octave: update to 7.3.0.

[tornaria]

• Fix cross build
• Use SuiteSparse
• Use rapidjson
• Use --disable-docs to skip building docs, instead install the docs in
  the tarball. This way we don't need texlive/texinfo/gnuplot to build.
• Avoid warnings due to egrep deprecation
• xlint fixes

Testing the changes

• I tested the changes in this PR: briefly

---

I wonder if we really want all those build options. I am tempted to remove almost all of them.

I also wonder if the choice of GraphicsMagick over ImageMagick is the best one.

NOTE: this PR includes #39846 for the current SuiteSparse. Could work with the current SuiteSparse, but I didn't try (#39846 is ready to merge IMO; the CI failures in that PR are addressed in the comments)

[hervyqa]

can you please rebase? CI runs again
igraph merged #40224

[Piraty]

remove the suitesparse commits from this branch. #39846 is merged

[tornaria]

The testsuite was giving a segfault on i686, when using qr on a sparse matrix. In fact this is trivial to reproduce:

octave:1> qr(sparse(1,1));
fatal: caught signal Segmentation fault -- stopping myself...
Segmentation fault

I "solved" it by using --without-spqr on i686 builds.

---

Then there are a couple of test failures seemingly due to numerical unstability, but they are not fixed with the standard tricks (-ffloat-store, -mfpmath=sse, etc). At least one is unrelated to octave as I can reproduce with:

$ cat sin.cc 
#include <complex>
#include <iostream>

int main()
{
	std::complex<double> c(-1.3962634015954636, 3.141592653589793);
	std::complex<double> d(-11.415845458288851, 2.0054175437381652);
	std::complex<double> e = std::sin(c) - d;

	std::cout << e << std::endl;
}
$ g++ sin.cc -o sin && ./sin                    
(-1.77636e-15,0)

The same code on x86_64 gives (0,0).

[tornaria]

Summary of testsuite (since failures don't error out):

• on x86_64, everything looks ok:

Summary:

  PASS                            17107
  FAIL                                0
  XFAIL (reported bug)               33
  SKIP (missing feature)            140
  SKIP (run-time condition)          52

• on x86_64-musl, 6 regressions:

  libinterp/corefcn/mappers.cc-tst ............................... pass  412/420 
                                                              REGRESSION   6
...
Summary:

  PASS                            17099
  FAIL                                0
  REGRESSION                          6
  XFAIL (reported bug)               35
  SKIP (missing feature)            140
  SKIP (run-time condition)          52

• on i686, 2 failures and 1 regression:

  libinterp/corefcn/qz.cc-tst .................................... pass    3/4   
                                                                   FAIL    1
  libinterp/corefcn/sqrtm.cc-tst ................................. pass    3/4   
                                                              REGRESSION   1
...
  elfun/sind.m ................................................... pass    9/10  
                                                                   FAIL    1
...
Summary:

  PASS                            17095
  FAIL                                2
  REGRESSION                          1
  XFAIL (reported bug)               30
  SKIP (missing feature)            152
  SKIP (run-time condition)          52

Note that i686 has more skips because of the --without-spqr.

I'm satisfied as it is, I'm not going to work on fixing these unless this causes some real issue for someone.

[Piraty]

good work!!