Yes... Although I somehow think it should be allowed to compiled and tested even if nettle isn't in the dist, like it can be compiled and tested without gmp.

/ Mirar

Previous text:

2004-02-18 12:32: Subject: Crypto

---

Demanding Nettle would seem the way to go. Then we get to demand Gmp at the same time, since Nettle relies on Gmp. :-)

/ Marcus Comstedt (ACROSS) (Hail Ilpalazzo!)

Well, the parts of nettle that are used by the pike glue doesn't depend on Gmp. As far as I know.

About this bundling stuff, where should the source for the nettle and gmp libraries live in a pike dist? Pike-v47.11/gmp and Pike-v47.11/nettle seem reasonable. This is if we want fix places, otherwise Pike-v47.11/nettle-1.9 might be better.

Then next question is if and when to build them. There should be some rules in the top-level Makefile I guess, but it seems a little ugly to have src/configure invoke "make -f ../Makefile gmp".

/ Niels Möller (vässar rödpennan)

Previous text:

2004-02-18 12:42: Subject: Crypto

---

Yes. The configure scripts in GMP and Nettle should be fixed so that the bundled Nettle/GMP is built and used if none is found on the target system. And make export should be fixed so that GMP and Nettle is bundled.

/ Martin Nilsson (saturator)

One way would be to first run all pike's configurescripts. The nettle and gmp modules could use some PIKE_NEED_BUNDLED_LIB, which would (i) add necessary link and cpp flags to uses the bundled libraries, and (ii) write some file indicating to the top level Makefile which bundled libraries are needed.

Then make of the top level should build the bundled libraries and install them in some proper place in the buildtree, before running make in pike's build tree.

Installation of bundled libraries seem a little tricky. Either then can be installed in the build tree, and link the corresponding modules statically. Or we link dynamically and install (at make install time, not earlier!) them in some proper place on the system, and then we'll need some additional trickery to have a working pike before make install time.

/ Niels Möller (vässar rödpennan)

Previous text:

2004-02-18 13:28: Subject: Crypto

---

If we put some libraries in e.g. ./bundle then the configure/Makefile in that directory could just decide to build its children if no system counterpart was found. Perhaps that can be made general and abstract enough so that one can just drop in libraries with minimal extra configurations. But perhaps that would be too good to hope for...

/ Martin Nilsson (saturator)

Makes sense.

/ Niels Möller (vässar rödpennan)

Previous text:

2004-02-18 13:45: Subject: Crypto

---

We should link statically I would think.

/ Martin Nilsson (saturator)

One of the most annoying features of the configure script is that you get a fatal for missing Gmp after having run it for minutes, so if possible the tests for zlib, gmp and nettle should be the first to run.

/ Peter Bortas

Previous text:

2004-02-18 13:42: Subject: Crypto

---

One way would be to first run all pike's configurescripts. The nettle and gmp modules could use some PIKE_NEED_BUNDLED_LIB, which would (i) add necessary link and cpp flags to uses the bundled libraries, and (ii) write some file indicating to the top level Makefile which bundled libraries are needed.

Then make of the top level should build the bundled libraries and install them in some proper place in the buildtree, before running make in pike's build tree.

Installation of bundled libraries seem a little tricky. Either then can be installed in the build tree, and link the corresponding modules statically. Or we link dynamically and install (at make install time, not earlier!) them in some proper place on the system, and then we'll need some additional trickery to have a working pike before make install time.

/ Niels Möller (vässar rödpennan)

Because when it really hurts you are trying to build on something archaic like a cray and forgot to check the function of the libraries before starting the Pike build.

/ Peter Bortas

Previous text:

2004-02-18 14:21: Subject: Crypto

---

Why? If the script uses bundled versions instead of failing it shouldn't matter when the test is performed.

/ Marcus Comstedt (ACROSS) (Hail Ilpalazzo!)

No, after I have fixed that one of grubbas* sadistic tests will find a flaw in it and spit it back.

*Exempel Helge.

/ Peter Bortas

Previous text:

2004-02-18 14:30: Subject: Crypto

---

Eh? "Check the function of the libraries"? If Gmp doesn't compile on Cray, that's something which has to be fixed. That's a one-time job.

/ Marcus Comstedt (ACROSS) (Hail Ilpalazzo!)

In the case of Gmp it might be true. For any other random library (zlib, freetype, gl etc) I wouldn't be so sure.

/ Peter Bortas

Previous text:

2004-02-18 14:34: Subject: Crypto

---

One would expect that Gmp's own testsuite would be sufficient to find any problems with the library itself?

/ Marcus Comstedt (ACROSS) (Hail Ilpalazzo!)