Update MPFI to 1.5.1

[mhansen]

This is needed for #11666.

Use  http://perso.telecom-paristech.fr/~flori/sage/mpfi-1.5.1.spkg

Apply first trac_12171a.patch , then trac_12171b.patch

[mhansen]

There is spkg at  http://sage.math.washington.edu/home/mhansen/mpfi-1.5.0.spkg

[fbissey]

Have you checked the following doctests:

sage -t -long "devel/sage/sage/rings/qqbar.py"              
**********************************************************************
File "/usr/share/sage/devel/sage/sage/rings/qqbar.py", line 3808:
    sage: (x-2).real_exact(RR)
Expected:
    0.000000000000000
Got:
    -0.000000000000000
**********************************************************************
File "/usr/share/sage/devel/sage/sage/rings/qqbar.py", line 3810:
    sage: (x-2).real_exact(RealField(53, rnd='RNDD'))
Expected:
    0.000000000000000
Got:
    -0.000000000000000
**********************************************************************
File "/usr/share/sage/devel/sage/sage/rings/qqbar.py", line 3812:
    sage: (x-2).real_exact(RealField(53, rnd='RNDU'))
Expected:
    0.000000000000000
Got:
    -0.000000000000000
**********************************************************************
File "/usr/share/sage/devel/sage/sage/rings/qqbar.py", line 3814:
    sage: (x-2).real_exact(RealField(53, rnd='RNDZ'))
Expected:
    0.000000000000000
Got:
    -0.000000000000000
**********************************************************************
1 items had failures:
   4 of  20 in __main__.example_96
***Test Failed*** 4 failures.

and

sage -t -long "devel/sage/sage/rings/real_mpfi.pyx"         
**********************************************************************
File "/usr/share/sage/devel/sage/sage/rings/real_mpfi.pyx", line 1078:
    sage: b = R(1)/R(0); b
Expected:
    [+infinity .. +infinity]
Got:
    [-infinity .. +infinity]
**********************************************************************
File "/usr/share/sage/devel/sage/sage/rings/real_mpfi.pyx", line 1080:
    sage: loads(dumps(b)) == b
Expected:
    True
Got:
    False
**********************************************************************
File "/usr/share/sage/devel/sage/sage/rings/real_mpfi.pyx", line 1082:
    sage: b = R(-1)/R(0); b
Expected:
    [-infinity .. -infinity]
Got:
    [-infinity .. +infinity]
**********************************************************************
File "/usr/share/sage/devel/sage/sage/rings/real_mpfi.pyx", line 1084:
    sage: loads(dumps(b)) == b
Expected:
    True
Got:
    False
**********************************************************************
File "/usr/share/sage/devel/sage/sage/rings/real_mpfi.pyx", line 2360:
    sage: R(1)/R(0)
Expected:
    [+infinity .. +infinity]
Got:
    [-infinity .. +infinity]
**********************************************************************
File "/usr/share/sage/devel/sage/sage/rings/real_mpfi.pyx", line 3217:
    sage: RIF(-1, 1).min(0).endpoints()
Expected:
    (-1.00000000000000, 0.000000000000000)
Got:
    (-1.00000000000000, -0.000000000000000)
**********************************************************************
File "/usr/share/sage/devel/sage/sage/rings/real_mpfi.pyx", line 3221:
    sage: RIF(-1, 1).min(0).endpoints()
Expected:
    (-1.00000000000000, 0.000000000000000)
Got:
    (-1.00000000000000, -0.000000000000000)
**********************************************************************
3 items had failures:
   4 of  18 in __main__.example_29
   1 of   8 in __main__.example_55
   2 of   9 in __main__.example_80
***Test Failed*** 7 failures.

I did those test about one year ago  https://github.com/cschwan/sage-on-gentoo/issues/30 so some things have probably changed a little bit but I expect you'll have to patch sage.

[zimmerma]

by the way, MPFI 1.5.0 fixes a limitation to 1000 characters in mpfi_set_str, which can be demonstrated by the following test:

sage: R = RealIntervalField(53)
sage: R("1" * 1000)
1.111111111111112?e999

which failed previously. Maybe add a test for that?

Paul

[zimmerma]

two doctests are failing:

        sage -t  devel/sage-11666/sage/rings/qqbar.py # 4 doctests failed
        sage -t  devel/sage-11666/sage/rings/real_mpfi.pyx # 7 doctests failed

The issues are +0.0 versus -0.0, and also 1/0 which now gives [-Inf, +Inf] instead of [+Inf, +Inf].

Paul

[zimmerma]

there is one important change in MPFI 1.5.0 which might cause the failures: mpfi_set_ui on input 0 gives [+0.0,-0.0] instead of [+0.0,+0.0] previously. I will ask upstream why:

sage: R=RealIntervalField(53)
sage: a=R(0)
sage: a.lower(), a.upper()
(0.000000000000000, -0.000000000000000)

More annoying is the following, which I believe is a bug in MPFI 1.5.0:

sage: a=R(infinity)
sage: a.lower(), a.upper()
(+infinity, +infinity)
sage: b=1/a
sage: b.lower(), b.upper()
(0.000000000000000, -0.000000000000000)

(I would expect b=[+0.0,+0.0].)

Paul

[zimmerma]

the detailed failures (on a 32-bit computer) are:

macaron% ./sage -t devel/sage-12171/sage/rings/real_mpfi.pyx 
sage -t  "devel/sage-12171/sage/rings/real_mpfi.pyx"        
**********************************************************************
File "/localdisk/tmp/sage-4.7.2/devel/sage-12171/sage/rings/real_mpfi.pyx", line 4142:
    sage: 0 in (RealField(2000)(-19/3).gamma() - RealIntervalField(1000)(-19/3).gamma())
Expected:
    True
Got:
    False
**********************************************************************
File "/localdisk/tmp/sage-4.7.2/devel/sage-12171/sage/rings/real_mpfi.pyx", line 1082:
    sage: b = R(1)/R(0); b
Expected:
    [+infinity .. +infinity]
Got:
    [-infinity .. +infinity]
**********************************************************************
File "/localdisk/tmp/sage-4.7.2/devel/sage-12171/sage/rings/real_mpfi.pyx", line 1084:
    sage: loads(dumps(b)) == b
Expected:
    True
Got:
    False
**********************************************************************
File "/localdisk/tmp/sage-4.7.2/devel/sage-12171/sage/rings/real_mpfi.pyx", line 1086:
    sage: b = R(-1)/R(0); b
Expected:
    [-infinity .. -infinity]
Got:
    [-infinity .. +infinity]

**********************************************************************
File "/localdisk/tmp/sage-4.7.2/devel/sage-12171/sage/rings/real_mpfi.pyx", line 1088:
    sage: loads(dumps(b)) == b
Expected:
    True
Got:
    False
**********************************************************************
File "/localdisk/tmp/sage-4.7.2/devel/sage-12171/sage/rings/real_mpfi.pyx", line 2364:
    sage: R(1)/R(0)
Expected:
    [+infinity .. +infinity]
Got:
    [-infinity .. +infinity]
**********************************************************************
File "/localdisk/tmp/sage-4.7.2/devel/sage-12171/sage/rings/real_mpfi.pyx", line 3221:
    sage: RIF(-1, 1).min(0).endpoints()
Expected:
    (-1.00000000000000, 0.000000000000000)
Got:
    (-1.00000000000000, -0.000000000000000)
**********************************************************************
File "/localdisk/tmp/sage-4.7.2/devel/sage-12171/sage/rings/real_mpfi.pyx", line 3225:
    sage: RIF(-1, 1).min(0).endpoints()
Expected:
    (-1.00000000000000, 0.000000000000000)
Got:
    (-1.00000000000000, -0.000000000000000)

**********************************************************************
4 items had failures:
   1 of  19 in __main__.example_117
   4 of  19 in __main__.example_29
   1 of   9 in __main__.example_55
   2 of  10 in __main__.example_80
***Test Failed*** 8 failures.
For whitespace errors, see the file /users/caramel/zimmerma/.sage//tmp/real_mpfi_8949.py
         [8.1 s]

and:

macaron% ./sage -t devel/sage-12171/sage/rings/qqbar.py 
sage -t  "devel/sage-12171/sage/rings/qqbar.py"             
**********************************************************************
File "/localdisk/tmp/sage-4.7.2/devel/sage-12171/sage/rings/qqbar.py", line 4003:
    sage: (x-2).real_exact(RR)
Expected:
    0.000000000000000
Got:
    -0.000000000000000
**********************************************************************
File "/localdisk/tmp/sage-4.7.2/devel/sage-12171/sage/rings/qqbar.py", line 4005:
    sage: (x-2).real_exact(RealField(53, rnd='RNDD'))
Expected:
    0.000000000000000
Got:
    -0.000000000000000
**********************************************************************
File "/localdisk/tmp/sage-4.7.2/devel/sage-12171/sage/rings/qqbar.py", line 4007:
    sage: (x-2).real_exact(RealField(53, rnd='RNDU'))
Expected:
    0.000000000000000
Got:
    -0.000000000000000
**********************************************************************
File "/localdisk/tmp/sage-4.7.2/devel/sage-12171/sage/rings/qqbar.py", line 4009:
    sage: (x-2).real_exact(RealField(53, rnd='RNDZ'))

Expected:
    0.000000000000000
Got:
    -0.000000000000000
**********************************************************************
1 items had failures:
   4 of  21 in __main__.example_101
***Test Failed*** 4 failures.

Paul

[zimmerma]

the first issue with gamma seems to be due to that bug in Sage:

sage: RealIntervalField(54)(-19/3) > 1.462
True

(works for precision 53 or less).

Paul

[zimmerma]

I got a message from the developers of the Sollya package who say that they had similar problems when upgrading to MPFI 1.5.0. Their solution was to write a wrapper around MPFI to give sense to intervals like [+Inf,+Inf].

Paul

[zimmerma]

the problem from comment 9 seems to only apply on 32-bit machines. I'll first try to solve problems on 64-bit. There is one strange thing with the patch applied:

sage: x = AA(2).sqrt()^2
sage: (x-2).real_exact(RR)
0.000000000000000
sage: x.real_exact(RR)
2.00000000000000
sage: (x-2).real_exact(RR)
-0.000000000000000

The MPFI developers claim that in MPFI 1.5 the sign of 0 is irrelevant, but this is still a non-deterministic issue.

Paul

[zimmerma]

the attached trac_12171a.patch fixes all issues but two on 64-bit. The remaining issues are related to loads(dumps(b)) == b:

File "/usr/local/sage-4.7.2/sage/devel/sage-12171/sage/rings/real_mpfi.pyx", line 1084:
    sage: loads(dumps(b)) == b
Expected:
    True
Got:
    False
**********************************************************************
File "/usr/local/sage-4.7.2/sage/devel/sage-12171/sage/rings/real_mpfi.pyx", line 1088:
    sage: loads(dumps(b)) == b
Expected:
    True
Got:
    False

I see in real_mpfi.pyx that other similar tests are failing and were rewritten in a different way:

sage: R = RealIntervalField(4000)
sage: s = 1/R(3)
sage: t = loads(dumps(s))
sage: (t.upper(), t.lower()) == (s.upper(), s.lower())
True

Indeed t==s fails:

sage: t == s
False

Should I also rewrite the 2 failing tests like this? What is the reason for the failures?

Paul

[zimmerma]

Should I also rewrite the 2 failing tests like this? What is the reason for the failures?

Robert, Mike, are you there?

Paul

[zimmerma]

Jeroen, since I got no answer to my question I ask you: do you have an idea how to proceed?

Paul

[mhansen]

Hi Paul, sorry for the delay; I've been busy with traveling back to the states / the holidays. I'll look into why these are failing today. I started while I was in Warwick, but didn't make much progress.

[fbissey]

Replying to zimmerma:

I see in real_mpfi.pyx that other similar tests are failing and were rewritten in a different way:

sage: R = RealIntervalField(4000)
sage: s = 1/R(3)
sage: t = loads(dumps(s))
sage: (t.upper(), t.lower()) == (s.upper(), s.lower())
True

Indeed t==s fails:

sage: t == s
False

Should I also rewrite the 2 failing tests like this? What is the reason for the failures?

This is quite strange I cannot think of a reason this would fail unless the "==" operator is doing something slightly wrong or something unexpected.

[zimmerma]

ok I understand why the following "fails":

sage: R = RealIntervalField(100)
sage: s = 1/R(3)
sage: t = loads(dumps(s))
sage: t == s
False

In the real_mpfi.pyx file, it is said that s cmp t returns True if a cmp b is true for any element a in s and any element b in t. Obviously when cmp is == and t is the same interval as s this can hold only when both are the same point interval. Thus to check if s and t are the same interval, we should write:

sage: (t.upper(), t.lower()) == (s.upper(), s.lower())
True

Paul

[zimmerma]

this patch should be applied after trac_12171a.patch

[zimmerma]

the second patch attached fixes the remaining failing tests. Ready for review.

Paul

[fbissey]

There is no doubt that this second patch will do the trick and does the right thing. I am just wondering why "t == s" doesn't work as expected in the first place.

In any case I will do some basic reviewing shortly. The patch themselves look good to me. I need to check the spkg in details.

[zimmerma]

Francois, I explain in comment 18 that the semantics of t == s is special for intervals. If t=[a,b] and s=[c,d], it requires that x==y for any a <= x <= b and c <= y <= d, which obviously requires a=b=c=d.

The modified tests did work before because they had a=b=c=d, which is no longer the case due to change of semantics of MPFI.

Paul

[fbissey]

Just a quick update: it doesn't appear that mpfi runs a test suite when SAGE_CHECK is set can you fix that?

[fbissey]

I further inspected the spkg. There is one issue

 hg status
! patches/configure.ac
! patches/configure.ac.patch

You have to tell hg to remove them, just removing them with "rm" is not enough.

 hg addremove

will do the trick then you can commit and make the spkg. As mentioned above I would also like you to add a section to spkg-install that runs "make check" if SAGE_CHECK is set to yes. Once all that is done I'll have another look but it should be a positive review then.

[zimmerma]

Mike, please can you make the changes in the spkg? I'd like to make progress on this patch during SD 35.5, so that we can proceed on #11666.

Paul

[jpflori]

As I happened to be cc'ed on this ticket, I've uplaoded a package with the requested easy changes at  http://perso.telecom-paristech.fr/~flori/sage/mpfi-1.5.0.spkg

At least it install on a (vanilla? not sure... but nothing to exotic) sage 4.7.2 and runs the test suite (and succeed) when asked so on my machine.

[jpflori]

I've also put another version of the package at the same address + ".new" where the last test in spkg-install is slightly simplified...

[fbissey]

Looking at it again it may have been a little bit more verbose about the removal of the patches in SPKG.txt but it is otherwise fine now. It actually builds and tests fine on something exotic (SLES 11SP1 on power7, although there may be a slight clue about some problems on power7 in those tests).

I giving this a positive review.

[jpflori]

I've put the last package at the "normal" address (i.e. without .new at the end) and linked it in the ticket description.

[jdemeyer]

This should be rebased to #12131.

[fbissey]

I hadn't noticed that mpfi was one of the concerned package but I guess it make sense. This should be easy and quick to fix.

[jpflori]

The concerned line of spkg-install has been updated according to #12131.

You can find the updated package at the same address:

 http://perso.telecom-paristech.fr/~flori/sage/mpfi-1.5.0.spkg

[fbissey]

It is not how rebasing a spkg works Jean-Pierre (I had to do it several time for python-2.7). You have to start from the latest spkg available in #12131. The idea is that by doing what you just did you lost all the history from the spkg in #12131 that includes hg logs and the log entries from SPKG.txt.

So I am afraid you will have to take the spkg from #12131 and then add all the changes needed for 1.5.0.

[jpflori]

My bad, I did not really thought before adding the one line fix from #12131.

I'll do it the right way around later.

[zimmerma]

note that MPFI 1.5.1 will soon be out:  http://lists.gforge.inria.fr/pipermail/mpfi-users/2012-January/000031.html

Paul

[jpflori]

I'm currently properly rebuilding the spkg (for 1.5.0).

I see that in Mike package, he got rid of the previous patch for cygwin.

I guess it has been merged upstream. Anyway, I'll document it in SPKG.txt when I find a ref, remove src/configure.ac from tracked files and delete patches directory, unless someone has an objection.

If v1.5.1 happens to get released, I'll just use it instead of 1.5.

[jpflori]

I've contacted Philippe Théveny who is in charge of MPFI and he was not even aware of the patch we used in 1.3.4.something to compile under cygwin, so there is a priori no reason for the patch not to be needed anymore (I verified that the configure.ac file in Mike's spkg is the original one from MPFI 1.5).

The release is planned for the middle of next week, but if the patch is still needed Philippe is ready to get it merged upstream for that release. The original ticket for that issue is #3235.

Could someone who has access to cygwin try the last package I posted (or Mike's one) and report on success or failure ASAP?

[dimpase]

Replying to jpflori:

I've contacted Philippe Théveny who is in charge of MPFI and he was not even aware of the patch we used in 1.3.4.something to compile under cygwin, so there is a priori no reason for the patch not to be needed anymore (I verified that the configure.ac file in Mike's spkg is the original one from MPFI 1.5).

The release is planned for the middle of next week, but if the patch is still needed Philippe is ready to get it merged upstream for that release. The original ticket for that issue is #3235.

Could someone who has access to cygwin try the last package I posted (or Mike's one) and report on success or failure ASAP?

The spkg builds just fine on Sage 4.7.2 on Cygwin.

Dima

[zimmerma]

a release candidate for MPFI 1.5.1 is available:

Date: Fri, 13 Jan 2012 20:08:36 +0100
From: philipe theveny <philippe.theveny@ens-lyon.fr>
To: mpfi-users@lists.gforge.inria.fr
Subject: [Mpfi-users] announce: mpfi-1.5.1 release candidate
Content-Type: text/plain; charset="iso-8859-1"; Format="flowed"

The release of MPFI 1.5.1 is is imminent. Please help to make this
release as good as possible by downloading and testing this release
candidate:

https://gforge.inria.fr/frs/download.php/30117/mpfi-1.5.1-rc1.tar.bz2
https://gforge.inria.fr/frs/download.php/30116/mpfi-1.5.1-rc1.tar.gz
https://gforge.inria.fr/frs/download.php/30115/mpfi-1.5.1-rc1.zip

The MD5's:
ccff519222077407c3ca188a3e643c07  mpfi-1.5.1-rc1.tar.bz2
cd868c21d87afa05085d36d97dfde761  mpfi-1.5.1-rc1.tar.gz
5e8a29bfcd8cd8b3eb6d90ca86822763  mpfi-1.5.1-rc1.zip

The SHA1's:
5cacf98c882f217d8db937e12a4bc7104a5c322d  mpfi-1.5.1-rc1.tar.bz2
4f58755fbcc346ae23b04e9617e84c776e7b119f  mpfi-1.5.1-rc1.tar.gz
3e1d180c3640d9eae0a456efc4995cb1d6b99e03  mpfi-1.5.1-rc1.zip


Changes between MPFI version 1.5 and 1.5.1:
* fix bug in mpfi_sub_fr (reported by Takayuki YAMAGUCHI)
* now require at least GMP version 4.1.0 and MPFR version 2.4.2
* detect availability of mpfr_z_sub, mpfr_z_div, and mpfr_q_sub and use
   a custom implementation for the ones not provided by MPFR (mpfr_z_sub
   appeared in MPFR 3.1.0, the other ones are used in tests only)
* code cleanup (unused variables)
* tadd_q does no more check overflow by default (required too much
   memory)
* sign of zeros choice documented


Please send success and failure reports with "./config.guess" output
to <mpfi-users@lists.gforge.inria.fr>.

If no problems are found, MPFI 1.5.1 should be released around January
18, 2012.

Regards,
Philippe Theveny

_______________________________________________
Mpfi-users mailing list
Mpfi-users@lists.gforge.inria.fr
http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/mpfi-users

[fbissey]

Sorry I have fallen of the face of the internet in the past few days (not quite literally but almost). If there is a bit of time the test suite pass on power7 in a sage setting but with the following messages:

PASS: tinit
PASS: tinit_set
Warning: reverted endpoints line 21
Warning: reverted endpoints line 22
Warning: reverted endpoints line 23
Warning: reverted endpoints line 26
Warning: reverted endpoints line 30
Warning: reverted endpoints line 37
Warning: reverted endpoints line 38
PASS: tintersect

and

PASS: tsech
PASS: tset
Warning: reverted endpoints line 11
Warning: reverted endpoints line 12
Warning: reverted endpoints line 13
Warning: reverted endpoints line 14
Warning: reverted endpoints line 15
PASS: tset_d

Not sure if it can have an impact on some sage use or not.

[zimmerma]

Francois,

Not sure if it can have an impact on some sage use or not.

I reported the same warnings to the MPFI developers when 1.5.0 came out ( http://lists.gforge.inria.fr/pipermail/mpfi-users/2011-May/000023.html) but got no feedback. I will ask them again.

Paul

[jpflori]

FYI I get the same warnings, on the same kind of architecture.

And as far as the spkg is concerned, I'd prefer to wait until thursday when there is some sage meeting in Paris to package directly MPFI 1.5.1 on top of#12131. Unless someone has a very good reason for me to do it earlier with 1.5.1.rc1.

[jpflori]

Here comes finally a 1.5.1 spkg rebased on #12131

You can find it at  http://perso.telecom-paristech.fr/~flori/sage/mpfi-1.5.1.spkg

I slightly updated the description and license info as well.

[fbissey]

Shouldn't spkg-check actually be a proper shell script? By that I mean start with something like

#!/bin/sh

or possibly

#!/usr/bin/env bash

[jdemeyer]

fbissey: yes, it should be. It should also be executable.

[fbissey]

Replying to jdemeyer:

fbissey: yes, it should be. It should also be executable.

It is executable just not a proper shell script, putting it back to needs work. Everything else looks ok.

[jpflori]

Sorry about that, I definitely tried to repackage that one and mpfr one too quickly from scratch on my personal computer and mixed everything.

It should be ok now (same address as before).

[fbissey]

OK, I am putting it back to positive review, looks all right, survives basic testing.

[jdemeyer]

I actually get the doctest failure

sage -t  -force_lib devel/sage/sage/rings/real_mpfi.pyx
**********************************************************************
File "/mnt/usb1/scratch/jdemeyer/merger/sage-5.0.beta2/devel/sage-main/sage/rings/real_mpfi.pyx", line 4143:
    sage: 0 in (RealField(2000)(-19/3).gamma() - RealIntervalField(1000)(-19/3).gamma())
Expected:
    True
Got:
    False
**********************************************************************

[fbissey]

Replying to jdemeyer:

I actually get the doctest failure

sage -t  -force_lib devel/sage/sage/rings/real_mpfi.pyx
**********************************************************************
File "/mnt/usb1/scratch/jdemeyer/merger/sage-5.0.beta2/devel/sage-main/sage/rings/real_mpfi.pyx", line 4143:
    sage: 0 in (RealField(2000)(-19/3).gamma() - RealIntervalField(1000)(-19/3).gamma())
Expected:
    True
Got:
    False
**********************************************************************

Is it a 32 bit or 64 bit system? Or it doesn't matter? Was it 5.0_beta1?

[jdemeyer]

Sorry, should have mentioned this. It happened on sage.math, with sage-5.0.beta1 + various other patches + #12171. I have not tried a plain sage-5.0.beta1 + #12171.

[fbissey]

I haven't noticed any of that in the prealpha, interestingly enough the time out on power7 apparently happened on a call to .gamma(). I will try 5.0.beta1.

[fbissey]

On my 64 bit linux box after accounting for #12349 (which came out of nowhere suddenly) and applying this ticket

fbissey@QCD-nzi3 /home/work/fbissey/sandbox/sage-5.0.beta1 $ ./sage -t -long -force_lib devel/sage/sage/rings/real_mpfi.pyx
sage -t -long -force_lib "devel/sage/sage/rings/real_mpfi.pyx"
         [1.3 s]
 
----------------------------------------------------------------------
All tests passed!
Total time for all tests: 1.3 seconds

[zimmerma]

Replying to jdemeyer:

I actually get the doctest failure [...]

Jeroen, what do you get for the following?

sage: (RealField(2000)(-19/3).gamma() - RealIntervalField(1000)(-19/3).gamma())

My guess is that this problem is due to the one mentioned in comment 9, which is a bug already present in Sage, and thus separate from this ticket (however I don't know why the doctest worked before...)

Paul

[zimmerma]

the above problem can be simplified to:

sage: RealIntervalField(54)(-1) > RR(1)
True

which already happens in Sage 4.7.2, and seems to be due to the coercion model. I'll open a separate ticket for that (for precision between 2 and 53 we get the correct answer False).

Paul

[zimmerma]

I'll open a separate ticket for that

it is #12353.

Paul

[jdemeyer]

I don't understand what's happening here.

The test only seems to fail when running make ptest, not when running only the one test

jdemeyer@sage:/scratch/jdemeyer/merger/sage-5.0.beta2$ ./sage -t  -force_lib devel/sage/sage/rings/real_mpfi.pyx 2>&1 |cat
sage -t -force_lib "devel/sage/sage/rings/real_mpfi.pyx"
         [2.3 s]

----------------------------------------------------------------------
All tests passed!
Total time for all tests: 2.3 seconds

[zimmerma]

Replying to jdemeyer:

I don't understand what's happening here [...]

did you try the test case from #12353?

Paul

[jdemeyer]

I get

sage: RealIntervalField(53)(-1) > RR(1)
False
sage: RealIntervalField(54)(-1) > RR(1)
False
sage: (RealField(2000)(-19/3).gamma() - RealIntervalField(1000)(-19/3).gamma())
0.?e-302
sage: 0 in (RealField(2000)(-19/3).gamma() - RealIntervalField(1000)(-19/3).gamma())
True

[zimmerma]

strange, with 4.8 just compiled on a Core2 under Ubuntu 10.10 I get:

tarte% ./sage
----------------------------------------------------------------------
| Sage Version 4.8, Release Date: 2012-01-20                         |
| Type notebook() for the GUI, and license() for information.        |
----------------------------------------------------------------------
sage: RealIntervalField(53)(-1) > RR(1)
False
sage: RealIntervalField(54)(-1) > RR(1)
True

Paul

[fbissey]

I get (64 bit linux)

----------------------------------------------------------------------
| Sage Version 5.0.beta1, Release Date: 2012-01-22                   |
| Type notebook() for the GUI, and license() for information.        |
----------------------------------------------------------------------
**********************************************************************
*                                                                    *
* Warning: this is a prerelease version, and it may be unstable.     *
*                                                                    *
**********************************************************************
sage: RealIntervalField(53)(-1) > RR(1)
False
sage: RealIntervalField(54)(-1) > RR(1)
False
sage: (RealField(2000)(-19/3).gamma() - RealIntervalField(1000)(-19/3).gamma())
0.?e-302
sage: 0 in (RealField(2000)(-19/3).gamma() - RealIntervalField(1000)(-19/3).gamma())
True

On OS X 10.5.8 (32bit):

BlueFerniMac1:sage-5.0.beta1 frb15$ ./sage -t -long -force_lib devel/sage/sage/rings/real_mpfi.pyx
sage -t -long -force_lib "devel/sage/sage/rings/real_mpfi.pyx"
**********************************************************************
File "/Users/frb15/Desktop/sage-5.0.beta1/devel/sage/sage/rings/real_mpfi.pyx", line 4143:
    sage: 0 in (RealField(2000)(-19/3).gamma() - RealIntervalField(1000)(-19/3).gamma())
Expected:
    True
Got:
    False
**********************************************************************
1 items had failures:
   1 of  19 in __main__.example_117
***Test Failed*** 1 failures.
For whitespace errors, see the file /Users/frb15/.sage//tmp/real_mpfi_78068.py
	 [2.9 s]

and

BlueFerniMac1:sage-5.0.beta1 frb15$ ./sage
----------------------------------------------------------------------
| Sage Version 5.0.beta1, Release Date: 2012-01-22                   |
| Type notebook() for the GUI, and license() for information.        |
----------------------------------------------------------------------
**********************************************************************
*                                                                    *
* Warning: this is a prerelease version, and it may be unstable.     *
*                                                                    *
**********************************************************************
sage: RealIntervalField(53)(-1) > RR(1)
False
sage: RealIntervalField(54)(-1) > RR(1)
True
sage: (RealField(2000)(-19/3).gamma() - RealIntervalField(1000)(-19/3).gamma())
1.1?e-303
sage:  0 in (RealField(2000)(-19/3).gamma() - RealIntervalField(1000)(-19/3).gamma())
False

so there are differences of behavior between 32 and 64 bit naively.

[zimmerma]

Francois,

I just built Sage 5.0.beta1 on another 64 bit linux (Core2 under Ubuntu 10.10) and got:

tarte% ./sage
----------------------------------------------------------------------
| Sage Version 5.0.beta1, Release Date: 2012-01-22                   |
| Type notebook() for the GUI, and license() for information.        |
----------------------------------------------------------------------
**********************************************************************
*                                                                    *
* Warning: this is a prerelease version, and it may be unstable.     *
*                                                                    *
**********************************************************************
sage: RealIntervalField(53)(-1) > RR(1)
False
sage: RealIntervalField(54)(-1) > RR(1)
True

Paul

[fbissey]

This is way more subtle than I thought then. I did my tests without #11666, did you use #11666 as well as this one?

[zimmerma]

Francois,

no, I didn't use #11666. Can anybody reproduce the problem I have (cf #12353)? Anyway I believe this is independent from this ticket, since it already happened before (e.g., in Sage 4.7.2). Thus it should not prevent a positive review, if this is the only remaining issue, and we should solve #12353 independently.

Paul

[jdemeyer]

Everything I said was without #11666.

[fbissey]

Replying to zimmerma:

Francois,

no, I didn't use #11666. Can anybody reproduce the problem I have (cf #12353)? Anyway I believe this is independent from this ticket, since it already happened before (e.g., in Sage 4.7.2). Thus it should not prevent a positive review, if this is the only remaining issue, and we should solve #12353 independently.

I can reproduce it on OS X 10.5.8 (32bit) and linux on ppc64, but not on linux x86_64. You are probably right that's an orthogonal problem if it can happen as well in the current version of sage.

[zimmerma]

since this ticket got a positive review at comment 49, and the doctest failure at comment 50 is a separate issue (#12353) I put this back as "needs review".

Paul

[jdemeyer]

If nothing on this ticket changed, I guess we're back to positive review with a dependency on #12353.

[zimmerma]

Replying to jdemeyer:

If nothing on this ticket changed, I guess we're back to positive review with a dependency on #12353.

great! I'm working in #12353.

Paul

[zimmerma]

Jeroen, now that #12353 has been fixed, you can merge this one too.

Paul

[jdemeyer]

Replying to zimmerma:

Jeroen, now that #12353 has been fixed, you can merge this one too.

Don't worry, it's on my radar and should be merged in sage-5.0.beta5 if no further problems arise.

[jdemeyer]

In spkg-install, you should unset RM as old Sage versions set RM=rm which breaks libtool. For example, when upgrading from sage-4.5.3:

/bin/bash ../libtool --tag=CC   --mode=link gcc  -O2 -g -fPIC  -L/mnt/usb1/scratch/buildbot/sage/sage-1/sage_upgrade/build/sage-4.5.3-5.0.beta5/local/lib -L/mnt/usb1/scratch/buildbot/sage/sage-1/sage_upgrade/build/sage-4.5.3-5.0.beta5/local/lib  -o libmpfi.la -rpath /mnt/usb1/scratch/buildbot/sage/sage-1/sage_upgrade/build/sage-4.5.3-5.0.beta5/local/lib abs.lo acos.lo acosh.lo add.lo add_d.lo add_fr.lo add_q.lo add_si.lo add_ui.lo add_z.lo alea.lo asin.lo asinh.lo atan2.lo atan.lo atanh.lo bisect.lo blow.lo cbrt.lo clear.lo cmp.lo cmp_sym_pi.lo constants.lo cos.lo cosh.lo csc.lo csch.lo cot.lo coth.lo d_div.lo d_sub.lo diam.lo div.lo div_2exp.lo div_2si.lo div_2ui.lo div_d.lo div_fr.lo div_q.lo div_si.lo div_ui.lo div_z.lo error.lo exp.lo exp2.lo expm1.lo fr_div.lo fr_sub.lo get_endpoints.lo get_fr.lo get_d.lo get_prec.lo get_version.lo has_zero.lo hypot.lo increase.lo init.lo intersect.lo interv_d.lo interv_fr.lo interv_q.lo interv_si.lo interv_ui.lo interv_z.lo inp_str.lo inv.lo is_empty.lo is_inside.lo log.lo log10.lo log1p.lo log2.lo mag.lo mid.lo mig.lo mul.lo mul_2exp.lo mul_2si.lo mul_2ui.lo mul_d.lo mul_fr.lo mul_q.lo mul_si.lo mul_ui.lo mul_z.lo neg.lo out_str.lo predicates.lo print_binary.lo put.lo put_d.lo put_fr.lo put_q.lo put_si.lo put_ui.lo put_z.lo q_div.lo q_sub.lo quadrant.lo revert_if_needed.lo round_prec.lo sec.lo sech.lo set.lo set_d.lo set_fr.lo set_prec.lo set_q.lo set_si.lo set_str.lo set_ui.lo set_z.lo si_div.lo si_sub.lo sign.lo sin.lo sinh.lo sqr.lo sqrt.lo sub.lo sub_d.lo sub_fr.lo sub_q.lo sub_si.lo sub_ui.lo sub_z.lo swap.lo tan.lo tanh.lo ui_div.lo ui_sub.lo union.lo urandom.lo z_div.lo z_sub.lo  -lmpfr -lgmp
libtool: link: gcc -shared  -fPIC -DPIC  .libs/abs.o .libs/acos.o .libs/acosh.o .libs/add.o .libs/add_d.o .libs/add_fr.o .libs/add_q.o .libs/add_si.o .libs/add_ui.o .libs/add_z.o .libs/alea.o .libs/asin.o .libs/asinh.o .libs/atan2.o .libs/atan.o .libs/atanh.o .libs/bisect.o .libs/blow.o .libs/cbrt.o .libs/clear.o .libs/cmp.o .libs/cmp_sym_pi.o .libs/constants.o .libs/cos.o .libs/cosh.o .libs/csc.o .libs/csch.o .libs/cot.o .libs/coth.o .libs/d_div.o .libs/d_sub.o .libs/diam.o .libs/div.o .libs/div_2exp.o .libs/div_2si.o .libs/div_2ui.o .libs/div_d.o .libs/div_fr.o .libs/div_q.o .libs/div_si.o .libs/div_ui.o .libs/div_z.o .libs/error.o .libs/exp.o .libs/exp2.o .libs/expm1.o .libs/fr_div.o .libs/fr_sub.o .libs/get_endpoints.o .libs/get_fr.o .libs/get_d.o .libs/get_prec.o .libs/get_version.o .libs/has_zero.o .libs/hypot.o .libs/increase.o .libs/init.o .libs/intersect.o .libs/interv_d.o .libs/interv_fr.o .libs/interv_q.o .libs/interv_si.o .libs/interv_ui.o .libs/interv_z.o .libs/inp_str.o .libs/inv.o .libs/is_empty.o .libs/is_inside.o .libs/log.o .libs/log10.o .libs/log1p.o .libs/log2.o .libs/mag.o .libs/mid.o .libs/mig.o .libs/mul.o .libs/mul_2exp.o .libs/mul_2si.o .libs/mul_2ui.o .libs/mul_d.o .libs/mul_fr.o .libs/mul_q.o .libs/mul_si.o .libs/mul_ui.o .libs/mul_z.o .libs/neg.o .libs/out_str.o .libs/predicates.o .libs/print_binary.o .libs/put.o .libs/put_d.o .libs/put_fr.o .libs/put_q.o .libs/put_si.o .libs/put_ui.o .libs/put_z.o .libs/q_div.o .libs/q_sub.o .libs/quadrant.o .libs/revert_if_needed.o .libs/round_prec.o .libs/sec.o .libs/sech.o .libs/set.o .libs/set_d.o .libs/set_fr.o .libs/set_prec.o .libs/set_q.o .libs/set_si.o .libs/set_str.o .libs/set_ui.o .libs/set_z.o .libs/si_div.o .libs/si_sub.o .libs/sign.o .libs/sin.o .libs/sinh.o .libs/sqr.o .libs/sqrt.o .libs/sub.o .libs/sub_d.o .libs/sub_fr.o .libs/sub_q.o .libs/sub_si.o .libs/sub_ui.o .libs/sub_z.o .libs/swap.o .libs/tan.o .libs/tanh.o .libs/ui_div.o .libs/ui_sub.o .libs/union.o .libs/urandom.o .libs/z_div.o .libs/z_sub.o   -Wl,-rpath -Wl,/mnt/usb1/scratch/buildbot/sage/sage-1/sage_upgrade/build/sage-4.5.3-5.0.beta5/local/lib -Wl,-rpath -Wl,/mnt/usb1/scratch/buildbot/sage/sage-1/sage_upgrade/build/sage-4.5.3-5.0.beta5/local/lib -L/mnt/usb1/scratch/buildbot/sage/sage-1/sage_upgrade/build/sage-4.5.3-5.0.beta5/local/lib /mnt/usb1/scratch/buildbot/sage/sage-1/sage_upgrade/build/sage-4.5.3-5.0.beta5/local/lib/libmpfr.so /mnt/usb1/scratch/buildbot/sage/sage-1/sage_upgrade/build/sage-4.5.3-5.0.beta5/local/lib/libgmp.so  -O2   -Wl,-soname -Wl,libmpfi.so.0 -o .libs/libmpfi.so.0.0.0
libtool: link: (cd ".libs" && rm "libmpfi.so.0" && ln -s "libmpfi.so.0.0.0" "libmpfi.so.0")
rm: cannot remove `libmpfi.so.0': No such file or directory
make[2]: *** [libmpfi.la] Error 1

[jpflori]

An updated package is available at the same address.

I also added a mecurial tag for mpfi-1.5.1.

[vbraun]

Looks good!