Path: ...!eternal-september.org!feeder3.eternal-september.org!news.eternal-september.org!.POSTED!not-for-mail
From: BGB <cr88192@gmail.com>
Newsgroups: comp.lang.c
Subject: Re: transpiling to low level C
Date: Tue, 17 Dec 2024 00:40:54 -0600
Organization: A noiseless patient Spider
Lines: 76
Message-ID: <vjr6dn$1j57r$1@dont-email.me>
References: <vjlh19$8j4k$1@dont-email.me>
 <vjn9g5$n0vl$1@raubtier-asyl.eternal-september.org>
 <vjnhsq$oh1f$1@dont-email.me> <vjnq5s$pubt$1@dont-email.me>
 <vjpn29$17jub$1@dont-email.me> <vjps0l$18hon$1@dont-email.me>
 <vjq36r$19sis$1@dont-email.me> <vjqcut$1bld5$1@dont-email.me>
 <877c7z85t2.fsf@nosuchdomain.example.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Injection-Date: Tue, 17 Dec 2024 07:40:55 +0100 (CET)
Injection-Info: dont-email.me; posting-host="10e7c9ead457ab57c374f57f86cb174f";
	logging-data="1676539"; mail-complaints-to="abuse@eternal-september.org";	posting-account="U2FsdGVkX1/4k9YdvqqRcWBlKnNvA25jyIC9/OD00WQ="
User-Agent: Mozilla Thunderbird
Cancel-Lock: sha1:Y/6c9XiSlP0kBrUf/flqSc38rzE=
Content-Language: en-US
In-Reply-To: <877c7z85t2.fsf@nosuchdomain.example.com>
Bytes: 3765

On 12/16/2024 7:19 PM, Keith Thompson wrote:
> bart <bc@freeuk.com> writes:
> [SNIP]
>> In that case I've no idea what you were trying to say.
>>
>> When somebody says that 'goto' can emulate any control structure, then
>> clearly some of them need to be conditional; that is implied.
>>
>> Your reply suggested they you can do away with 'goto', and use
>> recursive functions, in a scenario where no other control structures
>> need exist.
>>
>> OK, if this is not for an IL, then it's not a language I would care
>> for either. Why tie one hand behind your back for no good reason?
> 
> I read Janis's post.  I saw a suggestion that certain constructs are
> *theoretically* unnecessary.  I saw no suggestion of any advocacy for
> such an approach.
> 
> """
> A 'goto' may be used but it isn't strictly *necessary*. What *is*
> necessary, though, that is an 'if' (some conditional branch), and
> either 'goto' or recursive functions.
> """
> 

While if-call is technically possible, it is likely to result in a 
significantly higher performance overhead if compared with if-goto.


Say:
   if-goto:
     can be a single CPU instruction on some targets.
     Both RISC-V and BJX2 can do single-instruction compare-and-branch.
   if-call:
     A whole lot more than 1 instruction...
     One may have call/return, stack frame creation, ...
       Or, a backend that is very clever about inlining.


Also: if-goto can readily express pretty much every other intra-function 
control flow construct:
   for, while, switch, break, continue, ...
And, is less awkward for a compiler than trying to map control flow to 
any of the other constructs. Also it is generally the fastest construct 
for implementing these higher level constructs.

Meanwhile, if-call, good luck... You are also going to need something 
beyond normal C style variable scoping to make this work acceptably. At 
best, it is likely to be very awkward and likely to perform poorly.


Though, it is possible that one could make a case for a 3-way branch 
operator in the IR, say:
   if3(x,y) A, B, C
Which does the equivalent of, say:
   if(x<y)goto A;
   else if(x>y)goto B;
   else goto C;

This would be a bit niche, but could be useful for things like 
implementing "switch()" via binary subdivision.

Though, functionally could decay into normal if-goto in various cases.


....


> [...]
> 
>> So what was your proposal about?
> 
> I saw no proposal.
>