search

sagemath / sage

Main repository of SageMath
https://www.sagemath.org
Other
1.41k stars 475 forks source link

Upgrade to lcalc-2.0.4 #32037

Closed orlitzky closed 2 years ago

orlitzky commented 3 years ago

The new lcalc 2.0.3 has been released. Note in particular:

In this ticket we upgrade to lcalc 2.0.3 and:

Depends on #32061 Depends on #31837

CC: @JohnCremona @antonio-rojas @kiwifb @dimpase

Component: packages: standard

Author: Michael Orlitzky, Frédéric Chapoton

Branch: cbb7908

Reviewer: Antonio Rojas, Dima Pasechnik

Issue created by migration from https://trac.sagemath.org/ticket/32037

mkoeppe commented 3 years ago
comment:1

Uh, a branch please?

dimpase commented 3 years ago
comment:2

John, L-functions question here...

fchapoton commented 3 years ago
comment:3

For the first doctest, the PARI implementation (only available for real input) gives a different answer at 0:

sage: chi = DirichletGroup(5)[2]                                                
sage: chi.lfunction().hardy(0)                                                  
0.793597201093544

EDIT

but Pari doc of "lfunhardy" says this gives a "Variant of the Hardy Z-function"

fchapoton commented 3 years ago
comment:5

For the second doctest, the values are indeed zeroes, according to pari :

sage: chi = DirichletGroup(5)[1]                                               
sage: f = chi.lfunction() 
sage: f(1/2+6.18357819545086*I)                                                 
2.61452836816118e-15 + 6.48370608217824e-15*I
sage: f(1/2+8.45722917442320*I)                                                 
3.51419504210656e-14 - 3.69407271406328e-14*I
sage: f(1/2-4.13290370521286*I)       
-4.69010768657543e-15 - 8.53389710587884e-15*I

EDIT

But that doc says that "L.find_zeros_via_N(3)" is supposed to find 3 zeroes, ignoring negative ones !

orlitzky commented 3 years ago
comment:6

Replying to @mkoeppe:

Uh, a branch please?

I hacked together the minimal changes needed to get this to compile (bonus: it doesn't crash), but all of the patches in the description should be viewed with great suspicion. It will take some more time to sort out the right things to do here.

orlitzky commented 3 years ago
comment:7

Replying to @fchapoton:

EDIT

But that doc says that "L.find_zeros_via_N(3)" is supposed to find 3 zeroes, ignoring negative ones !

There used to be a parameter called do_negative that made it look for zeros with a negative imaginary part, but the new libLfunction computes do_negative on its own:

bool do_negative=is_complex();
...
template <class ttype>
bool L_function <ttype>::
is_complex(){

    if(my_verbose>2) cout << "#            start is_complex()" << endl;

    bool do_negative=false;

    int n=0;
    if(what_type_L!=-1) //if not the zeta function, check if self dual. If so, do_negative stays false                                                     
    do{
        n++;
        if(abs(imag(Complex(dirichlet_coefficient[n])))>.00001){
            do_negative=true;
        if(my_verbose>2) cout << "#            is_complex: Dirichlet co\
eff " << n << " " << setprecision(DIGITS) << dirichlet_coefficient[n] << endl;
        }
    }while(n<number_of_dirichlet_coefficients&&!do_negative);
    if(my_verbose>2) cout << "#            is_complex() = " << do_negative<\
< endl;

    return do_negative;

}

So maybe it's just the documentation that needs to be updated. I don't know what self-dual means, or what it's ultimately checking to make that decision in this case.

DaveWitteMorris commented 3 years ago
comment:8

FWIW, here is my reading of this code: If all of the coefficients of the Dirichlet series are real (up to a tolerance of .00001), then the conjugate of any root is also a root. (This is what "self-dual" must mean.) We do not need to look for roots below the real axis in this case (because they are conjugates of roots that are above the real axis), so do_negative is set to False. Otherwise, do_negative is set to True.

Replying to @orlitzky:

Replying to @fchapoton: ... So maybe it's just the documentation that needs to be updated. I don't know what self-dual means, or what it's ultimately checking to make that decision in this case.

fchapoton commented 3 years ago

Branch: public/lcalc_version_2

fchapoton commented 3 years ago
comment:9

here is a first branch, just changing version, checksums and lcalc_sage.h

New commits:

61be05dfirst steps towards lcalc 2.0.0
fchapoton commented 3 years ago

Commit: 61be05d

fchapoton commented 3 years ago
comment:10

should we remove all the old patches ?

orlitzky commented 3 years ago
comment:11

Replying to @fchapoton:

should we remove all the old patches ?

Yep, all patches are obsolete. The spkg-install.in should now be,

sdh_configure --with-pari
sdh_make_install

and we can add an spkg-check.in containing,

sdh_make_check
7ed8c4ca-6d56-4ae9-953a-41e42b4ed313 commented 3 years ago

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

7f68936lcalc 2.0.0 removing all patches
7ed8c4ca-6d56-4ae9-953a-41e42b4ed313 commented 3 years ago

Changed commit from 61be05d to 7f68936

7ed8c4ca-6d56-4ae9-953a-41e42b4ed313 commented 3 years ago

Changed commit from 7f68936 to ef80bc3

7ed8c4ca-6d56-4ae9-953a-41e42b4ed313 commented 3 years ago

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

ef80bc3update installation scripts
7ed8c4ca-6d56-4ae9-953a-41e42b4ed313 commented 3 years ago

Changed commit from ef80bc3 to 76226d7

7ed8c4ca-6d56-4ae9-953a-41e42b4ed313 commented 3 years ago

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

76226d7lcalc 2.0.0 no dependency to mpfr
fchapoton commented 3 years ago
comment:15

should we "cd src" in spkg-install ?

I have not managed to install so far with the current branch, "sage -i lcalc" complains as follows

[lcalc-2.0.0] Uninstalling 'lcalc' with legacy uninstaller
[lcalc-2.0.0] Package 'lcalc' is currently not installed
[lcalc-2.0.0] Now building lcalc, example programs and the shared library...
[lcalc-2.0.0] Building lcalc-2.0.0
[lcalc-2.0.0] make[3]: *** No targets specified and no makefile found.  Stop.
DaveWitteMorris commented 3 years ago
comment:16

I don't understand anything about the build system, so it's likely that this is noise (in which case I apologize), but dimpase says not to use "sage -i":

we are trying to switch to a sane building scheme:

./bootstrap
./configure
make build
make pynormaliz

sage -i is an old hack noone should use...

So maybe you are supposed to try "make lcalc" ?

mkoeppe commented 3 years ago
comment:17

Replying to @fchapoton:

should we "cd src" in spkg-install ?

yes

7ed8c4ca-6d56-4ae9-953a-41e42b4ed313 commented 3 years ago

Changed commit from 76226d7 to aa2bb03

7ed8c4ca-6d56-4ae9-953a-41e42b4ed313 commented 3 years ago

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

aa2bb03Trac #32037: update spkg-configure.m4 for lcalc-2.0.0.
slel commented 3 years ago
comment:19

Should .m4 files mix tabs and spaces?

7ed8c4ca-6d56-4ae9-953a-41e42b4ed313 commented 3 years ago

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

849de4bTrac #32037: fixup: kill tabs in build/pkgs/lcalc/spkg-configure.m4.
7ed8c4ca-6d56-4ae9-953a-41e42b4ed313 commented 3 years ago

Changed commit from aa2bb03 to 849de4b

orlitzky commented 3 years ago
comment:21

Replying to @slel:

Should .m4 files mix tabs and spaces?

According to me, no. According to emacs, at every opportunity.

fchapoton commented 3 years ago
comment:22

Using "make lcalc", I am still not able to build, with or without "cd src".

Complains of missing makefile, as in comment:17

orlitzky commented 3 years ago
comment:23

Replying to @fchapoton:

Using "make lcalc", I am still not able to build, with or without "cd src".

Complains of missing makefile, as in comment:17

I think you have to get rid of spkg-build.in, too.

7ed8c4ca-6d56-4ae9-953a-41e42b4ed313 commented 3 years ago

Changed commit from 849de4b to 7327388

7ed8c4ca-6d56-4ae9-953a-41e42b4ed313 commented 3 years ago

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

7327388lcalc 2.0.0 towards a working make
fchapoton commented 3 years ago
comment:25

ok, now "make lcalc" starts doing something and stops at

[lcalc-2.0.0] Thread model: posix
[lcalc-2.0.0] gcc version 9.3.0 (Ubuntu 9.3.0-17ubuntu1~20.04) 
...
[lcalc-2.0.0] checking for gengetopt... no
[lcalc-2.0.0] configure: error: GNU gengetopt is required
orlitzky commented 3 years ago

Description changed:

--- 
+++ 
@@ -2,6 +2,7 @@

 There are a few issues that need to be addressed during the upgrade.

+0. The build requires GNU gengetopt, so we'll need a new gengetopt SPKG for lcalc to depend on. (It doesn't need to go in the DEPCHECK macro though, since it's a build-only dependency.)
 1. The build system is now sane. This should make life easier in the long run, but means that spkg-* need to be rewritten. `./configure --with-pari` should suffice.

 2. The lcalc spkg shouldn't depend on mpfr (I'm not sure why it ever did).
orlitzky commented 3 years ago
comment:26

Well the good news is that it's supposed to do that, but the bad news is that we'll have to add another gengetopt SPKG for lcalc to depend on.

orlitzky commented 3 years ago

Dependencies: #32061

orlitzky commented 3 years ago
comment:27

I posted a branch for review on #32061 to add the missing dependency.

orlitzky commented 3 years ago

Changed commit from 7327388 to 3b44220

orlitzky commented 3 years ago

Changed branch from public/lcalc_version_2 to u/mjo/ticket/32037

orlitzky commented 3 years ago
comment:28

I figured out how to fix the build scripts... and then found a typo in the upstream configure.ac that I had to release a 2.0.1 to fix. I think this will actually work now.

orlitzky commented 3 years ago
comment:29

Replying to @DaveWitteMorris:

FWIW, here is my reading of this code: If all of the coefficients of the Dirichlet series are real (up to a tolerance of .00001), then the conjugate of any root is also a root. (This is what "self-dual" must mean.) We do not need to look for roots below the real axis in this case (because they are conjugates of roots that are above the real axis), so do_negative is set to False. Otherwise, do_negative is set to True.

These functions are supposed to "return the first n zeros." That's vague, but scrutable if you're starting from the real axis and moving up: basically, the numbers (imaginary parts) it outputs get bigger. I guess when do_negative is true, we output the first n zeros in magnitude?

fchapoton commented 3 years ago

Description changed:

--- 
+++ 
@@ -1,4 +1,4 @@
-I've just tagged a 2.0.0 release of lcalc: https://gitlab.com/sagemath/lcalc/-/tags/2.0.0
+I've just tagged a 2.0.1 release of lcalc: https://gitlab.com/sagemath/lcalc/-/tags/2.0.1

 There are a few issues that need to be addressed during the upgrade.
7ed8c4ca-6d56-4ae9-953a-41e42b4ed313 commented 3 years ago

Branch pushed to git repo; I updated commit sha1. This was a forced push. Last 10 new commits:

f248552Trac #32061: add distro package information for GNU Gengetopt.
37299edTrac #32061: add spkg-check.in for gengetopt.
fd445a4first steps towards lcalc 2.0.0
62e2f2elcalc 2.0.0 removing all patches
2658f0dupdate installation scripts
a09e5a1lcalc 2.0.0 no dependency to mpfr
de1c43dTrac #32037: update spkg-configure.m4 for lcalc-2.0.0.
101ec5fTrac #32037: have lcalc depend on gengetopt.
24928e9Trac #32037: upgrade lcalc to v2.0.1.
4d5fc58Trac #32037: fix build scripts for lcalc SPKG.
7ed8c4ca-6d56-4ae9-953a-41e42b4ed313 commented 3 years ago

Changed commit from 3b44220 to 4d5fc58

7ed8c4ca-6d56-4ae9-953a-41e42b4ed313 commented 3 years ago

Changed commit from 4d5fc58 to 44f2e0d

7ed8c4ca-6d56-4ae9-953a-41e42b4ed313 commented 3 years ago

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

51f93f2Trac #32037: update libs/lcalc API for lcalc-2.x.
74ba355Trac #32037: use lcalc-2.x API in libs/lcalc/lcalc_Lfunction.pyx.
44f2e0dTrac #32037: update find_zeros_via_N() to match the lcalc-2.x API.
DaveWitteMorris commented 3 years ago
comment:34

Replying to @orlitzky:

Replying to @DaveWitteMorris:

... I guess when do_negative is true, we output the first n zeros in magnitude?

Yes, I'm pretty sure that is correct. For example, a comment introducing the find_zeros function says "Finds zeros of L(1/2+It) or of L(1/2+It)*L(1/2-It) (depending on whether self dual or not)". (The second form will combine the zeros with negative imaginary into the ones with positive imaginary part.) And a comment in the code of the function says "in the self dual case we only count half the zeros with |t|<T".

7ed8c4ca-6d56-4ae9-953a-41e42b4ed313 commented 3 years ago

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

13da394Trac #32037: ensure lcalc has PRECISION_DOUBLE and HAVE_PARI.
63bbfccTrac #32037: add Gentoo package information for lcalc.
7ed8c4ca-6d56-4ae9-953a-41e42b4ed313 commented 3 years ago

Changed commit from 44f2e0d to 63bbfcc

7ed8c4ca-6d56-4ae9-953a-41e42b4ed313 commented 3 years ago

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

8d70dfdTrac 32037: fixup lcalc header name in sage/libs/lcalc/lcalc_sage.h.
7ed8c4ca-6d56-4ae9-953a-41e42b4ed313 commented 3 years ago

Changed commit from 63bbfcc to 8d70dfd