--- Begin Message ---
|
Subject: |
Lisp reader mishandles large non-base-10 integers on 64-bit hosts |
|
Date: |
Sun, 01 May 2011 20:02:55 -0700 |
|
User-agent: |
Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.9.2.14) Gecko/20110223 Thunderbird/3.1.8 |
In the Emacs trunk I found some more problems with the Lisp
reader and large integers. It uses a floating-point number
to keep track of the integer's value, which leads to incorrect
answers with large integers on 64-bit hosts. In some cases the
errors are fairly extreme. I plan to install the following patch
after some more testing.
* lread.c (read_integer): Be more consistent with string-to-number.
Use string_to_number to do the actual conversion; this avoids
rounding errors and fixes some other screwups. Without this fix,
for example, #x1fffffffffffffff was misread as -2305843009213693952.
(digit_to_number): Move earlier, for benefit of read_integer.
Return -1 if the digit is out of range for the base, -2 if it is
not a digit in any supported base.
=== modified file 'src/lread.c'
--- src/lread.c 2011-04-29 07:55:25 +0000
+++ src/lread.c 2011-05-02 02:18:43 +0000
@@ -2245,6 +2245,26 @@
}
}
+/* Return the digit that CHARACTER stands for in the given BASE.
+ Return -1 if CHARACTER is out of range for BASE,
+ and -2 if CHARACTER is not valid for any supported BASE. */
+static inline int
+digit_to_number (int character, int base)
+{
+ int digit;
+
+ if ('0' <= character && character <= '9')
+ digit = character - '0';
+ else if ('a' <= character && character <= 'z')
+ digit = character - 'a' + 10;
+ else if ('A' <= character && character <= 'Z')
+ digit = character - 'A' + 10;
+ else
+ return -2;
+
+ return digit < base ? digit : -1;
+}
+
/* Read an integer in radix RADIX using READCHARFUN to read
characters. RADIX must be in the interval [2..36]; if it isn't, a
read error is signaled . Value is the integer read. Signals an
@@ -2254,59 +2274,64 @@
static Lisp_Object
read_integer (Lisp_Object readcharfun, int radix)
{
- int ndigits = 0, invalid_p, c, sign = 0;
- /* We use a floating point number because */
- double number = 0;
+ /* Room for sign, leading 0, other digits, trailing null byte. */
+ char buf[1 + 1 + sizeof (uintmax_t) * CHAR_BIT + 1];
+
+ int valid = -1; /* 1 if valid, 0 if not, -1 if incomplete. */
if (radix < 2 || radix > 36)
- invalid_p = 1;
+ valid = 0;
else
{
- number = ndigits = invalid_p = 0;
- sign = 1;
+ char *p = buf;
+ int c, digit;
c = READCHAR;
- if (c == '-')
+ if (c == '-' || c == '+')
{
+ *p++ = c;
c = READCHAR;
- sign = -1;
- }
- else if (c == '+')
- c = READCHAR;
-
- while (c >= 0)
- {
- int digit;
-
- if (c >= '0' && c <= '9')
- digit = c - '0';
- else if (c >= 'a' && c <= 'z')
- digit = c - 'a' + 10;
- else if (c >= 'A' && c <= 'Z')
- digit = c - 'A' + 10;
+ }
+
+ if (c == '0')
+ {
+ *p++ = c;
+ valid = 1;
+
+ /* Ignore redundant leading zeros, so the buffer doesn't
+ fill up with them. */
+ do
+ c = READCHAR;
+ while (c == '0');
+ }
+
+ while (-1 <= (digit = digit_to_number (c, radix)))
+ {
+ if (digit == -1)
+ valid = 0;
+ if (valid < 0)
+ valid = 1;
+
+ if (p < buf + sizeof buf - 1)
+ *p++ = c;
else
- {
- UNREAD (c);
- break;
- }
-
- if (digit < 0 || digit >= radix)
- invalid_p = 1;
-
- number = radix * number + digit;
- ++ndigits;
+ valid = 0;
+
c = READCHAR;
}
+
+ if (c >= 0)
+ UNREAD (c);
+ *p = '\0';
}
- if (ndigits == 0 || invalid_p)
+ if (! valid)
{
- char buf[50];
sprintf (buf, "integer, radix %d", radix);
invalid_syntax (buf, 0);
}
- return make_fixnum_or_float (sign * number);
+ return string_to_number (buf, radix, 0);
}
@@ -3165,23 +3190,6 @@
}
-static inline int
-digit_to_number (int character, int base)
-{
- int digit;
-
- if ('0' <= character && character <= '9')
- digit = character - '0';
- else if ('a' <= character && character <= 'z')
- digit = character - 'a' + 10;
- else if ('A' <= character && character <= 'Z')
- digit = character - 'A' + 10;
- else
- return -1;
-
- return digit < base ? digit : -1;
-}
-
#define LEAD_INT 1
#define DOT_CHAR 2
#define TRAIL_INT 4
--- End Message ---
--- Begin Message ---
|
Subject: |
Merged fixes for 8600, 8601, 8602, and (partially) for 8545 |
|
Date: |
Fri, 06 May 2011 00:29:56 -0700 |
|
User-agent: |
Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.9.2.17) Gecko/20110424 Thunderbird/3.1.10 |
I committed to the Emacs trunk a merge (bzr 104134) that has fixes for
the following bugs:
* Bug#8600 - The fix removes the garbage element of code_space.
* Bug#8601 - Here I assumed that the "* 2" is a typo.
* Bug#8602 - This fixes some large-int-to-float screwups in
the Lisp reader.
* Bug#8545 - This fixes the bug where the code should have called
va_copy, but didn't. Also, I changed a limit so that
the MOST_POSITIVE_FIXNUM limit for strings applies to
their length, i.e., does not include the null termination
byte. Stefan hasn't had time to chime in, but if this
last change turns out to be incorrect I will back it out.
This merge doesn't entirely fix Bug#8545, so I'll leave that bug open;
the others I'll close.
--- End Message ---