Deutsch   English   Français   Italiano  
<vpiujl$1fvqe$2@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: olcott <polcott333@gmail.com>
Newsgroups: comp.theory
Subject: Re: DD specifies non-terminating behavior to HHH --- RECURSIVE CHAIN
 --- Saving Democracy
Date: Mon, 24 Feb 2025 17:16:04 -0600
Organization: A noiseless patient Spider
Lines: 383
Message-ID: <vpiujl$1fvqe$2@dont-email.me>
References: <vo6420$3mpmf$1@dont-email.me> <vpd7s7$3e5k$1@dont-email.me>
 <vpd8pl$3h9q$1@dont-email.me> <vpd97q$3e5k$2@dont-email.me>
 <vpdaj5$3u9g$1@dont-email.me> <vpdatp$3e5k$3@dont-email.me>
 <vpddgj$3u9g$2@dont-email.me> <vpddqm$3e5k$4@dont-email.me>
 <vpdkhv$5kr2$1@dont-email.me> <vpdks8$5ga3$1@dont-email.me>
 <vpdqc8$6bqs$1@dont-email.me> <vpe1g3$7gnd$1@dont-email.me>
 <vpflv1$j7qb$1@dont-email.me> <vpfnhm$jena$1@dont-email.me>
 <vpgded$nkbd$1@dont-email.me> <vpgdn8$nlei$1@dont-email.me>
 <vpgh33$o4p7$1@dont-email.me> <vpghkq$o82o$1@dont-email.me>
 <vpgk2q$okhu$1@dont-email.me> <vpgo94$p8he$1@dont-email.me>
 <vpgoia$p9vl$1@dont-email.me> <vpgrdl$tdkf$1@dont-email.me>
 <vpgtb3$tiun$2@dont-email.me> <vpgth7$tdkf$3@dont-email.me>
 <vpgufr$truc$1@dont-email.me> <vpguru$tdkf$4@dont-email.me>
 <vpgvcv$tuuf$1@dont-email.me> <vphr67$13hrc$1@dont-email.me>
 <vpi0rc$14kaj$1@dont-email.me> <vpi1ni$13hrc$3@dont-email.me>
 <vpio66$1euhp$1@dont-email.me> <vpipdj$1f8pm$1@dont-email.me>
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
Injection-Date: Tue, 25 Feb 2025 00:16:06 +0100 (CET)
Injection-Info: dont-email.me; posting-host="5f1bdbe2bf9fd53383ef617592ba422f";
	logging-data="1572686"; mail-complaints-to="abuse@eternal-september.org";	posting-account="U2FsdGVkX1/biFZ8VlEiuROFL/ZFIElr"
User-Agent: Mozilla Thunderbird
Cancel-Lock: sha1:7mnaxIg0qjpHKMYJvxjXA++ECvQ=
X-Antivirus: Norton (VPS 250224-8, 2/24/2025), Outbound message
Content-Language: en-US
In-Reply-To: <vpipdj$1f8pm$1@dont-email.me>
X-Antivirus-Status: Clean

On 2/24/2025 3:47 PM, dbush wrote:
> On 2/24/2025 4:26 PM, olcott wrote:
>> On 2/24/2025 9:03 AM, dbush wrote:
>>> On 2/24/2025 9:48 AM, olcott wrote:
>>>> On 2/24/2025 7:11 AM, dbush wrote:
>>>>> On 2/24/2025 12:17 AM, olcott wrote:
>>>>>> On 2/23/2025 11:08 PM, dbush wrote:
>>>>>>> On 2/24/2025 12:01 AM, olcott wrote:
>>>>>>>> On 2/23/2025 10:45 PM, dbush wrote:
>>>>>>>>> On 2/23/2025 11:42 PM, olcott wrote:
>>>>>>>>>> On 2/23/2025 10:09 PM, dbush wrote:
>>>>>>>>>>> On 2/23/2025 10:20 PM, olcott wrote:
>>>>>>>>>>>> On 2/23/2025 9:15 PM, dbush wrote:
>>>>>>>>>>>>> On 2/23/2025 9:04 PM, olcott wrote:
>>>>>>>>>>>>>> On 2/23/2025 7:22 PM, dbush wrote:
>>>>>>>>>>>>>>> On 2/23/2025 8:13 PM, olcott wrote:
>>>>>>>>>>>>>>>> On 2/23/2025 6:15 PM, dbush wrote:
>>>>>>>>>>>>>>>>> On 2/23/2025 7:10 PM, olcott wrote:
>>>>>>>>>>>>>>>>>> On 2/23/2025 11:57 AM, dbush wrote:
>>>>>>>>>>>>>>>>>>> On 2/23/2025 12:30 PM, olcott wrote:
>>>>>>>>>>>>>>>>>>>> On 2/22/2025 8:34 PM, dbush wrote:
>>>>>>>>>>>>>>>>>>>>> On 2/22/2025 7:33 PM, olcott wrote:
>>>>>>>>>>>>>>>>>>>>>> On 2/22/2025 4:59 PM, dbush wrote:
>>>>>>>>>>>>>>>>>>>>>>> On 2/22/2025 5:53 PM, olcott wrote:
>>>>>>>>>>>>>>>>>>>>>>>> On 2/22/2025 2:59 PM, dbush wrote:
>>>>>>>>>>>>>>>>>>>>>>>>> On 2/22/2025 3:53 PM, olcott wrote:
>>>>>>>>>>>>>>>>>>>>>>>>>> On 2/22/2025 2:09 PM, dbush wrote:
>>>>>>>>>>>>>>>>>>>>>>>>>>> On 2/22/2025 3:03 PM, olcott wrote:
>>>>>>>>>>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>>>>>>>>> 01 int F(int i)
>>>>>>>>>>>>>>>>>>>>>>>>>> 02 {
>>>>>>>>>>>>>>>>>>>>>>>>>> 03   if (i > 10)
>>>>>>>>>>>>>>>>>>>>>>>>>> 04     return 0;
>>>>>>>>>>>>>>>>>>>>>>>>>> 05   else
>>>>>>>>>>>>>>>>>>>>>>>>>> 06     return F(i+1);
>>>>>>>>>>>>>>>>>>>>>>>>>> 07 }
>>>>>>>>>>>>>>>>>>>>>>>>>> 08
>>>>>>>>>>>>>>>>>>>>>>>>>> 09 int no_numbers_greater_than_10()
>>>>>>>>>>>>>>>>>>>>>>>>>> 10 {
>>>>>>>>>>>>>>>>>>>>>>>>>> 11   return F(0);
>>>>>>>>>>>>>>>>>>>>>>>>>> 12 }
>>>>>>>>>>>>>>>>>>>>>>>>>> 13
>>>>>>>>>>>>>>>>>>>>>>>>>> 14 int main()
>>>>>>>>>>>>>>>>>>>>>>>>>> 15 {
>>>>>>>>>>>>>>>>>>>>>>>>>> 16   F((int)no_numbers_greater_than_10);
>>>>>>>>>>>>>>>>>>>>>>>>>> 17   return 0;
>>>>>>>>>>>>>>>>>>>>>>>>>> 18 }
>>>>>>>>>>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>>>>>>>>>> So if the address of 
>>>>>>>>>>>>>>>>>>>>>>>>>>> no_numbers_greater_than_10 is greater than 10 
>>>>>>>>>>>>>>>>>>>>>>>>>>> then 0 is returned right away, otherwise as 
>>>>>>>>>>>>>>>>>>>>>>>>>>> most 10 recursive calls will be made before 
>>>>>>>>>>>>>>>>>>>>>>>>>>> the condition is matched and 0 is returned.
>>>>>>>>>>>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>>>>>>>>>> This doesn't change the fact that 
>>>>>>>>>>>>>>>>>>>>>>>>>>> no_numbers_greater_than_10 correctly
>>>>>>>>>>>>>>>>>>>>>>>>>>> simulated by F cannot possibly return so 
>>>>>>>>>>>>>>>>>>>>>>>>>>> F(no_numbers_greater_than_10)
>>>>>>>>>>>>>>>>>>>>>>>>>>> is correct to report non-halting, which means 
>>>>>>>>>>>>>>>>>>>>>>>>>>> that there is no natural
>>>>>>>>>>>>>>>>>>>>>>>>>>> number greater than 10.
>>>>>>>>>>>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>>>>>>>>>> Agreed?
>>>>>>>>>>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>>>>>>>>> I think that you will find more bugs when you 
>>>>>>>>>>>>>>>>>>>>>>>>>> try to
>>>>>>>>>>>>>>>>>>>>>>>>>> provide the line number by line number 
>>>>>>>>>>>>>>>>>>>>>>>>>> execution trace.
>>>>>>>>>>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>>>>>>>>> #1 bug F never simulates anything.
>>>>>>>>>>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>>>>>>>> It is a verified fact that 
>>>>>>>>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>>>>>>> F never simulates anything when i > 10.
>>>>>>>>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>>>>>> Remember, you agreed that the behavior of X 
>>>>>>>>>>>>>>>>>>>>>>> simulated by Y is defined by replacing the code 
>>>>>>>>>>>>>>>>>>>>>>> of Y with an unconditional simulator and running 
>>>>>>>>>>>>>>>>>>>>>>> Y(X):
>>>>>>>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>>>>>> On 2/22/2025 1:02 PM, olcott wrote:
>>>>>>>>>>>>>>>>>>>>>>>  > On 2/22/2025 11:10 AM, dbush wrote:
>>>>>>>>>>>>>>>>>>>>>>>  >> On 2/22/2025 11:43 AM, olcott wrote:
>>>>>>>>>>>>>>>>>>>>>>>  >>> The first point is DD correctly simulated by 
>>>>>>>>>>>>>>>>>>>>>>> HHH cannot
>>>>>>>>>>>>>>>>>>>>>>>  >>> possibly terminate normally by reaching its 
>>>>>>>>>>>>>>>>>>>>>>> own "return"
>>>>>>>>>>>>>>>>>>>>>>>  >>> instruction.
>>>>>>>>>>>>>>>>>>>>>>>  >>
>>>>>>>>>>>>>>>>>>>>>>>  >> In other words, if the code of HHH is 
>>>>>>>>>>>>>>>>>>>>>>> replaced with an unconditional simulator then it 
>>>>>>>>>>>>>>>>>>>>>>> can be shown that DD is non- halting and 
>>>>>>>>>>>>>>>>>>>>>>> therefore HHH(DD)==0 is correct.
>>>>>>>>>>>>>>>>>>>>>>>  >>
>>>>>>>>>>>>>>>>>>>>>>>  >
>>>>>>>>>>>>>>>>>>>>>>>  > Wow finally someone that totally gets it.
>>>>>>>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>>>>>> So the behavior of no_numbers_greater_than_10 
>>>>>>>>>>>>>>>>>>>>>>> simulated by F is defined by replacing the code 
>>>>>>>>>>>>>>>>>>>>>>> of F with an unconditional simulated and running 
>>>>>>>>>>>>>>>>>>>>>>> F(no_numbers_greater_than_10).
>>>>>>>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>>>>>> The finite string input to F proves that there 
>>>>>>>>>>>>>>>>>>>>>>> are no instructions in no_numbers_greater_than_10 
>>>>>>>>>>>>>>>>>>>>>>> that can break the recursive simulation.
>>>>>>>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>>>>>> Try to show how no_numbers_greater_than_10 
>>>>>>>>>>>>>>>>>>>>>>> correctly simulated by F can possibly halt.
>>>>>>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>>>>> Then is ceases to be analogous to HHH(DD) because
>>>>>>>>>>>>>>>>>>>>>> no_numbers_greater_than_10() always terminates 
>>>>>>>>>>>>>>>>>>>>>> normally
>>>>>>>>>>>>>>>>>>>>>> by reaching its own "return" instruction.
>>>>>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>>>> In other words, when we actually run 
>>>>>>>>>>>>>>>>>>>>> no_numbers_greater_than_10() it reaches its own 
>>>>>>>>>>>>>>>>>>>>> "return" instruction.
>>>>>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>>>> That means we've now established that the direct 
>>>>>>>>>>>>>>>>>>>>> execution of a program (which includes all the 
>>>>>>>>>>>>>>>>>>>>> functions it calls UNMODIFIED) defines whether or 
>>>>>>>>>>>>>>>>>>>>> not it halts.
>>>>>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>>>> Likewise, when we actually run DD() unmodified it 
>>>>>>>>>>>>>>>>>>>>> also reaches its own "return" instruction.
>>>>>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>>>> Therefore HHH(DD)==0 is wrong.
>>>>>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>>> _DD()
>>>>>>>>>>>>>>>>>>>> [00002133] 55         push ebp      ; housekeeping
>>>>>>>>>>>>>>>>>>>> [00002134] 8bec       mov ebp,esp   ; housekeeping
>>>>>>>>>>>>>>>>>>>> [00002136] 51         push ecx      ; make space for 
>>>>>>>>>>>>>>>>>>>> local
>>>>>>>>>>>>>>>>>>>> [00002137] 6833210000 push 00002133 ; push DD
>>>>>>>>>>>>>>>>>>>> [0000213c] e882f4ffff call 000015c3 ; call HHH(DD)
>>>>>>>>>>>>>>>>>>>> [00002141] 83c404     add esp,+04
>>>>>>>>>>>>>>>>>>>> [00002144] 8945fc     mov [ebp-04],eax
>>>>>>>>>>>>>>>>>>>> [00002147] 837dfc00   cmp dword [ebp-04],+00
>>>>>>>>>>>>>>>>>>>> [0000214b] 7402       jz 0000214f
>>>>>>>>>>>>>>>>>>>> [0000214d] ebfe       jmp 0000214d
>>>>>>>>>>>>>>>>>>>> [0000214f] 8b45fc     mov eax,[ebp-04]
>>>>>>>>>>>>>>>>>>>> [00002152] 8be5       mov esp,ebp
>>>>>>>>>>>>>>>>>>>> [00002154] 5d         pop ebp
>>>>>>>>>>>>>>>>>>>> [00002155] c3         ret
>>>>>>>>>>>>>>>>>>>> Size in bytes:(0035) [00002155]
>>>>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>>> When DD is correctly simulated by HHH according to 
>>>>>>>>>>>>>>>>>>>> the behavior
>>>>>>>>>>>>>>>>>>>> that the above machine code specifies then the call 
>>>>>>>>>>>>>>>>>>>> from DD
>>>>>>>>>>>>>>>>>>>> to HHH(DD) cannot possibly return and this correctly 
>>>>>>>>>>>>>>>>>>>> simulated
>>>>>>>>>>>>>>>>>>>> DD cannot possibly terminate  normally by reaching 
>>>>>>>>>>>>>>>>>>>> its own machine
>>>>>>>>>>>>>>>>>>>> address 00002155.
>>>>>>>>>>>>>>>>>>>>
========== REMAINDER OF ARTICLE TRUNCATED ==========