Disable by default (and eventually get rid) of "compatibility mode" in our build system

[fingolfin]

Our build system has a "compatibility mode" which is on by default, and exists to help packages with kernel extensions that expects certain things to be arranged exactly as they were in older GAP versions (before the new build system in 4.9).

I'd like to switch the default to "off" and at some point remove this mode. But that requires active testing for it and probably adjustments to a few packages. And possibly also to the GAP build system -- it may be that a few things that I back then only enabled in "compatibility mode" because I thought we don't need it anymore actually should be retained.

I plan to work on this, but I wouldn't mind help, so if anyone is interested... :-)

[ChrisJefferson]

How does one switch it off? I can check if anything terrible happens on windows

[fingolfin]

There is a configure switch, ask ./configure --help. As it is, I am not even sure it works at all; and it may already break stuff on Linux

[fingolfin]

Some things compat mode creates:

1. bin/gap.sh
   • this still used by a ~13 packages, and also by GAP itself.
2. symlink sysinfo.gap-default64 -> sysinfo.gap
3. a bunch of symlinks in bin/$GAPARCH:
   • config.h -> ../../build/config.h`
   • gac -> ../../gac`
   • gap -> ../../gap`
   • src -> ${srcdir}/src

Since a lot uses bin/gap.sh, I'd leave it in for now. But the other kinks are prime candidates for removal. I've done this now in PR #5101

[fingolfin]

Packages which still seem to be using gap.sh unconditionally:

• anupq (in docs and internal helper scripts)
• digraphs (in the doc target of Makefile.am) -- resolved in 1.6.2
• float (in the doc target of Makefile.am -- see buildsys: locate GAP executable using info from sysinfo.gap gap-packages/float#89)
• jupyterkernel (mentions it in README.md)
• semigroups (in the doc target of Makefile.am; already resolved on main branch, just needs a release)
• unipot (mentions it in README.md, install instructions)
• zeromqinterface (in the zgap script, which has been broken for a long time anyway)

So we are looking pretty good in this regard.