Bug #4005: YAML fails to roundtrip Time objects

• Status: Closed
• Assignee: -
• Target version: -
• ruby -v: ruby 1.9.3dev (2010-10-30 trunk 29630) [i686-linux]

[pweldon (Peter Weldon)]

=begin
$ rvm exec ruby -v -ryaml -e 't0 = Time.now; p YAML::load(YAML::dump(t0)) == t0'
ruby 1.9.3dev (2010-10-30 trunk 29630) [i686-linux]
false
ruby 1.8.7 (2010-08-16 patchlevel 302) [i686-linux]
true
ruby 1.9.1p378 (2010-01-10 revision 26273) [i686-linux]
false
ruby 1.9.2p0 (2010-08-18 revision 29036) [i686-linux]
false
=end

[zenspider (Ryan Davis)]

=begin

On Oct 29, 2010, at 11:26 , Peter Weldon wrote:

Bug #4005: YAML fails to roundtrip Time objects
http://redmine.ruby-lang.org/issues/show/4005

Author: Peter Weldon
Status: Open, Priority: Normal
ruby -v: ruby 1.9.3dev (2010-10-30 trunk 29630) [i686-linux]

$ rvm exec ruby -v -ryaml -e 't0 = Time.now; p YAML::load(YAML::dump(t0)) == t0'
ruby 1.9.3dev (2010-10-30 trunk 29630) [i686-linux]
false
ruby 1.8.7 (2010-08-16 patchlevel 302) [i686-linux]
true
ruby 1.9.1p378 (2010-01-10 revision 26273) [i686-linux]
false
ruby 1.9.2p0 (2010-08-18 revision 29036) [i686-linux]
false

I just got 4 trues... can you be more specific about what timezone you're in and roughly what that t0 was?

I'm in PST and ruby -e 'p Time.now' is currently "Fri Oct 29 17:08:27 -0700 2010". My env | grep TZ is empty.

=end

[pweldon (Peter Weldon)]

=begin
:~$ rvm exec ruby -v -ryaml -e 't0 = Time.now; p t0; p YAML::load(YAML::dump(t0)) == t0'
ruby 1.9.3dev (2010-10-30 trunk 29630) [i686-linux]
2010-10-30 02:30:09 +0000
false
ruby 1.8.7 (2010-08-16 patchlevel 302) [i686-linux]
Sat Oct 30 02:30:10 +0000 2010
true
ruby 1.9.1p378 (2010-01-10 revision 26273) [i686-linux]
2010-10-30 02:30:10 +0000
false
ruby 1.9.2p0 (2010-08-18 revision 29036) [i686-linux]
2010-10-30 02:30:11 +0000
false

~$ printenv | grep TZ
~$

=end

[naruse (Yui NARUSE)]

=begin
It seems from less than 1 second, so try following:
ruby -v -ryaml -e 't0 = Time.now;t2=YAML::load(t1=YAML::dump(t0));p [t0.to_f,t2.to_f];p t1'
=end

[tenderlovemaking (Aaron Patterson)]

=begin
On Sat, Oct 30, 2010 at 11:31:15AM +0900, Peter Weldon wrote:

Issue #4005 has been updated by Peter Weldon.

:~$ rvm exec ruby -v -ryaml -e 't0 = Time.now; p t0; p YAML::load(YAML::dump(t0)) == t0'
ruby 1.9.3dev (2010-10-30 trunk 29630) [i686-linux]
2010-10-30 02:30:09 +0000
false
ruby 1.8.7 (2010-08-16 patchlevel 302) [i686-linux]
Sat Oct 30 02:30:10 +0000 2010
true
ruby 1.9.1p378 (2010-01-10 revision 26273) [i686-linux]
2010-10-30 02:30:10 +0000
false
ruby 1.9.2p0 (2010-08-18 revision 29036) [i686-linux]
2010-10-30 02:30:11 +0000
false

~$ printenv | grep TZ
~$

I'm not able to reproduce this with Syck. Can you try with Psych
instead? Like this:

ruby -v -ryaml -e 'YAML::ENGINE.yamler = "psych"; t0 = Time.now; p t0; p YAML::load(YAML::dump(t0)) == t0'

I suspect your error is due to fractional second support. Observe the
output of Syck compared to Psych:

Using Syck

$ ruby -v -ryaml -e 't0 = Time.now; p t0; p YAML::dump(t0)'

"--- 2010-10-30 07:05:46.535362 -07:00\n"

Using Psych

ruby -v -ryaml -e 'YAML::ENGINE.yamler = "psych"; t0 = Time.now; p t0; p YAML::dump(t0)'

"--- 2010-10-30 07:06:48.118999000 -07:00\n...\n"

--
Aaron Patterson
http://tenderlovemaking.com/

Attachment: (unnamed)
=end

[darix (Marcus Rückert)]

=begin
On 2010-10-30 23:07:52 +0900, Aaron Patterson wrote:

I'm not able to reproduce this with Syck. Can you try with Psych
instead? Like this:

ruby -v -ryaml -e 'YAML::ENGINE.yamler = "psych"; t0 = Time.now; p t0; p YAML::load(YAML::dump(t0)) == t0'

I suspect your error is due to fractional second support. Observe the
output of Syck compared to Psych:

Using Syck

$ ruby -v -ryaml -e 't0 = Time.now; p t0; p YAML::dump(t0)'

"--- 2010-10-30 07:05:46.535362 -07:00\n"

Using Psych

ruby -v -ryaml -e 'YAML::ENGINE.yamler = "psych"; t0 = Time.now; p t0; p YAML::dump(t0)'

"--- 2010-10-30 07:06:48.118999000 -07:00\n...\n"

$ ruby1.9 -v -ryaml -e 'YAML::ENGINE.yamler = "psych"; t0 = Time.now; p t0; p YAML::load(YAML::dump(t0)) == t0'
ruby 1.9.2p0 (2010-08-18 revision 29036) [x86_64-linux]
/usr/lib64/ruby/1.9.1/psych/deprecated.rb:79: warning: method redefined; discarding old to_yaml_properties
/usr/lib64/ruby/1.9.1/syck/rubytypes.rb:13: warning: previous definition of to_yaml_properties was here
2010-10-30 16:11:41 +0200
true

$ ruby1.9 -v -ryaml -e 't0 = Time.now; p t0; p YAML::load(YAML::dump(t0)) == t0'
ruby 1.9.2p0 (2010-08-18 revision 29036) [x86_64-linux]
2010-10-30 16:11:50 +0200
false

$ ruby1.9 -v -ryaml -e 'YAML::ENGINE.yamler = "psych"; t0 = Time.now; p t0; p YAML::dump(t0)'
ruby 1.9.2p0 (2010-08-18 revision 29036) [x86_64-linux]
/usr/lib64/ruby/1.9.1/psych/deprecated.rb:79: warning: method redefined; discarding old to_yaml_properties
/usr/lib64/ruby/1.9.1/syck/rubytypes.rb:13: warning: previous definition of to_yaml_properties was here
2010-10-30 16:12:39 +0200
"--- 2010-10-30 16:12:39.608620883 +02:00\n...\n"

$ ruby1.9 -v -ryaml -e 't0 = Time.now; p t0; p YAML::dump(t0)'
ruby 1.9.2p0 (2010-08-18 revision 29036) [x86_64-linux]
2010-10-30 16:12:52 +0200
"--- 2010-10-30 16:12:52.775869 +02:00\n"

why does psych add the "...\n"?

hth

 darix

--
openSUSE - SUSE Linux is my linux
openSUSE is good for you
www.opensuse.org

=end

[zenspider (Ryan Davis)]

=begin

On Oct 30, 2010, at 07:14 , Marcus Rueckert wrote:

why does psych add the "...\n"?

Not psych exactly, but libyaml. It signifies the end of the document and is yaml 1.1 specific.

This uses psych and goes through libyaml:

% multiruby -1 1.9.2 -rpsych -ryaml -e 'p YAML::dump(42)'
"--- 42\n...\n"

This uses syck only:

% multiruby -1 1.9.2 -ryaml -e 'p YAML::dump(42)'
"--- 42\n"

I look forward to the day when syck is no longer in the ruby distribution and we have faster and more compliant yaml by default.

=end

[darix (Marcus Rückert)]

=begin
On 2010-10-31 05:03:56 +0900, Ryan Davis wrote:

On Oct 30, 2010, at 07:14 , Marcus Rueckert wrote:

why does psych add the "...\n"?

Not psych exactly, but libyaml. It signifies the end of the document
and is yaml 1.1 specific.

thanks for the explaination. can i help with more details with the date
roundtrip problem?

 darix

--
openSUSE - SUSE Linux is my linux
openSUSE is good for you
www.opensuse.org

=end

[pweldon (Peter Weldon)]

=begin
As pointed out fractional time does not appear to round trip very well:

:~$ rvm exec ruby -v -ryaml -e 't0 = Time.at(0.1); p YAML::load(YAML::dump(t0)) == t0'
ruby 1.9.3dev (2010-10-30 trunk 29648) [i686-linux]
false
ruby 1.8.7 (2010-08-16 patchlevel 302) [i686-linux]
true
ruby 1.9.1p378 (2010-01-10 revision 26273) [i686-linux]
true
ruby 1.9.2p0 (2010-08-18 revision 29036) [i686-linux]
false

It may be inherent in the Time object as syck / psych / json suffer from it:

:~$ rvm exec ruby -v -rpsych -ryaml -e 't0 = Time.at(0.1); p YAML::load(YAML::dump(t0)) == t0'
ruby 1.9.3dev (2010-10-30 trunk 29648) [i686-linux]
false
ruby 1.9.2p0 (2010-08-18 revision 29036) [i686-linux]
false

:~$ ruby -v -rjson/add/core -e 't0 = Time.at(0.1); p JSON.parse(JSON.generate(t0)) == t0'
ruby 1.9.2p0 (2010-08-18 revision 29036) [i686-linux]
false

If you avoid using Time#at(Float) and explicitly pass in microseconds thing look better:

:~$ rvm exec ruby -v -ryaml -e 't0 = Time.at(0,100000); p YAML::load(YAML::dump(t0)) == t0'
ruby 1.9.3dev (2010-10-30 trunk 29648) [i686-linux]
true
ruby 1.8.7 (2010-08-16 patchlevel 302) [i686-linux]
true
ruby 1.9.1p378 (2010-01-10 revision 26273) [i686-linux]
true
ruby 1.9.2p0 (2010-08-18 revision 29036) [i686-linux]
true

Understanding why Time.at(0.1) and Time.at(0,100000) are not consistently equal/unequal might be a clue:

:~$ rvm exec ruby -v -e 'p Time.at(0.1) == Time.at(0,100000)'
ruby 1.9.3dev (2010-10-30 trunk 29648) [i686-linux]
false
ruby 1.8.7 (2010-08-16 patchlevel 302) [i686-linux]
true
ruby 1.9.1p378 (2010-01-10 revision 26273) [i686-linux]
true
ruby 1.9.2p0 (2010-08-18 revision 29036) [i686-linux]
false

=end

[naruse (Yui NARUSE)]

=begin

As pointed out fractional time does not appear to round trip very well:

Yes, this is originally intended.
From Ruby 1.9.2 Time stores its internal time as Rational, so it is more precise than before.
See Time#to_r or time.c (if you are interested).

I understand your trouble.
So if you suggest a reasonable feature, we can discuss about it.
=end

[pweldon (Peter Weldon)]

• File 0001-Make-syck-s-Time-timestamps-write-nsec-precision-fra.patch 0001-Make-syck-s-Time-timestamps-write-nsec-precision-fra.patch added

=begin
Patch syck's handling of fractional seconds. Round-tripping of nsec Time values now works.
=end

[naruse (Yui NARUSE)]

• Status changed from Open to Closed

1.9.3's YAML now dumps 9 digits of fractional (6 digits before).
So this seems resolved.