| Deutsch English Français Italiano |
|
<vvnt66$3j0g1$7@dont-email.me> View for Bookmarking (what is this?) Look up another Usenet article |
Path: news.eternal-september.org!eternal-september.org!.POSTED!not-for-mail
From: "Fred. Zwarts" <F.Zwarts@HetNet.nl>
Newsgroups: comp.theory
Subject: Re: Try and prove that DDD correctly emulated by HHH reaches its
final halt state
Date: Sat, 10 May 2025 18:00:04 +0200
Organization: A noiseless patient Spider
Lines: 105
Message-ID: <vvnt66$3j0g1$7@dont-email.me>
References: <vvma0u$34vcu$1@dont-email.me> <vvmvu6$3dt3n$1@dont-email.me>
<vvnrkv$3in62$7@dont-email.me>
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
Injection-Date: Sat, 10 May 2025 18:00:07 +0200 (CEST)
Injection-Info: dont-email.me; posting-host="6132ef5c9a5712f5fb4e052234097a74";
logging-data="3768833"; mail-complaints-to="abuse@eternal-september.org"; posting-account="U2FsdGVkX1/R5qk/tJkPZebPjJ2IypMn"
User-Agent: Mozilla Thunderbird
Cancel-Lock: sha1:OWPHu7GAO+zP15xSWOR5INrsyIQ=
In-Reply-To: <vvnrkv$3in62$7@dont-email.me>
Content-Language: nl, en-GB
Op 10.mei.2025 om 17:33 schreef olcott:
> On 5/10/2025 2:40 AM, Mikko wrote:
>> On 2025-05-10 01:26:54 +0000, olcott said:
>>
>>> void DDD()
>>> {
>>> HHH(DDD);
>>> return;
>>> }
>>>
>>> _DDD()
>>> [00002172] 55 push ebp ; housekeeping
>>> [00002173] 8bec mov ebp,esp ; housekeeping
>>> [00002175] 6872210000 push 00002172 ; push DDD
>>> [0000217a] e853f4ffff call 000015d2 ; call HHH(DDD)
>>> [0000217f] 83c404 add esp,+04
>>> [00002182] 5d pop ebp
>>> [00002183] c3 ret
>>> Size in bytes:(0018) [00002183]
>>>
>>> Try to show how DDD emulated by HHH according to the
>>> rules of the x86 language reaches its own "ret"
>>> instruction final halt state.
>>
>> If you do the impossible then everything happens, including
>> the reaching of the final "ret" instruction. In this case
>> the impossible is a correct emulation of DDD by HHH according
>> to the rules of the x86 language.
>>
>
> It is not impossible for DDD to be emulated
> by HHH according to the rules of the x86 language.
> Here it is actually doing this.
>
> _DDD()
> [0000219e] 55 push ebp
> [0000219f] 8bec mov ebp,esp
> [000021a1] 689e210000 push 0000219e
> [000021a6] e843f4ffff call 000015ee
> [000021ab] 83c404 add esp,+04
> [000021ae] 5d pop ebp
> [000021af] c3 ret
> Size in bytes:(0018) [000021af]
>
> _main()
> [000021be] 55 push ebp
> [000021bf] 8bec mov ebp,esp
> [000021c1] 689e210000 push 0000219e
> [000021c6] e823f4ffff call 000015ee
> [000021cb] 83c404 add esp,+04
> [000021ce] 50 push eax
> [000021cf] 685f070000 push 0000075f
> [000021d4] e8a5e5ffff call 0000077e
> [000021d9] 83c408 add esp,+08
> [000021dc] 33c0 xor eax,eax
> [000021de] 5d pop ebp
> [000021df] c3 ret
> Size in bytes:(0034) [000021df]
>
> machine stack stack machine assembly
> address address data code language
> ======== ======== ======== ============== =============
> [000021be][00103872][00000000] 55 push ebp
> [000021bf][00103872][00000000] 8bec mov ebp,esp
> [000021c1][0010386e][0000219e] 689e210000 push 0000219e // push DDD
> [000021c6][0010386a][000021cb] e823f4ffff call 000015ee // call HHH
> New slave_stack at:103916
>
> Begin Local Halt Decider Simulation Execution Trace Stored at:11391e
> [0000219e][0011390e][00113912] 55 push ebp
> [0000219f][0011390e][00113912] 8bec mov ebp,esp
> [000021a1][0011390a][0000219e] 689e210000 push 0000219e // push DDD
> [000021a6][00113906][000021ab] e843f4ffff call 000015ee // call HHH
>
> The following shows the result of HHH emulating
> itself emulating DDD. x86utm is a cooperative
> multi-tasking operating system.
>
> New slave_stack at:14e33e
> [0000219e][0015e336][0015e33a] 55 push ebp
> [0000219f][0015e336][0015e33a] 8bec mov ebp,esp
> [000021a1][0015e332][0000219e] 689e210000 push 0000219e // push DDD
> [000021a6][0015e32e][000021ab] e843f4ffff call 000015ee // call HHH
> Local Halt Decider: Infinite Recursion Detected Simulation Stopped
>
> [000021cb][00103872][00000000] 83c404 add esp,+04
> [000021ce][0010386e][00000000] 50 push eax
> [000021cf][0010386a][0000075f] 685f070000 push 0000075f
> [000021d4][0010386a][0000075f] e8a5e5ffff call 0000077e
> Input_Halts = 0
> [000021d9][00103872][00000000] 83c408 add esp,+08
> [000021dc][00103872][00000000] 33c0 xor eax,eax
> [000021de][00103876][00000018] 5d pop ebp
> [000021df][0010387a][00000000] c3 ret
> Number of Instructions Executed(10069) == 150 Pages
>
>
You seems to forget what you have said before. This has been posted
several times already and each time it was shown to be irrelevant. What
you show is that the simulated HHH aborts, but that the simulating HHH
misses that verifiable fact in its simulation.
This buggy HHH is part of the input, which makes that the input
specifies a buggy halting program. That HHH is blind for the
specification does not mean that the input does not specify it.