Issue 7090 in pharo: Should Float>>#printOn:base: use the slower exact printing ?

Status: Accepted
Owner: ----
Labels: Milestone-2.0 Type-Feature

New issue 7090 by [hidden email]: Should Float>>#printOn:base: use  
the slower exact printing ?
http://code.google.com/p/pharo/issues/detail?id=7090

Some recent Float printing enhancements led to the decision to use exact  
printing in #printOn:base:

The performance difference with inexact printing is however an order of a  
magnitude.

Should be not revise that decision ?

| data |
data := (1 to: 100000) collect: [ :each | each reciprocal asFloat ].
[ String streamContents: [ :stream |
        data do: [ :each | stream print: each; space ] ] ] timeToRun.  17776

| data |
data := (1 to: 100000) collect: [ :each | each reciprocal asFloat ].
[ String streamContents: [ :stream |
        data do: [ :each | each absPrintOn: stream base: 10. stream space ] ] ]  
timeToRun.  1324


_______________________________________________
Pharo-bugtracker mailing list
[hidden email]
http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-bugtracker

Comment #1 on issue 7090 by [hidden email]: Should  
Float>>#printOn:base: use the slower exact printing ?
http://code.google.com/p/pharo/issues/detail?id=7090

Make that "Should we not revise that decision ?"


_______________________________________________
Pharo-bugtracker mailing list
[hidden email]
http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-bugtracker

Comment #2 on issue 7090 by [hidden email]: Should  
Float>>#printOn:base: use the slower exact printing ?
http://code.google.com/p/pharo/issues/detail?id=7090

I would say yes (to revise)


_______________________________________________
Pharo-bugtracker mailing list
[hidden email]
http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-bugtracker

Comment #3 on issue 7090 by [hidden email]: Should  
Float>>#printOn:base: use the slower exact printing ?
http://code.google.com/p/pharo/issues/detail?id=7090

I would say no, don't revise.
Otherwise, Smalltalk will be the only language in the world to print  
approximately.
And that will break REPL and ASCII storage...
If speed matters that much, then let's increase speed with a primitive.
Do you know that LargeInteger still perform operation byte-wise in this 64  
bit era?
There is room for speed improvement.


_______________________________________________
Pharo-bugtracker mailing list
[hidden email]
http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-bugtracker

Comment #4 on issue 7090 by [hidden email]: Should Float>>#printOn:base: use  
the slower exact printing ?
http://code.google.com/p/pharo/issues/detail?id=7090

The speed difference is pretty big, wouldn't you say ?

[ String new: 2000000 streamContents: [ :stream |
        1 to: 100000 do: [ :each | each * Float pi absPrintExactlyOn: stream base:  
10. stream space ] ] ] timeToRun. 3471

[ String new: 2000000 streamContents: [ :stream |
        1 to: 100000 do: [ :each | each * Float pi absPrintOn: stream base: 10.  
stream space ] ] ] timeToRun. 754

I am not saying that we have to give up a good property.
If you can help optimize it a bit, that would be great.

What I am actually looking for is a way for client code to print  
objects/numbers faster, without having to test for Float and do something  
entirely different, which isn't very polymorphic. But I don't know the  
answer: #printOnFast: seems very silly.


_______________________________________________
Pharo-bugtracker mailing list
[hidden email]
http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-bugtracker

Updates:
        Cc: [hidden email] [hidden email]
        Labels: -Type-Feature Type-Enh

Comment #5 on issue 7090 by [hidden email]: Should  
Float>>#printOn:base: use the slower exact printing ?
http://code.google.com/p/pharo/issues/detail?id=7090

Is this a show-stopper for 2.0?


_______________________________________________
Pharo-bugtracker mailing list
[hidden email]
http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-bugtracker

Comment #6 on issue 7090 by [hidden email]: Should  
Float>>#printOn:base: use the slower exact printing ?
http://code.google.com/p/pharo/issues/detail?id=7090

No, it is not a showstopper, but anyone trying to print or write out a file  
(CSV, JSON, XML, ...) with 1000s of numbers in it will eventually hit this  
and ask themselves why...


_______________________________________________
Pharo-bugtracker mailing list
[hidden email]
http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-bugtracker

Comment #7 on issue 7090 by [hidden email]: Should  
Float>>#printOn:base: use the slower exact printing ?
http://code.google.com/p/pharo/issues/detail?id=7090

I am going to try to make a patch where a dynamic variable controls the  
float printing behavior.

Then we can test if it is fast enough.


_______________________________________________
Pharo-bugtracker mailing list
[hidden email]
http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-bugtracker

Updates:
        Status: FixReviewNeeded

Comment #8 on issue 7090 by [hidden email]: Should  
Float>>#printOn:base: use the slower exact printing ?
http://code.google.com/p/pharo/issues/detail?id=7090

In inbox:

---
Name:  
SLICE-Issue-7090-Should-FloatgtgtprintOnbase-use-the-slower-exact-printing--SvenVanCaekenberghe.1
Author: SvenVanCaekenberghe
Time: 8 February 2013, 12:15:38.79 pm
UUID: 79a1110f-0a4e-4731-908e-be7a84ffbf4d
Ancestors:
Dependencies: Kernel-SvenVanCaekenberghe.1303

introduction of FloatPrintPolicy, a dynamic variable to control how  
Float>>#printOn:base: works;
ExactFloatPrintPolicy double dispatches to Float>>#absPrintExactlyOn:base:  
for slower but accurate printing [ the current default ];
InexactFloatPrintPolicy double dispatches to Float>>#absPrintOn:base: for  
faster but potentially less accurate printing.

FloatPrintPolicy
        value: InexactFloatPrintPolicy new
        during: [ Float pi printString ]
---


_______________________________________________
Pharo-bugtracker mailing list
[hidden email]
http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-bugtracker

Comment #9 on issue 7090 by [hidden email]: Should  
Float>>#printOn:base: use the slower exact printing ?
http://code.google.com/p/pharo/issues/detail?id=7090

To compare the speed:

data := (1 to: 100000) collect: [ :each | each reciprocal asFloat ].
data := (1 to: 100000) collect: [ :each | each * Float pi ].

[ FloatPrintPolicy
        value: InexactFloatPrintPolicy new
        during: [
                String new: 2500000 streamContents: [ :stream |
                        data do: [ :each | stream print: each; space ] ] ] ] timeToRun.

[ String new: 2500000 streamContents: [ :stream |
        data do: [ :each | stream print: each; space ] ] ] timeToRun.



_______________________________________________
Pharo-bugtracker mailing list
[hidden email]
http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-bugtracker

Updates:
        Status: MonkeyIsChecking

Comment #10 on issue 7090 by [hidden email]: Should  
Float>>#printOn:base: use the slower exact printing ?
http://code.google.com/p/pharo/issues/detail?id=7090#c10

The Monkey is currently checking this issue. Please don't change it!


_______________________________________________
Pharo-bugtracker mailing list
[hidden email]
http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-bugtracker

Updates:
        Status: ValidatedByTheMonkey
        Labels: CheckedIn20524

Comment #11 on issue 7090 by [hidden email]: Should  
Float>>#printOn:base: use the slower exact printing ?
http://code.google.com/p/pharo/issues/detail?id=7090#c11

This Issue has been checked by Ulysse the Monkey
       
7632 tests passed in 00:02:57s:
===============================
        CollectionsTests-Arrayed (553)
        CollectionsTests-Atomic (12)
        CollectionsTests-Sequenceable (912)
        CollectionsTests-SplitJoin (27)
        CollectionsTests-Stack (16)
        CollectionsTests-Streams (37)
        CollectionsTests-Strings (611)
        CollectionsTests-Support (12)
        CollectionsTests-Unordered (1954)
        CollectionsTests-Weak (739)
        CompilerTests (181)
        KernelTests-Chronology (1186)
        KernelTests-Classes (138)
        KernelTests-Exception (5)
        KernelTests-Methods (359)
        KernelTests-Numbers (552)
        KernelTests-Objects (172)
        KernelTests-Pragmas (6)
        KernelTests-Processes (76)
        SUnit-Core-Extensions (3)
        SUnit-Core-Utilities (3)
        SUnit-Tests-Core (78)

----------------------------------------------------------
Loaded Source:  
SLICE-Issue-7090-Should-FloatgtgtprintOnbase-use-the-slower-exact-printing--SvenVanCaekenberghe.1  
from http://ss3.gemstone.com/ss/PharoInbox
Tested using Pharo-2.0-20524-a on NBCoInterpreter  
NativeBoost-CogPlugin-IgorStasenko.15 uuid:  
44b6b681-38f1-4a9e-b6ee-8769b499576a Nov 27 2012
NBCogit NativeBoost-CogPlugin-IgorStasenko.15 uuid:  
44b6b681-38f1-4a9e-b6ee-8769b499576a Nov 27 2012
https://git.gitorious.org/cogvm/blessed.git Commit:  
40ac7e7bdec6fef0e934d2c019b86db996053912 Date: 2012-11-19 18:54:49 +0100  
By: Mariano Martinez Peck <[hidden email]> Jenkins build #146


_______________________________________________
Pharo-bugtracker mailing list
[hidden email]
http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-bugtracker

Comment #12 on issue 7090 by [hidden email]: Should  
Float>>#printOn:base: use the slower exact printing ?
http://code.google.com/p/pharo/issues/detail?id=7090

Some human review would not hurt.

The code works, it is just that it adds to Kernel-Numbers


_______________________________________________
Pharo-bugtracker mailing list
[hidden email]
http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-bugtracker

Updates:
        Status: HumanReviewNeeded

Comment #13 on issue 7090 by [hidden email]: Should  
Float>>#printOn:base: use the slower exact printing ?
http://code.google.com/p/pharo/issues/detail?id=7090

(No comment was entered for this change.)


_______________________________________________
Pharo-bugtracker mailing list
[hidden email]
http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-bugtracker