From nobody Wed Jul  5 10:39:02 2006
From: Aharon Robbins <arnold@skeeve.com>
Subject: Re: conversion error
To: bug-gawk@gnu.org, Heiner.Marxen@DrB.Insel.DE
Cc: 
Date: Tue, 04 Jul 2006 22:46:47 +0300

Greetings. Re this:

> Date: Tue, 04 Jul 2006 21:06:14 +0200 (MEST)
> From: Heiner Marxen <Heiner.Marxen@DrB.Insel.DE>
> Subject: conversion error
> To: bug-gawk@gnu.org
>
> Hello,
>
> The following awk script fails for gawk 3.1.4 and 3.1.5.
> Older versions did not do this, but I cannot say, how old they were.
>
> BEGIN {
>     if( 0 ) {		#ok
> 	t = "8"
>     }else {		#fails
> 	t = ""
> 	t = t "8"
>     }
>     printf("8  = %d\n", 0+t)	# ok without this line
>     t = t "8"			# does not invalidate numeric interpretation
>     printf("88 = %s\n", 0+t)
>     ## The above prints "88 = 8" with gawk 3.1.4 and 3.1.5
> }
>
> The following one-liner already exhibits the bug:
>
> gawk 'BEGIN{t=""; t=t "8";printf("8=%d\n", 0+t);t=t "8";printf("88=%s\n", 0+t)}'
>
>
> Preliminary observation: under somewhat strange conditions a variable
> does retain its numeric interpretation although something is appended to it.
> -- 
> Heiner Marxen				http://www.drb.insel.de/~heiner/

This is an excellent bug report and test case. Much thanks. The following
patch fixes the problem.

Arnold
----------------------------------------------------
Tue Jul  4 22:43:05 2006  Arnold D. Robbins  <arnold@skeeve.com>

	* eval.c (interpret): Node_assign_concat case: Turn off NUMBER and NUMCUR
	flags in result.  Sheesh. Thanks to <Heiner.Marxen@DrB.Insel.DE> for finding
	the problem.

--- ../gawk-3.1.5/eval.c	2005-07-26 21:07:43.000000000 +0300
+++ eval.c	2006-07-04 22:41:25.000000000 +0300
@@ -1186,6 +1187,7 @@
 			unref(*lhs);
 			*lhs = make_str_node(nval, l->stlen + r->stlen, ALREADY_MALLOCED);
 		}
+		(*lhs)->flags &= ~(NUMCUR|NUMBER);
 		free_temp(r);
 
 		if (after_assign)


_______________________________________________
bug-gnu-utils@gnu.org
http://lists.gnu.org/mailman/listinfo/bug-gnu-utils