[Tarantool-patches] [PATCH luajit 1/2] Prevent recording of loops with -0 step or NaN values.
Sergey Kaplun
skaplun at tarantool.org
Fri Mar 13 17:41:59 MSK 2026
Hi, Sergey!
See my answer below.
On 13.03.26, Sergey Bronnikov wrote:
> Hi, Sergey,
>
> thanks for the fixes!
>
> Sergey
>
> On 3/13/26 13:07, Sergey Kaplun wrote:
>
>
> <snipped>
>
> >>> +local function test_trace_recorded(test_payload)
> >>> + jit.flush()
> >>> + -- Reset hotcounters.
> >> nit: comment can be omitted
> > I prefer not to. There may be the question: why we don't declare this
> > parameters once? The reason is that the hotcounters may cause collisions
> > and lead to the false-positive tests failures. Should I make the comment
> > more verbose?
>
> We reset hotcounters in tests about 470 times (grep -R -B 1 "hotloop=1"
> test | wc -l) and only
>
> 15 times we add a comment like "Reset hotcounters.". You add a comment
> here but missed it in the patch
>
> "MIPS64: Avoid unaligned load in lj_vm_exit_interp.". Why we should
> leave comment here and
>
> omit it the aforementioned patch? I'll not insist removing it, just
> interesting, it is not an issue for blocking merge.
The main idea is to prevent the hotcount collisions between any other
functions that may possibly get hot. Unaligned load isn't a problem then
since we have no check for JIT semantics (no calls to `jit.util.traceinfo()`).
The same approach is vital for all checks that assume the specific trace
recording (or abortion). Hence, this comment is added in the first
place to attract the attention of the reader to these "standard lines",
which are not standard at all (since it is not done in the main chunk only
once).
Than the question is: should I make the comment more verbose and
specific?
>
> >
> <snipped>
--
Best regards,
Sergey Kaplun
More information about the Tarantool-patches
mailing list