Re: Undocumented LPeg behavior with table and group captures and non-string names

[Daurnimator <quae@...>]

On 20 July 2015 at 19:13, Daurnimator <quae@daurnimator.com> wrote:
> On 20 July 2015 at 18:07, Dirk Laurie <dirk.laurie@gmail.com> wrote:
>> 2015-07-19 23:40 GMT+02:00 Parke <parke.nexus@gmail.com>:
>>
>>> If a named group capture has a name that is a number (as opposed to a
>>> string), then when that named group is captured inside a table
>>> capture, the key used to store the value will be the number converted
>>> to a string (rather than the number itself).
>>
>> The second parameter to Cg is retrieved using luaL_checkstring.
>> This function does number-to-string coercion.
>>
>> The key is called a "name" in the documentation, which gives
>> a light hint that its type is expected to be string.
>>
>
> Reading through the code I can't seem to seem to see any reason for
> that checkstring call to be there.
> Removing it, all tests still pass, and this works:
>
> local lpeg = require "lpeg"
> local t = {}
> local p = lpeg.Ct(lpeg.Cg(lpeg.P("foo"), t))
> local r = p:match("foo")
> assert(next(r) == t)
>
> Roberto, you said you were making a new lpeg release soon; perhaps you
> can remove this un-required luaL_checkstring?

For the sake of the archives: this change was made in the lpeg 1.0 release.