Re: Lua 5.2.3 Tests?

[Dirk Laurie <dirk.laurie@...>]

2014-10-20 3:25 GMT+02:00 iggyvolz . <iggyvolz@gmail.com>:
> I did a little more digging... I set up Lua 5.2.3 on a friend's computer
> (Windows 64-bit) and the 5.2.2 tests passed.
>
> I also tried out the 5.2.1 tests for kicks and they failed on 5.2.3...
> meaning that the tests most certainly depend on patch version.

The idea is that each test suite should pass when run by
the corresponding Lua executable. A new minor number means
that some bugs, either in the code or in the manual, have been
fixed. In some cases,

Since the manual quite deliberately often
defines only what happens when you do the right thing, and
leaves unspecified what happens when you do something
you shouldn't. I.e. it leaves a few loose ends aka implementation
details.  On this list we spend a lot of time hassling the Lua
authors to tie up some of those, and occasionally they respond.

In particular, the table library used to be a perennial bone
of contention, because it assumes that the input table is
a proper sequence but it is not stated what happens
otherwise. Checking for that property would severely impact
speed. In 5.2.2, though, some minimal checks that don't cost
much were introduced: table.insert and table.remove made
sure that the specified index was in range. That meant that
the behaviour on part of the 5.2.1 test suite changed, making
a 5.2.2 test suite necessary.

In the case of Lua 5.2.3, the 5.2.2 test suite still passes.
But you are right, there should eventually be a 5.2.3 test suite
that is failed by 5.2.2. If you really need that now, have a look
at http://www.lua.org/bugs.html and write your own additional
test.