[JW Lua] Jan Angermüller's Lua 5.2/5.4 issues

[Robert Patterson]

I was able to duplicate basically all reported math issues. I am noticing
some slight discrepancies between JW Lua and the production release of RGP
Lua, even though both are Lua 5.2. I suspect JW Lua is using an earlier
iteration of Lua 5.2 than RGP Lua is. There also could be a difference
between the Win and Mac compilations of Lua or the C compiler Jari used vs.
the one I'm using.

The biggest find in all of this has to do with `math.pow`. The `pow`
function is deprecated as of Lua 5.3 in favor of the ^ operator. However,
there is a compiler flag to include the deprecated math functions, and I
have now enabled that flag in the interest of maintaining maximum backward
compatibility. Kudos to Jan for finding it.

The `tonumber` function only complains about non-string arguments if you
specify a base. Since you are specifying base 10, which is the default, you
could omit the base and then the error would not occur. That is:

tonumber(12) -- no error
tonumber(12, 10) -- error

I haven't found any compiler switch to address this, and it actually looks
to me like it could be a bug. I can't see any motivation for treating one
as an error but not the other. Nor can I find any documentation declaring
the change as intentional.

Also, there doesn't appear to be any way to assign 1/0 (or 0/0) to an
integer on Lua 5.4. Considering that on Lua 5.2 it ended up as -1, that's
probably actually a benefit.

The Lua 5.4 behavior of entering the nan loop seems more correct to me than
Lua 5.2 not entering it.

Let me know if any of these are showstoppers for anyone. While it is easy
to go back, I am leaning more towards upgrading to 5.4 with the next
release. But that release is not imminent, and there is plenty of room for
more regression testing with it.

Robert
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://jwmusic.nu/pipermail/jwlua_jwmusic.nu/attachments/20230322/329cc10d/attachment.htm>