--- TODO 2001/11/13 14:43:43 1.4
+++ TODO 2001/11/14 12:59:24 1.5
@@ -4,3 +4,101 @@
Regexp-Ausdrücken enthalten.
- Unterstützung von PCRE.
+
+
+---- RSE: --------------------------------------------------------------------
+
+- HAS TO BE: adding a var_error() function which translates var_rc_t
+ into a text version ala strerror(3).
+
+- HAS TO BE: still missing PCRE support plus corresponding Autoconf stuff
+
+- CAN BE: as in L2, reduce the large 1024 auto-variable with a a lot
+ smaller (on 32-bit boxes just 21 bytes) but still fully
+ sufficiently-sized variable:
+
+Index: var.c
+===================================================================
+RCS file: /u/rse/arc/cvs/ossp/ossp-pkg/var/var.c,v
+retrieving revision 1.14
+diff -u -d -u -d -r1.14 var.c
+--- var.c 2001/11/14 11:11:01 1.14
++++ var.c 2001/11/14 11:37:54
+@@ -1261,8 +1261,8 @@
+
+ case '#': /* Substitute length of the string. */
+ if (data->begin) {
+- char buf[1024];
+- sprintf(buf, "%d", data->end - data->begin);
++ char buf[((sizeof(int)*8)/3)+10]; /* sufficient size: <#bits> x log_10(2) + safety */
++ sprintf(buf, "%d", (int)(data->end - data->begin));
+ tokenbuf_free(data);
+ if (!tokenbuf_assign(data, buf, strlen(buf))) {
+ rc = VAR_ERR_OUT_OF_MEMORY;
+
+- SHOULD BE: the "magic" number 256 in name class definitions would be
+ nice if replaced by a symbolic name
+
+- SHOULD BE: the expand_octal() function checks that in the octal number
+ NNN (representing a character) the first number is not larger than
+ 3 because the 0xff is (octal) 377. Unfortunately this still allows
+ things like 399 which is illegal because larger than 0xff (the maximum
+ fitting into an 8-bit character). My suggestions:
+
+Index: var.c
+===================================================================
+RCS file: /u/rse/arc/cvs/ossp/ossp-pkg/var/var.c,v
+retrieving revision 1.15
+diff -u -d -u -d -r1.15 var.c
+--- var.c 2001/11/14 12:02:40 1.15
++++ var.c 2001/11/14 12:14:49
+@@ -209,7 +209,7 @@
+
+ static var_rc_t expand_octal(const char **src, char **dst, const char *end)
+ {
+- unsigned char c;
++ unsigned int c;
+
+ if (end - *src < 3)
+ return VAR_ERR_INCOMPLETE_OCTAL;
+@@ -217,8 +217,6 @@
+ return VAR_ERR_INVALID_OCTAL;
+
+ c = **src - '0';
+- if (c > 3)
+- return VAR_ERR_OCTAL_TOO_LARGE;
+ c *= 8;
+ ++(*src);
+
+@@ -227,6 +225,9 @@
+ ++(*src);
+
+ c += **src - '0';
++
++ if (c > 0xff)
++ return VAR_ERR_OCTAL_TOO_LARGE;
+
+ **dst = (char) c;
+ ++(*dst);
+
+- SHOULD BE: the tokenbuf functions like tokenbuf_append() return
+ 1 or 0 and the other routines convert this into var_rc_t values. I
+ recommend to change the tokenbuf functions to already return var_rc_t
+ values and pass this value through in other functions. This way the
+ different errors in tokenbuf_append() are visible, too.
+
+- CAN BE: I IMHO would reduce the code size of the recursive descent parser
+ by 1. making the prototype and corresponding parameter names of all
+ functions identical (see for example text() for a current exclusion)
+ and then use a macro to replace the long prototype description on
+ every function.
+
+- SHOULD BE: the return code semantics are still not clear to me
+ (perhaps it is solved by documentation). Especially it confuses me a lot
+ that internally it uses "int" instead of "var_rc_t" and does arithmetics
+ like "1 + rc" before finanlly it is casted to an var_rc_t. So how should
+ anyone can do anything reasonable with return codes > 0. He does not
+ know what they mean, doesn't he?
+
+------------------------------------------------------------------------
+
|
