Fragile GMP header hack in Singular and track all files

[leif]

One (variant of possible) symptom(s):

In file included from ../kernel/si_gmp.h:4:0,
                  from ../kernel/structs.h:15,
                  from weight0.c:13:
../factory/cf_gmp.h:2501:28: fatal error: bits/c++config.h: No such file or directory
  #include <bits/c++config.h>
                             ^
compilation terminated.
make[4]: *** [weight0.o] Error 1

This is still present in Singular 3-1-6.

---

The file factory/cf_gmp.h is supposed to contain a copy of the otherwise used gmp.h found on the system, with some C++ definitions removed, later on to be used both by Singular C and C++ files (i.e., those include cf_gmp.h instead of gmp.h).

The actual problem is getting the location (i.e., the absolute path) of the header file used by the C/C++ compiler / preprocessor, in order to create a modified copy of it.

The latter is achieved by using GCC's -M switch to output Makefile dependencies (also taking into account include directory search paths eventually specified by the user, probably via --with-gmp=...), but isn't robust w.r.t. the formatting of the generated rules.

In particular, undesired additional header files (which usually include others) may end up in cf_gmp.h, leading to (potentially arbitrary) problems at least when C files include it, as the folders searched for C headers usually differ from those used when compiling / preprocessing C++. (This is the case in the rather obscure error above, where bits/c++config.h isn't found.)

The current patch to the Singular spkg both simplifies the generation of cf_gmp.h (no longer needing gen_cf_gmp.template and [to ship] gen_cf_gmp.cc, nor the generation of a temporary shell script, gen_cf_gmp.sh), and of course makes it more robust, also making sure only the contents of gmp.h ends up in its modified copy.

---

fixed spkg: ​http://boxen.math.washington.edu/home/vbraun/spkg/singular-3-1-5.p8.spkg

[leif]

(More to come...)

[leif]

For the impatient:

An initial fixed spkg can be found ​here. (Currently just adds a patch to src/factory/GNUmakefile.in, also simplifying the header generation a bit.)

[mmezzarobba]

I used the patch to build sage 5.11β3 on a system that used to be affected by this bug (debian sid, gcc 4.8). It works for me.

[leif]

Replying to mmezzarobba:

I used the patch to build sage 5.11β3 on a system that used to be affected by this bug (debian sid, gcc 4.8). It works for me.

Someone else just reported the same error for Debian testing... :-)

[leif]

Slightly changed the spkg; patch now removes unneeded files as well.

[leif]

Proposed patch to upstream (included in the .p8 spkg). For reference / review.

[ncohen]

This spkg fixed the problem for me too ! And I also run debian sid. Thank you very much ! ;-)

Nathann

[vbraun]

Leif, are you still working on the patch or is this ready for review?

[vbraun]

The GMP header patch looks good to me (positive review)

Somebody needs to review the sage-src script and then we are ready.

[jpflori]

I'm ok with spkg-src.