[Tarantool-patches] [PATCH luajit] test: adapt test checking reversed function

[Tarantool-patches] [PATCH luajit] test: adapt test checking reversed function

[Maxim Kokryashkin via Tarantool-patches]

The first fiber in Tarantool has only 512Kb of the stack which is not enough to
handle such a deep call chain.
The test is adapted to Tarantool by decreasing the string length.

Closes tarantool/tarantool#5782
Part of tarantool/tarantool#5845
Part of tarantool/tarantool#4473
---
GitHub branch: https://github.com/tarantool/luajit/tree/fckxorg/gh-5782-adapt-deep-nest-gsub-PUC-Rio
Issue: https://github.com/tarantool/tarantool/issues/5782

 test/PUC-Rio-Lua-5.1-tests/pm.lua | 9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)

diff --git a/test/PUC-Rio-Lua-5.1-tests/pm.lua b/test/PUC-Rio-Lua-5.1-tests/pm.lua
index e364ff9d..7da3ef4a 100644
--- a/test/PUC-Rio-Lua-5.1-tests/pm.lua
+++ b/test/PUC-Rio-Lua-5.1-tests/pm.lua
@@ -206,12 +206,11 @@ function rev (s)
   return string.gsub(s, "(.)(.+)", function (c,s1) return rev(s1)..c end)
 end
 
-local x = string.rep('012345', 10)
--- FIXME: The first Tarantool's fiber has only 512Kb of stack.
--- It is not enough for this recursive call.
+-- This test is adapted to match the stack size (512Kb) of the first fiber in
+-- Tarantool.
 -- See also https://github.com/tarantool/tarantool/issues/5782.
--- The test is disabled for Tarantool binary.
--- assert(rev(rev(x)) == x)
+local x = string.rep('01234', 10)
+assert(rev(rev(x)) == x)
 
 
 -- gsub with tables
-- 
2.33.0

Re: [Tarantool-patches] [PATCH luajit] test: adapt test checking reversed function

[Sergey Kaplun via Tarantool-patches]

Hi, Maxim!

Thanks for the patch!

Please consider my comments below.

On 30.09.21, Maxim Kokryashkin wrote:
> The first fiber in Tarantool has only 512Kb of the stack which is not enough to
> handle such a deep call chain.
> The test is adapted to Tarantool by decreasing the string length.
> 
> Closes tarantool/tarantool#5782
> Part of tarantool/tarantool#5845
> Part of tarantool/tarantool#4473

Looks like it should be 5870 instead 4473. Also, 5845 is already
closed.

> ---

Please show the Tarantool branch as well, to show that problem is gone.

> GitHub branch: https://github.com/tarantool/luajit/tree/fckxorg/gh-5782-adapt-deep-nest-gsub-PUC-Rio
> Issue: https://github.com/tarantool/tarantool/issues/5782
> 
>  test/PUC-Rio-Lua-5.1-tests/pm.lua | 9 ++++-----
>  1 file changed, 4 insertions(+), 5 deletions(-)
> 
> diff --git a/test/PUC-Rio-Lua-5.1-tests/pm.lua b/test/PUC-Rio-Lua-5.1-tests/pm.lua
> index e364ff9d..7da3ef4a 100644
> --- a/test/PUC-Rio-Lua-5.1-tests/pm.lua
> +++ b/test/PUC-Rio-Lua-5.1-tests/pm.lua
> @@ -206,12 +206,11 @@ function rev (s)
>    return string.gsub(s, "(.)(.+)", function (c,s1) return rev(s1)..c end)
>  end
>  
> -local x = string.rep('012345', 10)
> --- FIXME: The first Tarantool's fiber has only 512Kb of stack.
> --- It is not enough for this recursive call.
> +-- This test is adapted to match the stack size (512Kb) of the first fiber in
> +-- Tarantool.
>  -- See also https://github.com/tarantool/tarantool/issues/5782.
> --- The test is disabled for Tarantool binary.
> --- assert(rev(rev(x)) == x)
> +local x = string.rep('01234', 10)
> +assert(rev(rev(x)) == x)

The patch is looks OK to me, but the main problem is still here:
LuaJIT is badly managing C stack overflow. The same chunk rases an error
for Lua 5.1, but crashes for LuaJIT.

| $ luajit -e 'local function rev (s) return string.gsub(s, "(.)(.+)", function (c,s1) return rev(s1)..c end) end local x = string.rep("0", 1000) rev(x)'
| Segmentation fault

| $ lua    -e 'local function rev (s) return string.gsub(s, "(.)(.+)", function (c,s1) return rev(s1)..c end) end local x = string.rep("0", 1000) rev(x)'
| lua: C stack overflow
| stack traceback:
|         [C]: in function 'gsub'
|         (command line):1: in function 'rev'
|         (command line):1: in function <(command line):1>
|         [C]: in function 'gsub'
|         (command line):1: in function 'rev'
|         (command line):1: in function <(command line):1>
|         [C]: in function 'gsub'
|         (command line):1: in function 'rev'
|         (command line):1: in function <(command line):1>
|         [C]: in function 'gsub'
|         ...
|         [C]: in function 'gsub'
|         (command line):1: in function 'rev'
|         (command line):1: in function <(command line):1>
|         [C]: in function 'gsub'
|         (command line):1: in function 'rev'
|         (command line):1: in function <(command line):1>
|         [C]: in function 'gsub'
|         (command line):1: in function 'rev'
|         (command line):1: in main chunk
|         [C]: ?

So I suppose it is strange to add test that tests nothing.

Thoughts?

>  
>  
>  -- gsub with tables
> -- 
> 2.33.0
> 

-- 
Best regards,
Sergey Kaplun

Re: [Tarantool-patches] [PATCH luajit] test: adapt test checking reversed function

[Максим Корякшин via Tarantool-patches]

[-- Attachment #1: Type: text/plain, Size: 3395 bytes --]


Hi! Thanks for the comments. 
 
I agree that it is strange to add this test, as long as it does not test
anything, so I propose to remove it.
 
Best regards,
Maxim Kokryashkin
 
> 
>>Hi, Maxim!
>>
>>Thanks for the patch!
>>
>>Please consider my comments below.
>>
>>On 30.09.21, Maxim Kokryashkin wrote:
>>> The first fiber in Tarantool has only 512Kb of the stack which is not enough to
>>> handle such a deep call chain.
>>> The test is adapted to Tarantool by decreasing the string length.
>>>
>>> Closes tarantool/tarantool#5782
>>> Part of tarantool/tarantool#5845
>>> Part of tarantool/tarantool#4473
>>
>>Looks like it should be 5870 instead 4473. Also, 5845 is already
>>closed.
>>
>>> ---
>>
>>Please show the Tarantool branch as well, to show that problem is gone.
>>
>>> GitHub branch:  https://github.com/tarantool/luajit/tree/fckxorg/gh-5782-adapt-deep-nest-gsub-PUC-Rio
>>> Issue:  https://github.com/tarantool/tarantool/issues/5782
>>>
>>> test/PUC-Rio-Lua-5.1-tests/pm.lua | 9 ++++-----
>>> 1 file changed, 4 insertions(+), 5 deletions(-)
>>>
>>> diff --git a/test/PUC-Rio-Lua-5.1-tests/pm.lua b/test/PUC-Rio-Lua-5.1-tests/pm.lua
>>> index e364ff9d..7da3ef4a 100644
>>> --- a/test/PUC-Rio-Lua-5.1-tests/pm.lua
>>> +++ b/test/PUC-Rio-Lua-5.1-tests/pm.lua
>>> @@ -206,12 +206,11 @@ function rev (s)
>>> return string.gsub(s, "(.)(.+)", function (c,s1) return rev(s1)..c end)
>>> end
>>>
>>> -local x = string.rep('012345', 10)
>>> --- FIXME: The first Tarantool's fiber has only 512Kb of stack.
>>> --- It is not enough for this recursive call.
>>> +-- This test is adapted to match the stack size (512Kb) of the first fiber in
>>> +-- Tarantool.
>>> -- See also  https://github.com/tarantool/tarantool/issues/5782 .
>>> --- The test is disabled for Tarantool binary.
>>> --- assert(rev(rev(x)) == x)
>>> +local x = string.rep('01234', 10)
>>> +assert(rev(rev(x)) == x)
>>
>>The patch is looks OK to me, but the main problem is still here:
>>LuaJIT is badly managing C stack overflow. The same chunk rases an error
>>for Lua 5.1, but crashes for LuaJIT.
>>
>>| $ luajit -e 'local function rev (s) return string.gsub(s, "(.)(.+)", function (c,s1) return rev(s1)..c end) end local x = string.rep("0", 1000) rev(x)'
>>| Segmentation fault
>>
>>| $ lua -e 'local function rev (s) return string.gsub(s, "(.)(.+)", function (c,s1) return rev(s1)..c end) end local x = string.rep("0", 1000) rev(x)'
>>| lua: C stack overflow
>>| stack traceback:
>>| [C]: in function 'gsub'
>>| (command line):1: in function 'rev'
>>| (command line):1: in function <(command line):1>
>>| [C]: in function 'gsub'
>>| (command line):1: in function 'rev'
>>| (command line):1: in function <(command line):1>
>>| [C]: in function 'gsub'
>>| (command line):1: in function 'rev'
>>| (command line):1: in function <(command line):1>
>>| [C]: in function 'gsub'
>>| ...
>>| [C]: in function 'gsub'
>>| (command line):1: in function 'rev'
>>| (command line):1: in function <(command line):1>
>>| [C]: in function 'gsub'
>>| (command line):1: in function 'rev'
>>| (command line):1: in function <(command line):1>
>>| [C]: in function 'gsub'
>>| (command line):1: in function 'rev'
>>| (command line):1: in main chunk
>>| [C]: ?
>>
>>So I suppose it is strange to add test that tests nothing.
>>
>>Thoughts?
>>
>>>
>>>
>>> -- gsub with tables
>>> --
>>> 2.33.0
>>>
>>
>>--
>>Best regards,
>>Sergey Kaplun
> 

[-- Attachment #2: Type: text/html, Size: 4370 bytes --]

Re: [Tarantool-patches] [PATCH luajit] test: adapt test checking reversed function

[Sergey Kaplun via Tarantool-patches]

Hi!

Thanks for the answer!

On 11.10.21, Максим Корякшин wrote:
> 
> Hi! Thanks for the comments. 
>  
> I agree that it is strange to add this test, as long as it does not test
> anything, so I propose to remove it.

May be as an option, but IMO it is better to leave it commented instead,
as a reminder about the problem. Lets wait for the second reviewer here.
I'm OK with both solutions.

>  
> Best regards,
> Maxim Kokryashkin
>  
> > 
> >>Hi, Maxim!
> >>
> >>Thanks for the patch!
> >>
> >>Please consider my comments below.
> >>
> >>On 30.09.21, Maxim Kokryashkin wrote:
> >>> The first fiber in Tarantool has only 512Kb of the stack which is not enough to
> >>> handle such a deep call chain.
> >>> The test is adapted to Tarantool by decreasing the string length.
> >>>
> >>> Closes tarantool/tarantool#5782
> >>> Part of tarantool/tarantool#5845
> >>> Part of tarantool/tarantool#4473
> >>
> >>Looks like it should be 5870 instead 4473. Also, 5845 is already
> >>closed.
> >>
> >>> ---
> >>
> >>Please show the Tarantool branch as well, to show that problem is gone.
> >>
> >>> GitHub branch:  https://github.com/tarantool/luajit/tree/fckxorg/gh-5782-adapt-deep-nest-gsub-PUC-Rio
> >>> Issue:  https://github.com/tarantool/tarantool/issues/5782
> >>>
> >>> test/PUC-Rio-Lua-5.1-tests/pm.lua | 9 ++++-----
> >>> 1 file changed, 4 insertions(+), 5 deletions(-)
> >>>
> >>> diff --git a/test/PUC-Rio-Lua-5.1-tests/pm.lua b/test/PUC-Rio-Lua-5.1-tests/pm.lua
> >>> index e364ff9d..7da3ef4a 100644
> >>> --- a/test/PUC-Rio-Lua-5.1-tests/pm.lua
> >>> +++ b/test/PUC-Rio-Lua-5.1-tests/pm.lua
> >>> @@ -206,12 +206,11 @@ function rev (s)
> >>> return string.gsub(s, "(.)(.+)", function (c,s1) return rev(s1)..c end)
> >>> end
> >>>
> >>> -local x = string.rep('012345', 10)
> >>> --- FIXME: The first Tarantool's fiber has only 512Kb of stack.
> >>> --- It is not enough for this recursive call.
> >>> +-- This test is adapted to match the stack size (512Kb) of the first fiber in
> >>> +-- Tarantool.
> >>> -- See also  https://github.com/tarantool/tarantool/issues/5782 .
> >>> --- The test is disabled for Tarantool binary.
> >>> --- assert(rev(rev(x)) == x)
> >>> +local x = string.rep('01234', 10)
> >>> +assert(rev(rev(x)) == x)
> >>
> >>The patch is looks OK to me, but the main problem is still here:
> >>LuaJIT is badly managing C stack overflow. The same chunk rases an error
> >>for Lua 5.1, but crashes for LuaJIT.
> >>
> >>| $ luajit -e 'local function rev (s) return string.gsub(s, "(.)(.+)", function (c,s1) return rev(s1)..c end) end local x = string.rep("0", 1000) rev(x)'
> >>| Segmentation fault
> >>
> >>| $ lua -e 'local function rev (s) return string.gsub(s, "(.)(.+)", function (c,s1) return rev(s1)..c end) end local x = string.rep("0", 1000) rev(x)'
> >>| lua: C stack overflow
> >>| stack traceback:
> >>| [C]: in function 'gsub'
> >>| (command line):1: in function 'rev'
> >>| (command line):1: in function <(command line):1>
> >>| [C]: in function 'gsub'
> >>| (command line):1: in function 'rev'
> >>| (command line):1: in function <(command line):1>
> >>| [C]: in function 'gsub'
> >>| (command line):1: in function 'rev'
> >>| (command line):1: in function <(command line):1>
> >>| [C]: in function 'gsub'
> >>| ...
> >>| [C]: in function 'gsub'
> >>| (command line):1: in function 'rev'
> >>| (command line):1: in function <(command line):1>
> >>| [C]: in function 'gsub'
> >>| (command line):1: in function 'rev'
> >>| (command line):1: in function <(command line):1>
> >>| [C]: in function 'gsub'
> >>| (command line):1: in function 'rev'
> >>| (command line):1: in main chunk
> >>| [C]: ?
> >>
> >>So I suppose it is strange to add test that tests nothing.
> >>
> >>Thoughts?
> >>
> >>>
> >>>
> >>> -- gsub with tables
> >>> --
> >>> 2.33.0
> >>>
> >>
> >>--
> >>Best regards,
> >>Sergey Kaplun
> > 

-- 
Best regards,
Sergey Kaplun

Re: [Tarantool-patches] [PATCH luajit] test: adapt test checking reversed function

[Igor Munkin via Tarantool-patches]

Max,

Thanks for the patch!

On 11.10.21, Максим Корякшин wrote:
> 
> Hi! Thanks for the comments. 
>  
> I agree that it is strange to add this test, as long as it does not test
> anything, so I propose to remove it.

Why do you think this test checks nothing? It checks <string.gsub> with
recursive handler. If author wanted to check deep recursion exception,
he would add <pcall> at least (if Lua successfully handles this).

I propose to apply the original fix[1], but adjust the comment
considering Sergey review: it provides the description, why we can't use
the original test, or at least add <pcall> to avoid the crash.

>  
> Best regards,
> Maxim Kokryashkin
>  
> > 
> >>Hi, Maxim!
> >>
> >>Thanks for the patch!
> >>
> >>Please consider my comments below.
> >>
> >>On 30.09.21, Maxim Kokryashkin wrote:
> >>> The first fiber in Tarantool has only 512Kb of the stack which is not enough to
> >>> handle such a deep call chain.
> >>> The test is adapted to Tarantool by decreasing the string length.
> >>>
> >>> Closes tarantool/tarantool#5782

Typo: s/Closes/Resolves/.

> >>> Part of tarantool/tarantool#5845
> >>> Part of tarantool/tarantool#4473

Minor: Replace both tags above with "Part of tarantool/tarantool/5870".

> >>
> >>Looks like it should be 5870 instead 4473. Also, 5845 is already
> >>closed.
> >>
> >>> ---
> >>
> >>Please show the Tarantool branch as well, to show that problem is gone.
> >>
> >>> GitHub branch:  https://github.com/tarantool/luajit/tree/fckxorg/gh-5782-adapt-deep-nest-gsub-PUC-Rio
> >>> Issue:  https://github.com/tarantool/tarantool/issues/5782
> >>>
> >>> test/PUC-Rio-Lua-5.1-tests/pm.lua | 9 ++++-----
> >>> 1 file changed, 4 insertions(+), 5 deletions(-)
> >>>

<snipped>

> >>> --
> >>> 2.33.0
> >>>
> >>
> >>--
> >>Best regards,
> >>Sergey Kaplun
> > 

[1]: https://lists.tarantool.org/tarantool-patches/20210930114337.1575120-1-m.kokryashkin@tarantool.org/

-- 
Best regards,
IM