Upgrade autotools & cleanup

[saraedum]

This tries to modernize the use of autotools in cddlib (see also https://groups.google.com/forum/#!msg/sage-devel/uHsTOd5sTxY/ivpNHT1MAwAJ)

Most significant changes:

• Drop all (?) autogenerated files from the repository.
• Add ./bootstrap script to regenerate files used by autotools and friends.
• Make make distcheck work again which creates cddlib-0.94i.tar.gz for distribution. (So in the future to create a releasable tarball one would increase the version number in configure.ac, run ./bootstrap, run ./configure and make distcheck.)

Many of the changes are originally due to @embray.

TODO

• Check that this works for OSX Ok. TravisCI takes care of that.
• Update the README so it talks about ./bootstrap for people who clone from here instead of downloading a tarball. Let's postpone this to a separate PR. The README needs some fixes to make it looks nice on github anyway.
• Check the contents of the tarball (see below)
• Things such as export gmpdir := /usr/local in Makefile.am seem dubious. ./configure should decide that I guess? I just set -lgmp conditionally now. I think that's good enough for GMP (otherwise we could use pkg-config?) Note that this is also what most distributions patch into the configure.in file.
• cddlib installs itself as libcdd.so.0.0.0 and libcddgmp.so.0.0.0. This can now be changed in configure.ac.
• should the binaries really install to /usr/bin/? I do not really understand what people use in the end, but maybe some things should not be on the PATH? (So maybe these should go into /usr/libexec instead? Or not be installed at all?) Let's put everything into PREFIX/bin for now.
• documentation does not install with make install. Should it? It does now.
• should the documentation be built as part of make dist? the currently distributed .pdf, .dvi, .ps is now built by make dist.
• examples do not install with make install. Should they? They do now.
• Have a look at Debian's patches.
• Have a look at Archlinux's patches.
• Have a look at SageMath's patches.
• Have a look at Conda's patches.

[saraedum]

TODO

These files are part of the repository but not part of the tarball. Check which of the are actually missing:

• bootstrap. Ok. Now part of the tarball.
• doc. Ok. We now ship the pdf, ps, dvi, but not the html documentation (like before). Is that Ok?
• .git, Ok, the git history should not be in the tarball.
• .gitignore, Ok, nothing git related should be in the tarball.
• HISTORY, not a standard file and identical to NEWS, so I think it's Ok if it's missing.
• README.core2processor, it's included again.
• .gitignore, Ok, nothing git related should be in the tarball.
• src/minkowski.c, included
• src/redexter.c, included
• src/redundancies.c, included
• src/redundancies_clarkson.c, included
• src/testuniq.c, included
• src-gmp/.gitignore, Ok, nothing git related should be in the tarball.

These files are part of the tarball but not in the repository. That's ok. They are all generated by autotools.

• aclocal.m4
• compile
• config.guess
• config.sub
• configure
• depcomp
• INSTALL
• install-sh
• Makefile.in
• lib-src-gmp/Makefile.in
• ltmain.sh
• m4
• Makefile.in
• missing
• src/Makefile.in
• src-gmp/Makefile.in

[saraedum]

I am not sure what to do with these source files:

• src/minkowski.c
• src/redexter.c
• src/redundancies.c
• src/redundancies_clarkson.c
• src/testuniq.c
• src-gmp/redundancies.c
• src-gmp/redundancies_clarkson.c
• src-gmp/testuniq.c

They are apparently not used in any of the Makefiles. Is that on purpose or should they be built as well?

[saraedum]

I had a look at all the patches used elsewhere.

• I create Include cdd_both_reps #3 to add an additional binary that everybody is using apparently (probably because Sage is using it.)
• SageMath adds a "portable random number generator". I guess the reason is to have the randomness of results not be dependent of the underlying implementation of rand. I can see why we do that at SageMath, but I think it's a bad idea anyway. If stuff is random, we should just live with that in our doctests. (Or we should link a patched libc that gives us portable random numbers…but we shouldn't patch this in a single package.); this is now https://trac.sagemath.org/ticket/25238
• Everybody is patching the GMP detection (which I did here as well.)
• Debian has a man-page for the additional binary but that's probably not worth it for just one binary.

[saraedum]

The author answered about the .c file that are not in the Makefiles:

Those .c files not used are experimental programs that may be
of interest for researchers and advanced users. They are there so that
anyone who has advanced questions can see how some advanced techniques
in computational geometry might test the behaviour of certain algorithms
that might be interesting. I do not know these examples should be in src directories.

So, I think it makes sense to build them, so the interested developer does not need to figure out how to do that. But they should not be installed.

[saraedum]

The author answered about the examples:

I do not know what is the usual practice. If the user expects that these
example files are in a certain directory, they should be copied there.
But, I do not like these files are copied to the primary locations under /usr .
I much prefer all third party softwares are in /usr/local, because one can easily
separate the core binaries/libraries from the third party ones.

Since the documentation mentions these example files, they should probably be part of the documentation? Whether they get installed to /usr or /usr/local depends on how you call ./configure. The default is indeed for everything to go to /usr/local. But if you install this with say Debian, then they would of course go into /usr.

[saraedum]

@komeifukuda dc05ae0 makes your sed script unnecessary I think.

[saraedum]

@komeifukuda strangely, testuniq.c in src and src-gmp are not the same. Here is the diff.

--- src/testuniq.c	2018-04-22 22:25:01.636505790 +0200
+++ src-gmp/testuniq.c	2018-04-22 22:25:01.635505776 +0200
@@ -59,17 +59,15 @@
 
 int main(int argc, char *argv[])
 {
-  dd_MatrixPtr M=NULL,M1=NULL,M2=NULL;
+  dd_MatrixPtr M=NULL,M1=NULL,M2=NULL,M3=NULL,M4=NULL;
   dd_rowrange i;
   dd_colrange d;
   dd_ErrorType err=dd_NoError;
-  dd_rowset redrows,linrows,posset;
-  dd_rowindex newpos=NULL, newpos1=NULL, newpos2=NULL;
+  dd_rowset redrows,linrows;
+  dd_rowindex newpos1=NULL, newpos2=NULL,newpos4=NULL;
   mytype val;
   dd_DataFileType inputfile;
   FILE *reading=NULL;
-  int foi;
-  dd_Arow certificate;
 
   dd_set_global_constants();  /* First, this must be called. */
 
@@ -92,69 +90,67 @@
 
   d=M->colsize;
   
-  /* 
   printf("\nInput Matrix.\n");
-  dd_WriteMatrix(stdout, M); 
-  */
- 
-  M1=dd_MatrixSortedUniqueCopy(M,&newpos1);
-  printf("\nSort and remove duplicates with dd_MatrixSortedUniqueCopy.\n");
+  dd_WriteMatrix(stdout, M);
+  M1=dd_MatrixSortedCopy(M,&newpos1);
+  printf("\nNormalize and sort the matrix with dd_MatrixSortedCopy.\n");
   printf(" Row index changes -- original:new\n");
   for (i=1;i<=M->rowsize; i++){
     printf(" %ld:%ld",i,newpos1[i]);
   }
   printf("\n");
-  /* dd_WriteMatrix(stdout, M1); */
 
-  dd_InitializeArow(M1->colsize+2, &certificate);
-  fprintf(stdout, "\nCheck whether the system contains an implicit linearity.\n");
-  foi=dd_FreeOfImplicitLinearity(M1, certificate, &posset, &err);
-  switch (foi) {
-    case 1:
-         fprintf(stdout, "  It is free of implicit linearity.\n");
-      break;
-      
-    case 0:
-         fprintf(stdout, "  It is not free of implicit linearity.\n");
-      break;
-
-    case -1:
-         fprintf(stdout, "  The input system is trivial (i.e. the empty H-polytope or the V-rep of the whole space.\n");
-      break;
+  dd_WriteMatrix(stdout, M1);
+  
+  M2=dd_MatrixUniqueCopy(M1,&newpos2);
+  printf("\nRemove row (consecutive) duplicates with dd_MatrixUniqueCopy.\n");
+  printf(" Row index changes -- original:new\n");
+  for (i=1;i<=M1->rowsize; i++){
+    printf(" %ld:%ld",i,newpos2[i]);
+  }
+  printf("\n");
+  dd_WriteMatrix(stdout, M2);
+   
+   
+  M4=dd_MatrixSortedUniqueCopy(M,&newpos4);
+  printf("\nTwo operations can be done at once with dd_MatrixSortedUniqueCopy.\n");
+  printf(" Row index changes -- original:new\n");
+  for (i=1;i<=M->rowsize; i++){
+    printf(" %ld:%ld",i,newpos4[i]);
   }
+  printf("\n");
+  dd_WriteMatrix(stdout, M4);
 
 
-  fprintf(stdout, "\nOne can then remove nontrivial redundant rows with dd_RedundantRows.\n Redundant rows:\n");
-  redrows=dd_RedundantRowsWithPresorting(M1, &err);
-  /* redrows=dd_RedundantRows(M1, &err);  */
+  fprintf(stdout, "\nOne can then remove nontrivial redundant rows with dd_RedundantRows.\n Redundant rows:");
+  redrows=dd_RedundantRows(M2, &err);
   set_fwrite(stdout, redrows);
-  printf("\n");
 
-  M2=dd_MatrixSubmatrix2(M1, redrows, &newpos);
+  M3=dd_MatrixSubmatrix(M2, redrows);
   dd_FreeMatrix(M1);
+  dd_FreeMatrix(M2);
   set_free(redrows);
-  free(newpos);
-  
+
   fprintf(stdout, " Implicit linearity (after removal of redundant rows): ");
-  linrows=dd_ImplicitLinearityRows(M2, &err);
+  linrows=dd_ImplicitLinearityRows(M3, &err);
 
   if (M->representation==dd_Generator)
     fprintf(stdout," %ld  ", set_card(linrows));
   else 
     fprintf(stdout," %ld  ", set_card(linrows));
   set_fwrite(stdout,linrows);
-  set_uni(M2->linset, M2->linset, linrows); 
+  set_uni(M3->linset, M3->linset, linrows); 
     /* add the implicit linrows to the given linearity rows */
 
   printf("\nNonredundant representation (except for the linearity part):\n");
-  dd_WriteMatrix(stdout, M2);
-  dd_WriteLPStats(stdout);
+  dd_WriteMatrix(stdout, M3);
   set_free(linrows);
 
   dd_FreeMatrix(M);
-  dd_FreeMatrix(M2);
+  dd_FreeMatrix(M3);
+  dd_FreeMatrix(M4);
   dd_clear(val);
-  free(newpos1); free(newpos2);
+  free(newpos1); free(newpos2); free(newpos4);
 
 _L99:;
   /* if (err!=dd_NoError) dd_WriteErrorMessages(stderr,err); */

[saraedum]

@komeifukuda Is there any reason why one would want to build the binaries without GMP support if GMP is available? I think it's a bit non-standard to build two flavours of every binary.

[saraedum]

Or rather: I think if one really wants this, then it should simply be:

./configure --without-gmp
make
./configure --program_suffix=_gmp
make

[saraedum]

So. What is the right default: with or without GMP support? @komeifukuda

[saraedum]

@komeifukuda I think I have fixed all the things that I wanted to. If you are happy with this this, just press "Merge pull request". Otherwise I am looking forward to hearing your suggestions.

[saraedum]

If you want to try this out:

git clone git@github.com:saraedum/cddlib.git -b autotool-cleanup
cd cddlib
./bootstrap
./configure
make dist