Upgrade giac to 1.9.0-15 #31563

mkoeppe · 2021-03-26T01:22:41Z

https://www-fourier.ujf-grenoble.fr/~parisse/debian/dists/stable/main/source/

Previous update: #30537

Depends on #33226
Depends on #34037
Depends on #31403
Depends on #34088

CC: @orlitzky @kiwifb @antonio-rojas @frederichan-IMJPRG @sagetrac-parisse @sagetrac-tmonteil @dimpase

Component: packages: standard

Author: Matthias Koeppe

Branch: b4e5c6e

Reviewer: François Bissey

Issue created by migration from https://trac.sagemath.org/ticket/31563

The text was updated successfully, but these errors were encountered:

antonio-rojas · 2021-03-26T08:34:09Z

comment:1

FTR, this is straightforward and requires no changes to sagelib.

orlitzky · 2021-04-02T13:35:55Z

comment:4

I've updated the system version check in #31594 because it should be a trivial review.

mkoeppe · 2021-09-02T19:18:37Z

Changed dependencies from #31562 to #32449

mkoeppe · 2021-09-02T19:34:32Z

Branch: u/mkoeppe/upgrade_giac_to_1_7_0

mkoeppe · 2021-09-02T19:36:38Z

Attachment: giac-1.7.0.25p0.tar.bz2.gz

mkoeppe · 2021-09-02T19:42:58Z

Commit: 4df014a

mkoeppe · 2021-09-02T19:42:58Z

comment:8

I have made a new tarball using spkg-src. Patches in build/pkgs/giac/patches/ still need updating/review, help welcome

New commits:

`7d6a7b8`	`Make test pass with giac 1.7`
`ffa2679`	`Merge #32449`
`1127597`	`build/pkgs/giac/patches/autotools/0001-configure.ac-Do-not-link-to-libintl-if-USE_NLS-is-no.patch: Remove, not needed for giac 1.7`
`35111c9`	`build/pkgs/giac: Update to 1.7.0.25`
`4df014a`	`build/pkgs/giac/checksums.ini: Update upstream_url`

mkoeppe · 2021-09-03T19:03:03Z

comment:9

These are the current patches:
https://github.com/sagemath/sage-prod/tree/develop/build/pkgs/giac/patches
(I have already taken care of the ones in autotools/)

kiwifb · 2021-09-26T23:27:02Z

comment:10

I just tested the latest giac (1.7.0.33) against sage 9.5.beta2. The only doctests failing are

sage -t --long --random-seed=0 /usr/lib/python3.9/site-packages/sage/functions/min_max.py
too many failed tests, not using stored timings
Running doctests with ID 2021-09-27-12-21-06-b85d81e2.
Using --optional=dochtml,pip,sage
Doctesting 1 file.
sage -t --long --random-seed=0 /usr/lib/python3.9/site-packages/sage/functions/min_max.py
**********************************************************************
File "/usr/lib/python3.9/site-packages/sage/functions/min_max.py", line 236, in sage.functions.min_max.MaxSymbolic._evalf_
Failed example:
    r
Expected:
    sqrt(2) - cos(1)
Got:
    0.873911256505000
**********************************************************************
File "/usr/lib/python3.9/site-packages/sage/functions/min_max.py", line 238, in sage.functions.min_max.MaxSymbolic._evalf_
Failed example:
    r.n()
Expected:
    0.873911256504955
Got:
    0.873911256505000
**********************************************************************
1 item had failures:
   2 of  11 in sage.functions.min_max.MaxSymbolic._evalf_
    [69 tests, 2 failures, 0.53 s]

I'll note that recent giac unconditionally look for usb connecting libraries. It seems to be related to giac being able to go in embedded devices as far as I can tell. No real impact on the maths but the functionality is not clearly or cleanly separated.

kiwifb · 2021-10-18T22:08:56Z

comment:12

1.7.0.39 out. Did anyone notice that it was looking for curl before or is it recent?

dnl AC_CHECK_LIB(curl, main)
CONFIG_CURL="yes"
AC_ARG_ENABLE([curl],
	[AS_HELP_STRING([--enable-curl], [Use CURL [[default=yes]]])],
	[if test "$enableval" = "no"; then CONFIG_CURL="no"; fi], [])

if test "$CONFIG_CURL" = "yes"; then
	AC_CHECK_HEADER(curl/curl.h, [], [CONFIG_CURL="no"])
  fi
if test "$CONFIG_CURL" = "yes"; then
	AC_CHECK_LIB(curl, main, [], [CONFIG_CURL="no"])
  fi
AC_SUBST(CONFIG_CURL)
AC_SUBST(CURL_LIBS)
AM_CONDITIONAL(CONFIG_CURL, [test "$CONFIG_CURL" = "yes"])

It may have been there some time because it is an alternative to a code branch that can never be reached (pragma constant EMCC and EMCC2 are never defined anywhere).

kiwifb · 2021-10-18T22:10:11Z

comment:13

curl is in my dependency list in Gentoo, so I just didn't look properly the first _couple_ of times.

sagetrac-parisse · 2021-10-19T13:01:30Z

comment:14

curl is used for the read command, if the argument is a string beginning with http.
For example read("https://www-fourier.univ-grenoble-alpes.fr/~parisse/numworks/py.py")

kiwifb · 2021-10-19T20:29:11Z

comment:15

Replying to @sagetrac-parisse:

curl is used for the read command, if the argument is a string beginning with http.
For example read("https://www-fourier.univ-grenoble-alpes.fr/~parisse/numworks/py.py")

That's certainly an interesting option. What about all that EMCC(2) code for emscript all around it that has no way of being enabled. Is it dead code or something in gestation?

kiwifb · 2021-10-19T21:04:18Z

comment:16

Replying to @kiwifb:

I just tested the latest giac (1.7.0.33) against sage 9.5.beta2. The only doctests failing are

sage -t --long --random-seed=0 /usr/lib/python3.9/site-packages/sage/functions/min_max.py
too many failed tests, not using stored timings
Running doctests with ID 2021-09-27-12-21-06-b85d81e2.
Using --optional=dochtml,pip,sage
Doctesting 1 file.
sage -t --long --random-seed=0 /usr/lib/python3.9/site-packages/sage/functions/min_max.py
**********************************************************************
File "/usr/lib/python3.9/site-packages/sage/functions/min_max.py", line 236, in sage.functions.min_max.MaxSymbolic._evalf_
Failed example:
    r
Expected:
    sqrt(2) - cos(1)
Got:
    0.873911256505000
**********************************************************************
File "/usr/lib/python3.9/site-packages/sage/functions/min_max.py", line 238, in sage.functions.min_max.MaxSymbolic._evalf_
Failed example:
    r.n()
Expected:
    0.873911256504955
Got:
    0.873911256505000
**********************************************************************
1 item had failures:
   2 of  11 in sage.functions.min_max.MaxSymbolic._evalf_
    [69 tests, 2 failures, 0.53 s]

This is really bizarre. min_max uses sympy not giac - at least directly. The only thing I can think of, is that giac is used to perform the integral and the result is numerical instead of symbolic. See the context of the failing tests

            sage: f = max_symbolic(sin(x), cos(x))
            sage: r = integral(f, x, 0, 1)
            ...
            sage: r
            sqrt(2) - cos(1)
            sage: r.n()
            0.873911256504955

sagetrac-parisse · 2021-10-20T06:09:19Z

comment:17

Replying to @kiwifb:

Replying to @sagetrac-parisse:

curl is used for the read command, if the argument is a string beginning with http.
For example read("https://www-fourier.univ-grenoble-alpes.fr/~parisse/numworks/py.py")

That's certainly an interesting option. What about all that EMCC(2) code for emscript all around it that has no way of being enabled. Is it dead code or something in gestation?

It's not dead, it is enabled by commandline -D options, or by the compiler itself (emcc).
Try this [url=https://www-fourier.univ-grenoble-alpes.fr/%7eparisse/xcasen.html#exec&filename=%40session&python=1&radian=1&cas=0,0,read(%22https%3A%2F%2Fwww-fourier.univ-grenoble-alpes.fr%2F~parisse%2Fnumworks%2Fpy.py%22)&cas=0,200,go(f1)&cas=0,400,%3B&]session Xcas[/url]

tornaria · 2021-11-25T23:16:09Z

comment:18

The test failure above seems to be a regression in giac 1.7.0:

giac 1.6.0 (as compiled by sage)

0>> int(max(sin(x),cos(x)), x, 0, 1)
Warning, integration of abs or sign assumes constant sign by intervals (correct if the argument is real):
Check [abs(sin(x)-cos(x))]
Discontinuities at zeroes of sin(x)-cos(x) were not checked
sqrt(2)-cos(1)
// Time 0.11

giac 1.7.0 (void linux)

0>> int(max(sin(x),cos(x)), x, 0, 1)
Piecewise definite integration: can only handle linear < or > condition
0.873911256505
// Time 0.01

Note that sympy_integrator is able to correctly evaluate this integral, but the order is maxima, then giac, then sympy.

sage: f = max_symbolic(sin(x), cos(x))
sage: sage.symbolic.integration.external.sympy_integrator(f, x, 0, 1)
sqrt(2) - cos(1)

antonio-rojas · 2021-11-26T07:02:02Z

comment:19

Replying to @tornaria:

The test failure above seems to be a regression in giac 1.7.0:

Yes, more precisely it's a regression in 1.7.0-29. Since their forum is a gated community, it would be nice if someone with an account could report it.

sagetrac-parisse · 2021-11-26T08:16:47Z

comment:20

This is a consequence of the fact that certified definite integration is hard. For this example Xcas session
there is only one point of discontinuity and it can easily be found by solving sin(x)=cos(x) on x=0..1, but for a large interval (say x=0..10000) or for a more complicated equation it's not the case. In order to make sure that I do not return a false answer, I have decided to return an exact answer only if I'm certain I did not miss a discontinuity in the antiderivative. It's certainly possible to improve, but not easy...

tornaria · 2021-11-26T14:16:13Z

comment:21

OTOH, this still works on 1.7.0-39

0>> int((sin(x)+cos(x)+abs(cos(x)-sin(x)))/2, x, 0, 1)
Warning, integration of abs or sign assumes constant sign by intervals (correct if the argument is real):
Check [abs(cos(x)-sin(x))]
Discontinuities at zeroes of cos(x)-sin(x) were not checked
sqrt(2)-cos(1)
// Time 0.14

tornaria · 2021-11-26T14:20:24Z

comment:22

Note that the warnings get through into sage

sage: integral((cos(x)+sin(x)+abs(cos(x)-sin(x)))/2, x, 0, 1)
Warning, integration of abs or sign assumes constant sign by intervals (correct if the argument is real):
Check [abs(cos(sageVARx)-sin(sageVARx))]
Discontinuities at zeroes of cos(sageVARx)-sin(sageVARx) were not checked
sqrt(2) - cos(1)

Would it be possible to integrate max(A,B) with a similar warning?

tornaria · 2021-11-26T14:28:55Z

comment:23

BTW, did you notice that sage is reading giac output with less than 53-bit precision? We are getting 0.873911256505000 instead of 0.873911256504955, not because giac answer is incorrect but because its output is rounded to 0.873911256505 (which is correct rounding but less than 53-bit).

Is there a simple way to setup giac with more output precision so that sage gets the right answer correctly rounded?

sagetrac-parisse · 2021-11-26T19:37:29Z

comment:24

max is translated internally to piecewise, it does not use the formula with max, and the checks differ.
If you run Digits:=14, you will get 14 digits printed. You can not get 53 bits of mantissa without multiprecision floats in giac, because the generic type of giac stores double in 8 bytes, and 5 bits are used to store the type -> the mantissa is rounded to 0 with 48 bits precision.

tornaria · 2021-12-13T17:20:45Z

comment:25

Since there doesn't seem to be a fix available, I'm patching out the two failing doctests as follows:

--- a/src/sage/functions/min_max.py
+++ b/src/sage/functions/min_max.py
@@ -233,9 +233,9 @@ class MaxSymbolic(MinMax_base):
             sage: f = max_symbolic(sin(x), cos(x))
             sage: r = integral(f, x, 0, 1)
             ...
-            sage: r
+            sage: r             # not tested -- broken with giac 1.7.0
             sqrt(2) - cos(1)
-            sage: r.n()
+            sage: r.n()         # not tested -- broken with giac 1.7.0
             0.873911256504955
         """
         return max_symbolic(args)

I don't know if this is acceptable for including it in sagemath, but I can't downgrade giac and I rather skip this failing doctests. Is there a way to label a doctest as "expected fail" instead of "not tested" ?

orlitzky · 2021-12-13T18:06:02Z

comment:26

Ironically, I think this test was added to showcase the fact that, given a difficult integral, sage would iterate through its available backends until it got an answer it liked. It seems now that giac is evolving towards the status quo, and refusing to give a possibly-incorrect exact answer.

We should still test the fallback behavior somehow, but these integrals have given us headaches for decades. I think we should stop promising to integrate them exactly, i.e. change/delete the tests.

nasser1 · 2022-07-09T09:06:56Z

comment:93

Replying to @mkoeppe:

on macOS:

[giac-1.9.0.15p0] /Library/Developer/CommandLineTools/SDKs/MacOSX.sdk/usr/include/sys/cdefs.h:208:48: note: expanded from macro '__deprecated_msg'
[giac-1.9.0.15p0]         #define __deprecated_msg(_msg) __attribute__((__deprecated__(_msg)))
[giac-1.9.0.15p0]                                                       ^
[giac-1.9.0.15p0] icas.cc:1435:9: error: no member named 'localisation' in namespace 'xcas'
[giac-1.9.0.15p0]   xcas::localisation(); // change by L. Marohnić
[giac-1.9.0.15p0]   ~~~~~~^

FYI
I've reported this error xcas::localisation(); few days ago to giac forum. Here is the link

https://xcas.univ-grenoble-alpes.fr/forum/viewtopic.php?f=3&t=2789

--Nasser

kiwifb · 2022-07-09T09:07:39Z

comment:94

I didn't spot that one before because I always build with fltk support. And that part is in a "#elseblock when fltk is not requested. Blocks withxcas::localisation` have very funny guards

#ifndef USE_OBJET_BIDON
xcas::localisation(); // change by L. Marohnić
#endif

literally "use fake object". Passing -DUSE_OBJET_BIDON may help.

kiwifb · 2022-07-09T09:22:06Z

comment:95

Yes, I compiled it without fltk by adding -DUSE_OBJET_BIDON=1 to CPPFLAGS. Be careful "OBJET" without a "C", this is french and google translate will tell you that "bidon" is a "can", but colloquially it means "something that isn't what it seems", the can may look big but inside it is just empty.

sagetrac-git · 2022-07-09T16:52:40Z

Branch pushed to git repo; I updated commit sha1. New commits:

`d2ade90`	`build/pkgs/giac/spkg-install.in: Work around xcas::localisation error`

sagetrac-git · 2022-07-09T16:52:40Z

Changed commit from 4160a64 to d2ade90

mkoeppe · 2022-07-09T16:53:25Z

comment:97

Thanks both, this builds now

mkoeppe · 2022-07-09T16:54:54Z

Changed dependencies from #33226, #34037, #31403 to #33226, #34037, #31403, #34088

sagetrac-git · 2022-07-09T16:55:03Z

Branch pushed to git repo; I updated commit sha1. New commits:

`d835afc`	`build/pkgs/python3/distros/cygwin.txt: Add packages needed for ensurepip`
`d11f705`	`Merge #34088`

sagetrac-git · 2022-07-09T16:55:03Z

Changed commit from d2ade90 to d11f705

mkoeppe · 2022-07-09T16:58:01Z

Changed reviewer from https://github.com/sagemath/sagetrac-mirror/actions/runs/2178226151 to https://github.com/mkoeppe/sage/actions/runs/2641761864, https://github.com/mkoeppe/sage/actions/runs/2641761870

sagetrac-git · 2022-07-11T04:01:39Z

Changed commit from d11f705 to b4e5c6e

sagetrac-git · 2022-07-11T04:01:39Z

Branch pushed to git repo; I updated commit sha1. New commits:

`b4e5c6e`	`Merge tag '9.7.beta5' into t/31563/upgrade_giac_to_1_7_0`

mkoeppe · 2022-07-11T15:54:13Z

comment:103

Still fails on cygwin with same error as shown in comment:55 - https://github.com/mkoeppe/sage/runs/7274399777?check_suite_focus=true

mkoeppe · 2022-07-11T21:00:41Z

comment:104

The other platforms looks OK.
I'd suggest that we ignore Cygwin for now and merge the ticket.

kiwifb · 2022-07-11T21:05:37Z

comment:105

Cygwin is low on my own priority list but it feels a bit of a shame. If giac 1.7 builds on cygwin this is clearly a regression, but while we upgrade the package here, there is nothing that prevent the use of 1.7 or specifically require 1.9, so if someone wants to build on cygwin, they can bring their own giac 1.7.

mkoeppe · 2022-07-11T21:08:51Z

comment:106

I think it will be easy to fix the Cygwin build failure in a follow up ticket. Best done in one go together with other current failures of the platform.

kiwifb · 2022-07-11T21:11:01Z

comment:107

I am OK with that plan. I am OK being a positive reviewer for this.

mkoeppe · 2022-07-11T23:38:34Z

comment:108

Thanks!

kiwifb · 2022-07-12T00:38:54Z

Changed reviewer from https://github.com/mkoeppe/sage/actions/runs/2641761864, https://github.com/mkoeppe/sage/actions/runs/2641761870 to François Bissey

vbraun · 2022-07-21T21:12:21Z

Changed branch from u/mkoeppe/upgrade_giac_to_1_7_0 to b4e5c6e

mkoeppe · 2022-08-03T16:58:49Z

Changed commit from b4e5c6e to none

mkoeppe · 2022-08-03T16:58:49Z

comment:111

Replying to @mkoeppe:

I think it will be easy to fix the Cygwin build failure in a follow up ticket. Best done in one go together with other current failures of the platform.

This is now #34269.

mkoeppe added this to the sage-9.4 milestone Mar 26, 2021

mkoeppe added c: packages: standard labels Mar 26, 2021

mkoeppe modified the milestones: sage-9.4, sage-9.5 Jul 19, 2021

mkoeppe added s: needs review and removed s: needs work labels Jul 9, 2022

kiwifb added s: positive review and removed s: needs review labels Jul 12, 2022

vbraun removed the s: positive review label Jul 21, 2022

vbraun closed this as completed in 1c98f0a Jul 21, 2022

videlec mentioned this issue Jan 7, 2023

make libgiac quiet #31655

Open

sagetrac-tmonteil mannequin mentioned this issue Feb 6, 2022

giac 1.6.0.47p3 does not pass self-tests #32588

Closed

Upgrade giac to 1.9.0-15 #31563

Upgrade giac to 1.9.0-15 #31563

Comments

mkoeppe commented Mar 26, 2021

antonio-rojas commented Mar 26, 2021

orlitzky commented Apr 2, 2021

mkoeppe commented Sep 2, 2021

mkoeppe commented Sep 2, 2021

mkoeppe commented Sep 2, 2021

mkoeppe commented Sep 2, 2021

mkoeppe commented Sep 2, 2021

mkoeppe commented Sep 3, 2021

kiwifb commented Sep 26, 2021

kiwifb commented Oct 18, 2021

kiwifb commented Oct 18, 2021

sagetrac-parisse mannequin commented Oct 19, 2021

kiwifb commented Oct 19, 2021

kiwifb commented Oct 19, 2021

sagetrac-parisse mannequin commented Oct 20, 2021

tornaria commented Nov 25, 2021

antonio-rojas commented Nov 26, 2021

sagetrac-parisse mannequin commented Nov 26, 2021

tornaria commented Nov 26, 2021

tornaria commented Nov 26, 2021

tornaria commented Nov 26, 2021

sagetrac-parisse mannequin commented Nov 26, 2021

tornaria commented Dec 13, 2021

orlitzky commented Dec 13, 2021

nasser1 commented Jul 9, 2022

kiwifb commented Jul 9, 2022

kiwifb commented Jul 9, 2022

sagetrac-git mannequin commented Jul 9, 2022

sagetrac-git mannequin commented Jul 9, 2022

mkoeppe commented Jul 9, 2022

mkoeppe commented Jul 9, 2022

sagetrac-git mannequin commented Jul 9, 2022

sagetrac-git mannequin commented Jul 9, 2022

mkoeppe commented Jul 9, 2022

sagetrac-git mannequin commented Jul 11, 2022

sagetrac-git mannequin commented Jul 11, 2022

mkoeppe commented Jul 11, 2022

mkoeppe commented Jul 11, 2022

kiwifb commented Jul 11, 2022

mkoeppe commented Jul 11, 2022

kiwifb commented Jul 11, 2022

mkoeppe commented Jul 11, 2022

kiwifb commented Jul 12, 2022

vbraun commented Jul 21, 2022

mkoeppe commented Aug 3, 2022

mkoeppe commented Aug 3, 2022