Enea Zaffanella writes:
[About make check].
This looks strange to me: the message seems to suggest that the system finds the OCaml mlgmp wrapper, but it does not find the C++ wrapper libgmpxx.
I am definitely not an expert of OCaml related builds, but I think that I have read somewhere that OCaml native library wrappers were encoding in their inside the right link options, so that one should not need to issue them each time.
I will need to investigate further, but for the immediate I can imagine two possible explanations :
a) Maybe you did something wrong when compiling the mlgmp wrapper. As said in file interfaces/Ocaml/README.ocaml: =========== In order to install MLGMP you should make sure that GMP_INCLUDES and GMP_LIBDIR are correctly set (these variables are defined at the beginning of Makefile in the MLGMP's root directory). =========== Could it be the case that you did something wrong here?
No, MLGMP's OK.
b) What is the output of the command echo $LD_LIBRARY_PATH on your machine? Does it include the path to your installation place, that is /afs/inf.ed.ac.uk/user/k/kwxm/libraries/lib ?
Let us know if either a) or b) solves the problem.
I have to admit that I'm very confused by this. I don't have LD_LIBRARY_PATH set, and you're correct in suggesting that if I set it to point to the location of my libraries then the test works OK. However, I've been able to compile and link my own programs without the use of LD_LIBRARY_PATH with no problems, and I can also compile your test program without using it. Here are the commands which interfaces/OCcaml/tests/Makefile executes when it's compiling test1:
OCAMLRUNPARAM='l=1M' ocamlc -o test1.cmo -c -I /home/kwxm/libraries/lib -I .. -ccopt -g test1.ml OCAMLRUNPARAM='l=1M' ocamlc -o test1 \ -cc "g++" -I /home/kwxm/libraries/lib -I .. -ccopt -g \ -cclib ../../../src/.libs/libppl.so -cclib ../../../Watchdog/src/.libs/libpwl.so -cclib -lmlgmp -cclib -lmpfr `echo " -lm -L/home/kwxm/libraries/lib -lgmpxx -L/home/kwxm/libraries/lib -lgmp -R/home/kwxm/libraries/lib -R/home/kwxm/libraries/lib " | /bin/sed -e "s/ -R[^ ]*//g" -e "s/ -/ -cclib -/g"` \ ppl_ocaml.cma test1.cmo
Note that the second command, which does the linking, explicitly mentions things like libppl.so and -lgmpxx.
If you run the above commands and then use ldd, you get the following:
ldd test1 linux-gate.so.1 => (0xb7fc4000) libgmpxx.so.4 => not found libgmp.so.3 => /usr/lib/libgmp.so.3 (0xb7f4e000) libppl.so.7 => not found libpwl.so.4 => not found libmpfr.so.1 => /usr/lib/libmpfr.so.1 (0xb7f0b000) ...
So in spite of all the options, it still doesn't know where to find libgmpxx.so. It turns out that if you make a directory containing only libgmpxx.so.4 and point LD_LIBRARY_PATH to it, then test1 still fails because it can't find libppl.so.7, and similarly for libpwl.so.4. Once it's got all of these it works OK though.
I spent some time experimenting and found that I could do the compilation with much simpler commands. To avoid the danger of interacting with my own previously-compiled version of PPL and other libraries, I made a directory called ~/gmp containing libgmp.a and libgmp.so*, together with the files gmp.a, gmp.cma and gmp.cmxa from mlgmp. I can compile test1.ml by saying
ocamlc -c -o test1.cmo -I ~/gmp -I .. test1.ml
This is pretty much the same as what's in the makefile. I think that at this stage it just needs to be able to find the .cma files for the OCaml modules used by test1.
After this, I can link test1 with
ocamlc -o test1 ../ppl_ocaml.cma ~/gmp/gmp.cma test1.cmo \ -cclib -lppl -cclib -lpwl
(instead of using ../ppl_ocaml.cma you can also say -I .. ppl_ocaml.cma, and similarly for gmp)
After this test1 runs successfully.
Note that I don't have to refer to -lgmp or any of the other libraries (but you can't leave out -lppl or -lpwl). If you type "strings ../ppl_ocaml.cma" then at the very end you see some strings which mention stuff like -L/home/kwxm/libraries/lib and -lgmpxx, so I think the ocaml compiler records some of the information that's necessary for the final linking.
[Some time later...]
Ah, now I see that what makes the difference is including gmp.cma in the linking command. If you insert it just after libppl.so in the makefile then everything works OK, and I think most of the other options are unnecessary. I still don't understand exactly what's going on though. :(
Kenneth
PS: my problematic BHRZ03 widening just finished, after 31 hours and 40 minutes. The result was the same as that of the H79 widening.