanks. However, perhaps a better fix would be to change R_EOF? All
the isxxx macros/functions work by indexing an array designed to take
values between -128 and +255, so passing 65535 is playing with fire,
and we have many more instances littered all over the code, where I
don't feel absolutely sure that R_EOF couldn't sneak in.

Any one remember why R_EOF is different from the customary EOF of -1??

-- 
   O__  ---- Peter Dalgaard             Blegdamsvej 3  
  c/ /'_ --- Dept. of Biostatistics     2200 Cph. N   
 (*) \(*) -- University of Copenhagen   Denmark      Ph: (+45) 35327918
~~~~~~~~~~ - (p.dalgaard@biostat.ku.dk)             FAX: (+45) 35327907
-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To:
r-devel-request@stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._