Re: Bug in Lua when using signed long

[Mike Pall <mikelu-0912@...>]

Luiz Henrique de Figueiredo wrote:
> > That's one *big* reminder I want to put out for the Lua authors:
> > whatever you supply as the out-of-the-box configuration *will* be
> > used by distributions. And it will stick, since they require a
> > stable ABI. So better plan ahead and work with the maintainers.
> 
> Any concrete suggestions about luaconf.h etc, Mike?

Of course it's too late for Lua 5.1 now. One issue for Lua 5.2 is
to reconsider the type of lua_Integer in the light of this
discussion. But I suggest to discuss this again during the release
process.

Another issue I have with the current luaconf.h is that it mixes
stuff that is only relevant to compiling the Lua core with stuff
that is relevant for the public API and ABI.

Also it would help to disentangle unrelated features. E.g. why
should LUA_USE_LINUX always imply the use of LUA_USE_READLINE?
As I've previously pointed out, it's not always desirable to load
readline unconditionally.

Most platform-specific features can be either auto-detected in the
headers or in the Makefile. Currently this is all mixed up.

Allowing to override or augment a couple more features via make
arguments would certainly help (like the module path).

In LJ2 I decided to strip luaconf.h to the bare minimum and do
most of the configuration in the (GNU) Makefile. Arguably the LJ2
setup is much more complicated than Lua, so you'd probably get
away with a smaller Makefile. IMHO the end result is both more
user-friendly and distro-friendly. So far I have not received any
complaints about not supporting a 'dumb' Makefile. Maybe one could
find an appropriate compromise for Lua, too.

More specific feedback about luaconf.h will obviously have to wait
until the first Lua 5.2 alpha is out.

--Mike