Build PPL with its C interface

[SimonKing]

Currently, the standard PPL spkg only builds the C++ interface. However, CLooG-PPL (see #12666) needs the C-header ppl_c.h as well.

Therefore I suggest to slightly modify the install script of our PPL spkg:

• Do make install instead of make install-strip -- I admit that I am not sure whether this is needed, and perhaps make install-strip yields faster code. A reviewer should comment on that.
• Modify the configuration options, such that the C interface is built.

spkg: ​http://boxen.math.washington.edu/home/SimonKing/SAGE/spkgs/ppl-0.11.2.p2.spkg

[SimonKing]

Interestingly, when I tried doc tests with a Sage version that used massive optimization (via graphite), a lot of tests went a lot slower (by a factor of 2 in some cases) than without the optimization.

I had the impression that the particularly bad cases were related to polyhedra. Therefore, I try to build ppl with make install-strip again, and keep my fingers crossed that it will improve the situation.

[SimonKing]

It could be that the bad performance came from using GMP instead of MPIR. It turns out that cloog-ppl can be built with MPIR (originally, I thought GMP is required). Anyway, I am now building Sage without GMP but with high optimization, and hope that it'll work better.

[aginiewicz]

This also fixes #11391, where on Arch Linux and some other new 32-bit distros there are missing symbols in ppl during building of pycrypto - enabling C interface makes it work again. I haven't noticed this ticket, as I started testing fix for 11391 before this one was created, so I posted nearly same spkg in #11391 (with C interface, but without install vs install-strip change).

As I worked on the same thing just few hours ago, after quick check on sage 4.8 I just made in VM I can say that it works, but I don't feel experienced enough to comment on install vs install-strip, so I'm leaving it as needing review.

[jdemeyer]

make install-strip removes debugging information from the PPL library. I don't see the need for that, so I actually prefer install over install-strip.

[jdemeyer]

Unfortunately, this seems to require a recent vesion of M4 (which isn't present on OS X 10.4):

checking whether to build the Parma Watchdog Library... yes
checking whether to build the ppl_lcdd program... yes
checking whether to build the ppl_lpsol program... yes
checking whether to build the ppl_pips program... yes
checking which interfaces are enabled... cxx c
checking for GNU M4 that supports accurate traces... configure: error: no acceptable m4 could be found in $PATH.
GNU M4 1.4.5 or later is required; 1.4.11 is recommended
Error configuring the Parma Polyhedra Library.

[jdemeyer]

This is a duplicate of #11391 (which has the same solution for a different problem).

[SimonKing]

Replying to jdemeyer:

This is a duplicate of #11391 (which has the same solution for a different problem).

I think it is not a duplicate. #11391 is about building ppl on ArchLinux and OpenSuse and so on. The ticket here is about building ppl with its c interface. So, these are orthogonal problems.

[SimonKing]

It remains "needs work", because of the problem on OS X you mentioned. So, do we need an M4 package??

[kcrisman]

I don't see why one couldn't just do the usual thing we do in spkg-install for unusual systems - just a check for Darwin with version 8 or 9 or something. Or even check for m4's existence and version (can this be done easily?). In either case, just don't do

-    --enable-interfaces=c++ --disable-fpmath
+    --enable-interfaces="c++ c" --disable-fpmath

for that specific situation, maybe echoing a message. Then #12666 can raise an error message if one doesn't have the correct header, saying "please get a system with a recent enough m4, do sage -i ppl, and then reinstall this optional spkg".

[jdemeyer]

Replying to kcrisman:

just a check for Darwin with version 8 or 9 or something.

That's too fragile, nobody guarantees that this is the only system with a bad m4.

Or even check for m4's existence and version

If you create a patch, I'll be happy to review it. Note that we already require the existence of m4 in prereq (MPIR needs it).

[kcrisman]

Replying to jdemeyer:

Replying to kcrisman:

just a check for Darwin with version 8 or 9 or something.

That's too fragile, nobody guarantees that this is the only system with a bad m4.

True, in fact it's quite likely others will be thus.

Or even check for m4's existence and version

If you create a patch, I'll be happy to review it. Note that we already require the existence of m4 in prereq (MPIR needs it).

[kcrisman]

Here is a first try.

spkg-install

@@ -39 +39 @@
     fi
 done
 
-./configure --prefix="$SAGE_LOCAL" --libdir="$SAGE_LOCAL/lib" \
-    --with-gmp-prefix="$SAGE_LOCAL" --enable-coefficients=mpz \
-    --enable-interfaces="c++ c" --disable-fpmath
+# If m4 is not recent enough, we fall back to only enabling the C++ interface
+M4_VERSION=`m4 --version | grep GNU | cut -d " " -f 3`
+if [[ $M4_VERSION < 1.4.5 ]]; then
+    echo "Configuring only for C++ interface"
+    echo "Install GNU M4 1.4.5 or later for C interface"
+    ./configure --prefix="$SAGE_LOCAL" --libdir="$SAGE_LOCAL/lib" \
+        --with-gmp-prefix="$SAGE_LOCAL" --enable-coefficients=mpz \
+        --enable-interfaces=c++ --disable-fpmath
+else
+    echo "Great, you have GNU M4 1.4.5 or later"
+    echo "Configuring for C++ and C interfaces"
+    ./configure --prefix="$SAGE_LOCAL" --libdir="$SAGE_LOCAL/lib" \
+        --with-gmp-prefix="$SAGE_LOCAL" --enable-coefficients=mpz \
+        --enable-interfaces="c++ c" --disable-fpmath
+fi
 if [ $? -ne 0 ]; then
    echo >&2 "Error configuring the Parma Polyhedra Library."
    exit 1

But it doesn't work, because the extended bash test doesn't quite do what we want.

$ if [[ 1.4.11 > 1.4.5 ]]; then echo "bigger"; fi
$

I cringe at the more comprehensive solutions to this problem I found on the internet, though. The sort -V command might have been helpful, but that flag isn't universally available.

[jdemeyer]

Fixed by #14232.