Problem with lpeg 0.12

[catwell]

Cosmo is broken with lpeg 0.12. This stacktrace use LuaJIT but the problem also happens with PUC Lua.

loop body may accept empty string
stack traceback:
    [C]: at 0x7fa6d89eca20
    [C]: in function 'match'
    /usr/local/share/lua/5.1/re.lua:208: in function 'new'
    /usr/local/share/lua/5.1/cosmo/grammar.lua:166: in main chunk
    [C]: in function 'require'
    /usr/local/share/lua/5.1/cosmo.lua:3: in main chunk
    [C]: in function 'require'
    stdin:1: in main chunk
    [C]: at 0x00405010

[h4rm]

I got the same issue here. Looked into the re.lua and it is something with this piece of code, starting at line 170 - 183 in re.lua:

Suffix = m.Cf(m.V"Primary" * S *
          ( ( m.P"+" * m.Cc(1, mt.__pow)
            + m.P"*" * m.Cc(0, mt.__pow)
            + m.P"?" * m.Cc(-1, mt.__pow)
            + "^" * ( m.Cg(num * m.Cc(mult))
                    + m.Cg(m.C(m.S"+-" * m.R"09"^1) * m.Cc(mt.__pow))
                    )
            + "->" * S * ( m.Cg((String + num) * m.Cc(mt.__div))
                         + m.P"{}" * m.Cc(nil, m.Ct)
                         + m.Cg(Def / getdef * m.Cc(mt.__div))
                         )
            + "=>" * S * m.Cg(Def / getdef * m.Cc(m.Cmt))
            ) * S
          )^0

[catwell]

I asked about this on the Lua mailing list. Here's what Tom Harris answered:

The problem is the construction

local longstring = #("[" * lpeg.S"[=") * (longstring1 + longstring2)

longstring1 is fine, but longstring2 is a match-time-capture that consumes an
unknown amount of input. Although the function will match a long string or
fail, the compiler doesn't know that and assumes it will match with no input.

I changed Cosmo's longstring pattern to this from the LPEG manual

local ls_equals = lpeg.P'='^0
local ls_open = '[' * lpeg.Cg(ls_equals, "init") * '['
local ls_close = ']' * lpeg.C(ls_equals) * ']'
local ls_closeeq = lpeg.Cmt(ls_close * lpeg.Cb("init"),
      function (s, i, a, b)
        return a == b
      end)
longstring = ls_open * (lpeg.P(1) - ls_closeeq)^0 * ls_close / 1

And deleted the old longstring, longstring1, longstring2, start_ls, and start.
It compiles without error, though I haven't tested if it still matches
correctly.

Then Roberto Ierusalimschy wrote:

A less intrusive change is to change only longstring2 itself:

local longstring2 = lpeg.Cmt(start, function (s, i, start)
  local p = string.gsub(start, "%[", "]")
  local _, e = string.find(s, p, i)
  return (e and e + 1)
end)

(Many thanks to Tom for doing the hard part.)

I will test his solution and submit a pull request asap.