Copy lib/lua-$(VERSION)/src/liblua.h to lib/lua/ rather than lib/.
Instead of copying any of the header files or telling the compiler
where to find the lua ones, generate include/nhlua.h on the fly and
restrict the knowledge of where they are to it (paths are relative
to include/).
|/* nhlua.h - generated by top Makefile */
|#include "../lib/lua-5.3.5/src/lua.h"
|LUA_API int (lua_error) (lua_State *L) NORETURN;
|#include "../lib/lua-5.3.5/src/lualib.h"
|#include "../lib/lua-5.3.5/src/lauxlib.h"
|/*nhlua.h*/
This might need to be redone (or augmented by having CFLAGS add back
'-I path-to-lua') if some compiler can't find '#include "luaconf.h"'
issued by lua.h.
undefined reference to g.variables if certain parts of
drawing.c are included in host-side utilities, so
surround the offending code in
-> #if !defined(CROSSCOMPILE) || defined(CROSSCOMPILE_TARGET)
A reddit thread about an unaligned altar in an aligned temple was
a tipoff that mimics posing as altars didn't have any particular
alignment. The look-at code was misusing an operloaded field of the
underlying terrain. Pick an alignment at random when taking on the
appearance of an altar, store it in the mimic's mon->mextra->mcorpsenm
field, and have look-at use that.
Also, dropping a ring of polymorph into a sink can transform it, and
one possible outcome is an altar. In this case, the alignment is
part of the location's topology, but code setting that up was using
Align2amask(rn2(foo)). That's a macro which evaluates its argument
more than once. The first evaluation was effectively a no-op. If
the second evaluation picked lawful then the result was lawful as
intended. But if the second picked non-lawful and the third picked
lawful, the result would end up as none-of-the-above (a value of 3
when it needs to be a single-bit mask of 1, 2, or 4).
Reduce the implied reliance of a specific version of lua.
Instead of copying liblua.a to src/, copy it to lib/. Instead of
telling the compiler to look for headers in lib/lua-5.3.5/src/ as
well as in include/, copy the relevant ones to lib/ and tell the
compiler to look for them there. 'make spotless' in src/ will
remove both the object library and the header files from lib/ but
there really should be a new Makefile.lib to take care of that
directory.
Update Makefile.src to be able to build lua in case someone starts
with 'make all' there instead of in the top Makefile. It doesn't
duplicate the option to fetch the lua source package though.
NHinternal/../genFiles.c has been updated to mention lib/liblua.a
and lib/lua*.h as 'generated at compile time' in Files and to skip
lib/lua-* entirely if it comes across that (so not operating on a
completely clean tree). But it won't be accurate unless/until
other ports stage their lua files there instead of in src/ and
lib/lua-$(VERSION)/src/.
I haven't tried 'make depend' to see what it makes of the numerous
changes....
Developed for 3.6 but deferred to 3.7. Most of the testing was with
the earlier incarnation.
Report was that pronouns were accurate for the underlying monsters
when hallucination was describing something random, and also that the
gender prefix flag from bogusmon.txt wasn't being used. The latter
is still the case, but pronouns are now chosen at random while under
the influence of hallucination. One of the choices is plural and an
attempt is made to make the monster name and verb fit that usage.
|The homunculus picks up a wand of speed monster.
|The large cats zap themselves with a wand of speed monster!
|The blue dragon is suddenly moving faster.
There is no attempt to match gender for the singular cases; you might
get
|The succubus zaps himself [...]
or
|The incubus zaps herself [...]
Unix Makefile.utl wasn't aware of the dependency of makedefs.o on
src/mdlib.c so didn't rebuild makedefs when it should have.
Eliminate several warnings:
mdlib.c - #if inside the arguments to macro Sprintf();
nhlua.c - nhl_error() ends with a call to lua_error() which doesn't
return, but neither of them were declared that way;
nhlsel.c - because of the previous, the 'else error' case of
l_selection_ellipse() led to complaints about uninitialized
variables;
sp_lev.c - missing 'const'.
I did minimal testing which went ok, but revisiting a couple of levels
gave me un-freed memory allocated by restore.c line 1337. (I haven't
looked at that at all.)
This is similar to the helm of opposite alignment case fixed some
time ago. Deferring the setting of foo->known until an item is fully
worn (because it used to get set earlier but gave away information if
the wear operation was interrupted) didn't take into account that foo
might end up Null in various circumstances. So Boots_on() needs to
validate uarmf before setting uarmf->known in case putting on boots
of levitation while on a sink caused them to come right back off.
I put similar validation into all foo_on() just in case (as far as
I'm aware, only Boots_on() and Helmet_on() actually need that).
