Rework spkg/install, spkg/standard/deps, spkg/standard/newest_version

[jdemeyer]

The current system of spkg/install setting environment variables like $TERMCAP with the version numbers of the various packages is quite bad.

In this ticket, I change spkg/install to create a file spkg/Makefile (which includes spkg/standard/deps) which is used as makefile instead of spkg/standard/deps. Instead of using environment variables for the versions, I simply use make variables.

The shell script newest_version is also removed in favor of a shell function in spkg/install. It gracefully handles non-existing packages. Previously a non-existing package will not signal any error or warning.

Apply:

1. 10492_deps_to_Makefile.patch to SAGE_ROOT
2. 10492_no_more_newest_version.patch to SCRIPTS.

See also #12609: fixing the developer manual to take into account the changes on this ticket.

[vbraun]

It would be nice if a shell function newest_version() would also be available for other spkg to check that required versions of dependencies are available. So I'd like it to be in 3rd file install-functions (or so) that is sourced by install. Perhaps with a better name, for example requires_version(mpir,1,1,2)...

Moreover, the current newest_version script does not check that it is called from $SAGE_ROOT/spkg. If it is called from the wrong directory, I get

/home/vbraun/Sage/sage/spkg/standard
[vbraun@volker-two standard]$ ./newest_version mpir
mpir-1.2.2.p2

on Linux and

vbraun@t2:standard$ pwd
/home/vbraun/t2/sage-4.6.1-sunos-32bit-5.10-sun4v-SunOS/spkg/standard
vbraun@t2:standard$ ./newest_version mpir

(no output) on t2/Solaris.

[jhpalmieri]

Is the code in data/extcode/debian dead? It uses newest_version, but is it relevant anymore?

Otherwise, the changes here make sense, and I've been using this (via #12369) for some time now with no problems. I'll test on a few more systems, and then (pending the answer to my question) I can give it a positive review.

[jhpalmieri]

I think that make distclean should delete spkg/Makefile.

[jdemeyer]

For Debian, see #12470. Yes, it's unused.

[jdemeyer]

Replying to jhpalmieri:

I think that make distclean should delete spkg/Makefile.

Yes, it should. Fixed.

[jdemeyer]

Thanks for the review!

Looks like the whole build system will be cleaned up nicely in sage-5.0 (#11073, #12479, #10492,...)

[leif]

It would be nicer (and more appropriate) to use basename (which is POSIX) in newest_version*().

Any reason to copy spkg/standard/deps rather than include it by the generated Makefile?

[jhpalmieri]

It turns out that this change breaks the optional Sage package database_gap, since that package uses the newest_version script. See #12712.

[jdemeyer]

Replying to leif:

Any reason to copy spkg/standard/deps rather than include it by the generated Makefile?

1) because it looks cleaner to me to have just one Makefile.

2) because I was worried about the portability of including files inside Makefiles.

[leif]

Replying to jdemeyer:

Replying to leif:

Any reason to copy spkg/standard/deps rather than include it by the generated Makefile?

1) because it looks cleaner to me to have just one Makefile.

Well, after copying, we still have two... ;-)

2) because I was worried about the portability of including files inside Makefiles.

Then you should avoid looking at the eclib package, which does (really) horrible things... B)

(Sage won't build with anything else but GNU make anyway, and this will not change in the foreseeable future, since many upstream packages rely on GNU make as well.)

[leif]

Any nice idea of how one could pass some flags just to the top-level make, but not to the instances building spkgs? (I'm thinking of -k.)

[jdemeyer]

Replying to leif:

Any nice idea of how one could pass some flags just to the top-level make, but not to the instances building spkgs? (I'm thinking of -k.)

Assuming that spkg/Makefile exists:

cd spkg
MAKE="make -S" make -k

(not tested)