Re: Should old (5.3) tests pass in new GC if running in non-generational mode

[Dibyendu Majumdar <mobile@...>]

On 8 January 2018 at 12:04, Roberto Ierusalimschy
<roberto@inf.puc-rio.br> wrote:
>> Just checking - as this would be a good way to ensure no regressions
>> in the non-generational case.
>
> Would you mind putting the question in the message?
>

Apologies for that - realised after having sent the message so it was too late.

>
>> Should old (5.3) tests pass in new GC if running in
>>  non-generational mode
>
> Please read the test documentation. GC tests (and other tests) can fail
> even in the same Lua version compiled in different platforms.
>

Of course. Interestingly though in my testing the gc.lua tests
(https://github.com/lua/tests/blob/master/gc.lua) fail at two
locations:

line 502 - if I comment out lines 502,503 then this passes.

line 553 - if I disable the entire if block then the rest goes through.

You must already know this of course ... it would be good to know
whether these are expected failures. My merge gives me the same
failures so that is reassuring.

I was wondering if the tests could be amended in a way that makes the
differences obvious - i.e. have the tests bracketed by version.
Although the GC is implementation detail and hence tests in one
version may not work in another, still it is useful to understand why
something breaks or is different, mainly so that bugs can be
identified.

I have zero knowledge of how the GC works and hence right now cannot
debug issues.

Thanks and Regards
Dibyendu