pipeline table.insert #2 argument out of bound

[suseyang]

in redis.lua file, pipeline function Line 516 request is {}, empty, so #request get 0, it's a bad argument for table.insert

[ignacio]

But the call to 'client.network.write' adds data to requests

If I'm not mistaken:

local reply = cmd(client, ...)  <--- this will end up inserting data in 'requests' table
-- so here #requests can't be zero
table_insert(parsers, #requests, reply.parser)
return reply

[suseyang]

sounds reasonable, but in particular situation, request is empty, that's a sure thing. and lua don't allow #2 argument to be 0, so an empty table can't use table.insert with #2 argument, which is a strange design i think. if Line 516 isn't wrong, then the responsibility falls to the code before it by not catching the error in advance.

[ignacio]

Do you have a test case to reproduce the error?

[nrk]

Not sure I understand what the issue is, any tiny snippet of code to reproduce it?

[benwilber]

I'm seeing this issue as well.

redis-lua
   2.0.4-1 (installed) - /usr/local/lib/luarocks/rocks

local redis = require("redis")
local r = redis.connect()

r:pipeline(function(pipe)
    pipe:zadd("FOO", 1, "BAR")
    pipe:zadd("FOO", 2, "BAZ")
end)

lua: /usr/local/share/lua/5.2/redis.lua:802: /usr/local/share/lua/5.2/redis.lua:471: bad argument #2 to 'table_insert' (position out of bounds)
stack traceback:
    [C]: in function 'error'
    /usr/local/share/lua/5.2/redis.lua:802: in function 'error'
    /usr/local/share/lua/5.2/redis.lua:480: in function 'pipeline'
    test.lua:5: in main chunk
    [C]: in ?

[janeczku]

@benwilber Did you find a fix for this?

[cedroyer]

I got the same problem, but it seems that it comes from lua version: in 5.1:

t = {}
table.insert(t, 1, nil)
print(#t) --> 0, but
table.insert(t, 2, nil) --> no error

in 5.2 and 5.3:

t = {}
table.insert(t, 1, nil)
print(#t) --> 0
table.insert(t, 2, nil) --> ERROR: bad argument #2 to 'insert' (position out of bounds)

corrected for me by changing in pipeline function:

table_insert(parsers, #requests, reply.parser)

by

parsers[#requests] = reply.parser

I'm working on a more complete patch and I submit a pull request

[janeczku]

:+1: well done @cedroyer! This seems to affect Lua >= 5.2 indeed.

[jaysoffian]

I think the correct fix is to change:

table_insert(parsers, #requests, reply.parser)

to:

table_insert(parsers, reply.parser)

This keeps the requests and parsers tables parallel, i.e. for requests[n] the correct parser is parsers[n].

[cedroyer]

I disagree, if you use: table_insert(parsers, reply.parser) and reply.parser is nil (and that seems to be the case), your table won't change and the next parser you add in parsers will be insert at the wrong place (n-1), that's why table_insert(parsers, #requests, reply.parser) was used in the library.

use parsers[#requests] = reply.parser works for me with lua 5.3 and I think it also works for lua 5.1 and 5.2.