Re: The Lua lexer's read_numeral() function and corresponding reference documentation

[Hugo Musso Gualandi <hgualandi@...>]

Em sex, 10 de set de 2021 às 17:58:17 -0700, Parke 
<parke.nexus@gmail.com> escreveu:
> The comment preceding the read_numeral() function in version 5.4.3 of
> Lua's lexer says:
>
> >   Roughly, [read_numeral()] accepts the following pattern:
> >
> >   %d(%x|%.|([Ee][+-]?))* | 0[Xx](%x|%.|([Pp][+-]?))*
>
> The above pattern starts with "%d(%x".
>
> Why the %x?

I suspect that one reason is that the implementation is a bit simpler. 
There is a single "%x|'.'" loop that covers both halves of the regex.

The only time that this difference could possibly matter is if a number 
is touching a letter. If we write a=1b=2, should that be parsed as a=1 
followed by b=2, or should the 1b be an "invalid number" error?

In Lua 5.3 or earlier, it depends on whether the letter is a hex digit. 
a=1b=2 is an invalid number error, but x=1y=2 is allowed. If the regex 
used %d in that place you asked, then a=1b=2 would become valid as well.

In Lua 5.4, numbers touching a letter always result in an invalid 
number error. It doesn't matter if the regex is %d or %x, or if the 
letter is a hex letter. If you ask me, I prefer this behavior instead 
of the 5.3 one.

-- Hugo