Most of these remarks are related to unknown versions of Math::Pari
and/or GP/PARI.

Need an additional field to ep_produce: where parsing ended.
fmt_nb only a placeholder.
NOEXP2 needs to be fixed.
init_opts declared in gp/gp.h.
wrong dependence for none/kernel0.o
Need to use $(AS) on sparcs, alpha, hppa.

changevalue() and installep() are statics.
numvar() disappeared.
l0asm.c, Gnuplot.h, plotport.c anal.h, init.c, anal.c edited.

Add for -g:

dummy$(OBJ_EXT): pariinl.h
	$(CCCMD) $(CCCDLFLAGS) $(DEFINE) -o $@ pariinl.h

pariinl.c: pariinl.h
	$(CPP) -I. $(INC) -I../../src/kernel/ix86/ $(DEFINE) $(CPPMINUS) < ../../src/kernel/none/level0.c | perl -ne "s/%s+/%/g; print unless /^s*#/" > $@

dummy$(OBJ_EXT): pariinl.c
	$(CCCMD) $(CCCDLFLAGS) $(DEFINE) -o $@ pariinl.c

install() installs at the head of the chain, but findentry() expects
that the installed entry goes into the end of the chain.

err_recover() was flush()ing before giving the caller the possibility to
die().

coinit() assumes that a long can be put in 9 chars ?!
p-adic was outputting extra space.

bruteall() was overwriting old sp.

a) T_PariPtr (to process &) is not implemented.
1) Lookup for autoloaded functions happens too early: after
   functions_hash, but before gp_hash.  Say, `kill' wants to be autoloaded.
2) default() uses a static.
3) tests/analyz does not check answers...