Skip to contentSkip to conversation

ip-link: in human readable output use dynamic precision length

PATCH Superseded Christian Hesse 2 participants 7 messages Nov 4, 2014 – Nov 5, 2014

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ip-link: in human readable output use dynamic precision length

Now that we use floating point numbers for human readable output we can calculate precision length on the fly.

---
 ip/ipaddress.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
ip/ipaddress.c +2 -2
index e240bb5..0ddcb0d 100644
@@ -343,8 +343,8 @@ static void print_num(FILE *fp, unsigned width, uint64_t count)
 		++prefix;
 	}
 
-	snprintf(buf, sizeof(buf), "%.1f%c%s", (double) count / powi, 
-		 *prefix, use_iec ? "i" : "");
+	snprintf(buf, sizeof(buf), "%.*f%c%s", 3 - snprintf(NULL, 0, "%"PRIu64, count / powi),
+		(double) count / powi, *prefix, use_iec ? "i" : "");
 
 	fprintf(fp, "%-*s ", width, buf);
 }
Signature · 1 line
2.1.3
#1 lore raw

Christian Hesse <[email protected]> on Tue, 2014/11/04 10:39:

Quoted 21 lines

Now that we use floating point numbers for human readable output we can calculate precision length on the fly.

---
 ip/ipaddress.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
ip/ipaddress.c +2 -2
index e240bb5..0ddcb0d 100644
@@ -343,8 +343,8 @@ static void print_num(FILE *fp, unsigned width,
 uint64_t count) ++prefix;
 	}
 
-	snprintf(buf, sizeof(buf), "%.1f%c%s", (double) count / powi, 
-		 *prefix, use_iec ? "i" : "");
+	snprintf(buf, sizeof(buf), "%.*f%c%s", 3 - snprintf(NULL, 0,
 "%"PRIu64, count / powi),
+		(double) count / powi, *prefix, use_iec ? "i" : "");
 
 	fprintf(fp, "%-*s ", width, buf);
Quoted 1 line

}

Damn... In IEC mode we have negative precision length for values between 1000 and 1024. Will send a new patch.

Signature · 2 lines
Best Regards,
Christian Hesse
#2 lore raw

PATCH v2 1/1 ip-link: in human readable output use dynamic precision length

---
 ip/ipaddress.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)
ip/ipaddress.c +6 -2
index e240bb5..55cbc77 100644
@@ -324,6 +324,7 @@ static void print_num(FILE *fp, unsigned width, uint64_t count)
 	const char *prefix = "kMGTPE";
 	const unsigned int base = use_iec ? 1024 : 1000;
 	uint64_t powi = 1;
+	int precision;
 	char buf[64];
 
 	if (!human_readable || count < base) {
@@ -343,8 +344,11 @@ static void print_num(FILE *fp, unsigned width, uint64_t count)
 		++prefix;
 	}
 
-	snprintf(buf, sizeof(buf), "%.1f%c%s", (double) count / powi, 
-		 *prefix, use_iec ? "i" : "");
+	if ((precision = 3 - snprintf(NULL, 0, "%"PRIu64, count / powi)) < 0)
+		precision = 0;
+
+	snprintf(buf, sizeof(buf), "%.*f%c%s", precision,
+		(double) count / powi, *prefix, use_iec ? "i" : "");
 
 	fprintf(fp, "%-*s ", width, buf);
 }
Signature · 1 line
2.1.3
#3 lore raw

From: Christian Hesse
...

Quoted 13 lines
ip/ipaddress.c +1 -0
index e240bb5..55cbc77 100644
@@ -324,6 +324,7 @@ static void print_num(FILE *fp, unsigned width, uint64_t count)
 	const char *prefix = "kMGTPE";
 	const unsigned int base = use_iec ? 1024 : 1000;
 	uint64_t powi = 1;
+	int precision;
 	char buf[64];
 
 	if (!human_readable || count < base) {

ip/ipaddress.c static void print_num(FILE *fp, unsigned width, uint64_t count) · line 347

Quoted 6 lines
ip/ipaddress.c +1 -2
@@ -343,8 +344,11 @@ static void print_num(FILE *fp, unsigned width, uint64_t count)
 		++prefix;
 	}
 
-	snprintf(buf, sizeof(buf), "%.1f%c%s", (double) count / powi,
-		 *prefix, use_iec ? "i" : "");
+	if ((precision = 3 - snprintf(NULL, 0, "%"PRIu64, count / powi)) < 0)

Don't put assignments in conditionals.

Quoted 9 lines
ip/ipaddress.c +4 -1
+		precision = 0;
+
+	snprintf(buf, sizeof(buf), "%.*f%c%s", precision,
+		(double) count / powi, *prefix, use_iec ? "i" : "");
 
 	fprintf(fp, "%-*s ", width, buf);
 }
--

2.1.3

The above will go wrong in all sorts of horrid ways.... For instance you are doing a truncating integer divide, but the FP value will get rounded for display.

It would be safer to use integers throughout.

Oh, and a 2Mbit E1 link is actually 2048000 :-)

David

#4 lore raw

David Laight <[email protected]> on Tue, 2014/11/04 11:06:

Quoted 7 lines

From: Christian Hesse
...

Quoted 5 lines

...

(quoted diff) +0 -0
@@ -343,8 +344,11 @@ static void print_num(FILE *fp, unsigned width,
 uint64_t count) ++prefix;
 	}
 
Quoted 6 lines

ip/ipaddress.c static void print_num(FILE *fp, unsigned width, uint64_t count)

Quoted 4 lines
ip/ipaddress.c +1 -2
-	snprintf(buf, sizeof(buf), "%.1f%c%s", (double) count / powi,
-		 *prefix, use_iec ? "i" : "");
+	if ((precision = 3 - snprintf(NULL, 0, "%"PRIu64, count / powi))
 < 0)

Don't put assignments in conditionals.

Ok. :D

I do not like this at all... snprintf() would be nice for a catch-all, but we have to take care of negative values. So let's try something different. I will think about it and send a new patch.

Quoted 5 lines
Quoted 5 lines
ip/ipaddress.c +4 -0
+		precision = 0;
+
+	snprintf(buf, sizeof(buf), "%.*f%c%s", precision,
+		(double) count / powi, *prefix, use_iec ? "i" : "");
 
Quoted 8 lines
Quoted 2 lines
ip/ipaddress.c +0 -0
 	fprintf(fp, "%-*s ", width, buf);

}

The above will go wrong in all sorts of horrid ways.... For instance you are doing a truncating integer divide, but the FP value will get rounded for display.

It would be safer to use integers throughout.

My implementation used integers, but Stephen changes this to floating point with his cleanups.

IMHO the rounding is ok. This is for *human* readability. ;) Whoever wants correct values should not ask ip to print human readable values but rely on pure numbers.

Quoted 1 line

Oh, and a 2Mbit E1 link is actually 2048000 :-)

Sorry? Did not get the point.

Signature · 2 lines
Best regards,
Chris
#5 lore raw

PATCH v3 1/1 ip-link: in human readable output use dynamic precision length

---
 ip/ipaddress.c | 13 +++++++++++--
 1 file changed, 11 insertions(+), 2 deletions(-)
ip/ipaddress.c +11 -2
index e240bb5..db39437 100644
@@ -324,6 +324,8 @@ static void print_num(FILE *fp, unsigned width, uint64_t count)
 	const char *prefix = "kMGTPE";
 	const unsigned int base = use_iec ? 1024 : 1000;
 	uint64_t powi = 1;
+	uint16_t powj = 1;
+	uint8_t precision = 2;
 	char buf[64];
 
 	if (!human_readable || count < base) {
@@ -343,8 +345,15 @@ static void print_num(FILE *fp, unsigned width, uint64_t count)
 		++prefix;
 	}
 
-	snprintf(buf, sizeof(buf), "%.1f%c%s", (double) count / powi, 
-		 *prefix, use_iec ? "i" : "");
+	/* try to guess a good number of digits for precision */
+	for (; precision > 0; precision--) {
+		powj *= 10;
+		if (count / powi < powj)
+			break;
+	}
+
+	snprintf(buf, sizeof(buf), "%.*f%c%s", precision,
+		(double) count / powi, *prefix, use_iec ? "i" : "");
 
 	fprintf(fp, "%-*s ", width, buf);
 }
Signature · 1 line
2.1.3
#6 lore raw
In reply to Christian Hesse

From: Christian Hesse
...

Quoted 17 lines

I do not like this at all... snprintf() would be nice for a catch-all, but we have to take care of negative values. So let's try something different. I will think about it and send a new patch.

Quoted 12 lines

ip/ipaddress.c static void print_num(FILE *fp, unsigned width, uint64_t count) · line 354

Quoted 7 lines
ip/ipaddress.c +4 -0
+		precision = 0;
+
+	snprintf(buf, sizeof(buf), "%.*f%c%s", precision,
+		(double) count / powi, *prefix, use_iec ? "i" : "");
 
 	fprintf(fp, "%-*s ", width, buf);
 }

The above will go wrong in all sorts of horrid ways.... For instance you are doing a truncating integer divide, but the FP value will get rounded for display.

Quoted 8 lines
Quoted 1 line

It would be safer to use integers throughout.

My implementation used integers, but Stephen changes this to floating point with his cleanups.

IMHO the rounding is ok. This is for *human* readability. ;) Whoever wants correct values should not ask ip to print human readable values but rely on pure numbers.

Rounding the value is fine, the difference between 1000 and 1024 is not that great either.
The problem is that the two calculations get rounded differently. If the FP value ends up being 9.999999 it will be printed as 10.0 but your field width calculation will only have allowed for a single digit.

Quoted 3 lines
Quoted 1 line

Oh, and a 2Mbit E1 link is actually 2048000 :-)

Sorry? Did not get the point.

That was really in reference to a much earlier comment about comms always using 1000.

David