bin/102694: df(1) can forget to print the unit with -h

[Ricardo Nabinger Sanchez]

The following reply was made to PR bin/102694; it has been noted by GNATS.

From: Ricardo Nabinger Sanchez <rnsanchez at wait4.org>
To: bug-followup at FreeBSD.org
Cc: ceri at FreeBSD.org
Subject: Re: bin/102694: df(1) can forget to print the unit with -h
Date: Fri, 6 Oct 2006 23:36:34 -0300

 > Not sure if this is df(1) or libutil's humanize_number(3)
 > to be honest.
 
 I was looking at it and was able to reproduce it and found what's going on.
 
 What happens is that, for numbers close to the next scale (1000<= x <1024),
 the scaled number is big for the 6-char buffer, which in fact is told be be
 6-1 bytes.  From the source:
 
 humanize_number(buf, sizeof(buf) - (bytes < 0 ? 0 : 1),
 	    bytes, "", HN_AUTOSCALE, flags);
 
 That "sizeof(buf) - (bytes < 0 ? 0 : 1)" causes humanize_number to think that
 the buffer has only 5 chars, thus producing:
 
 	[ ][1][0][0][0][\0]
 
 If one remove that restriction, the unit comes back:
 
 	[1][0][0][0][M][\0]
 
 The drawback is that it no longer scales "humanly" for numbers >= 1024, going
 up to 9999 (ugly).
 
 I spent a good time trying to figure out a clean and effective solution, but
 the best I got was to not use humanize_number, and scale the number locally.
 
 If that is good enough, I can clean up my code and send a patch.
 
 Regards.
 
 -- 
 Ricardo Nabinger Sanchez     <rnsanchez@{gmail.com,wait4.org}>
 Powered by FreeBSD
 
   "Left to themselves, things tend to go from bad to worse."