[Tarantool-patches] [PATCH luajit v1] tools: fix luajit-gdb stack dump

[Igor Munkin]

Sergey,

Thanks for the fixes! Everything is perfect now, I'll ask Sergos to
proceed with the review.

On 21.07.21, Sergey Kaplun wrote:
> Igor,
> 
> On 20.07.21, Igor Munkin wrote:
> > Sergey,
> > 
> > Thanks for the patch! Nice catch, actually! I wonder, how much impact
> > this bug made while digging the crash reports. Anyway, LGTM, but
> > consider several nits below.
> 
> Actually, not too much. It is the unique situation, when we have naked
> guest Lua stack without any function call.

Hm, I guess this bug affects all the cases when dummy frame was used,
but you know it better anyway :)

> 
> Branch is force-pushed with the changes below.
> 
> > 
> > On 18.06.21, Sergey Kaplun wrote:

<snipped>

> 
> Thanks for your comments!
> The new commit message version:
> 
> | Dummy frame is the "initial" coroutine state, when the framelink slot (i.e.
> | L->base - (1 + LJ_FR2)) is the bottom slot of the guest stack
> | (i.e. L->stack).  Since coroutine stack unwinding is implemented via
> | precondition loop, lj-stack doesn't dump the slots for the dummy frame, since
> | the framelink points to the stack bottom.
> |
> | The output looks like the following:
> |
> | | 0x7fb512ac40:0x7fb512ac70 [    ] 7 slots: Red zone
> | | 0x7fb512ac38              [   M]
> | | 0x7fb512ab28:0x7fb512ac30 [    ] 34 slots: Free stack slots
> | | 0x7fb512ab20              [  T ]
> | | 0x7fb512ab08:0x7fb512ab10 [S   ] FRAME: dummy L
> |
> | Python doesn't provide post-condition (do-while) syntax construction,
> | that fits better for this case, so the unwinding of the topmost frame is just manually
> | unrolled.
> |
> | As a result of the patch the output looks like the following:
> |
> | | 0x7fb512ac40:0x7fb512ac70 [    ] 7 slots: Red zone
> | | 0x7fb512ac38              [   M]
> | | 0x7fb512ab28:0x7fb512ac30 [    ] 34 slots: Free stack slots
> | | 0x7fb512ab20              [  T ]
> | | 0x7fb512ab18              [    ] VALUE: string 0 "/tmp/net_box.lua:6: err in ser" @ 0x7fb512ade8
> | | 0x7fb512ab10              [ B  ] VALUE: table @ 0x7fb512ac80 (asize: 0, hmask: 0x0)
> | | 0x7fb512ab00:0x7fb512ab08 [S   ] FRAME: dummy L

Side note: I guess this is a previous version, since everything is fine
on the branch. BTW, wording is neat!

> 

<snipped>

> > > diff --git a/src/luajit-gdb.py b/src/luajit-gdb.py
> > > index f1fd6230..a3dad585 100644
> > > --- a/src/luajit-gdb.py
> > > +++ b/src/luajit-gdb.py
> > > @@ -424,16 +424,26 @@ def dump_stack(L, base=None, top=None):

<snipped>

> 
> > 
> > >      while framelink > mref('TValue *', L['stack']):
> > > -        while slot > framelink + LJ_FR2:
> > > -            dump += dump_stack_slot(L, slot, base, top)
> > > -            slot -= 1
> > > +        assert slot == framelink + LJ_FR2, "Invalid slot during frame unwind"
> > 
> > Please, use parentheses for assert call.
> 
> It's done by design to avoid always true assertions [1]:
> 
> | $ python3
> | Python 3.7.10 (default, Mar 18 2021, 04:43:06)
> | [GCC 9.3.0] on linux
> | Type "help", "copyright", "credits" or "license" for more information.
> | >>> assert(1!=1, "stupid assert")
> | <stdin>:1: SyntaxWarning: assertion is always true, perhaps remove parentheses?

PEBKAC, didn't expect such a surprise here. I looked on the <assert>
usage below, but those parentheses are just excess. Everything is fine
here then, and thanks for clarification!

> 

<snipped>

> 
> [1]: https://docs.python.org/3/reference/simple_stmts.html#grammar-token-assert-stmt
> 
> -- 
> Best regards,
> Sergey Kaplun

-- 
Best regards,
IM