SICP chapter 4 implements begin as a sequence and does not allow empty sequences. But most Schemes allow an empty begin at least at top level.

For example "(begin)" is ok for Chez, but "(display (begin))" is not. Chibi does not care and displays undef.

Why is there so much inconsistency? What is the benefit?

ceving schrieb am Montag, 30. Oktober 2023 um 14:07:19 UTC+1:
> SICP chapter 4 implements begin as a sequence and does not allow empty sequences. But most Schemes allow an empty begin at least at top level.
>
> For example "(begin)" is ok for Chez, but "(display (begin))" is not. Chibi does not care and displays undef.
>
> Why is there so much inconsistency? What is the benefit?

Because there are at least two different `begin's. At the top-level and in definition context, `begin' is a splicing form and just splices its contents into the surrounding body. In expression context, `begin' is an expression and thus have to contain at least one expression because otherwise it would be unclear which values (if any) to deliver.

Marc Nieper-Wißkirchen schrieb am Montag, 30. Oktober 2023 um 15:05:05 UTC+1:
> Because there are at least two different `begin's.

Wow: one symbol with two different meanings. This sounds broken.

ceving schrieb am Montag, 30. Oktober 2023 um 16:50:49 UTC+1:
> Marc Nieper-Wißkirchen schrieb am Montag, 30. Oktober 2023 um 15:05:05 UTC+1:
> > Because there are at least two different `begin's.
> Wow: one symbol with two different meanings. This sounds broken.

Only if confusion or ambiguity could arise.

Scheme is not a context-free language.

Marc Nieper-Wißkirchen schrieb am Montag, 30. Oktober 2023 um 16:53:25 UTC+1:
>
> Scheme is not a context-free language.

Which one is wrong?

1.) Scheme's syntax is given in a Backus-Naur form.
2.) The Backus-Naur form defines context-free grammars.
3.) Context-free grammars define context-free languages.

ceving schrieb am Dienstag, 31. Oktober 2023 um 10:38:46 UTC+1:
> Marc Nieper-Wißkirchen schrieb am Montag, 30. Oktober 2023 um 16:53:25 UTC+1:
> >
> > Scheme is not a context-free language.
> Which one is wrong?
>
> 1.) Scheme's syntax is given in a Backus-Naur form.
> 2.) The Backus-Naur form defines context-free grammars.
> 3.) Context-free grammars define context-free languages.

Scheme's datum syntax is specified by BNF. But Scheme is more than its datum syntax. The same is true for most other programming languages; when a programming language's syntax is specified by BNF, the set of valid programs is usually a proper subset of the set of all syntactically valid programs.

See, for example, "Context Constraints in Wren" on page 13 of https://homepage.divms.uiowa.edu/~slonnegr/plf/Book/Chapter1.pdf for some other background.

On 30/10/2023 15:50, ceving wrote:
> Marc Nieper-Wißkirchen schrieb am Montag, 30. Oktober 2023 um 15:05:05 UTC+1:
>> Because there are at least two different `begin's.
>
> Wow: one symbol with two different meanings. This sounds broken.

I think it's because (lambda...) is a way to construct a value, and
(begin...) is a way to sequence actions.

The term 'lambda' as it applies to logic was described by a group of
logicians working on systems in general that might not have an atom for
*unspecified* so helpfully in scheme if you are to use lambda and return
*unspecified* you must do so by explicitly defining the value of the
lambda in term of something that may be *unspecified*. It's possible the
scheme language designers didn't appreciate the meaning in logic but
just followed habits of their forebears.

Some languages incorporate a lambda more wrongly and allow an implicit
*unspecified* because they're actually incorporating a stateful monad
because they followed *fewer* habits of their forebears.

cf. imperative vs pure functional languages.

On 30/10/2023 15:50, ceving wrote:
> Marc Nieper-Wißkirchen schrieb am Montag, 30. Oktober 2023 um 15:05:05 UTC+1:
>> Because there are at least two different `begin's.
>
> Wow: one symbol with two different meanings. This sounds broken.

Sorry for my other post whose context should have been an earlier
question but which is missing from my newsserver.

For this question, I did an experiment, and it looks like the special
meaning is only for a specific local context, in a define with a
subsequent (values...) expression.

This plurality is caused by the surrounding context and can be expressed
with a single meaning for 'begin' albeit a more complex one than you
would consider in order to not be broken. In particular the behaviour of
'if' that you seemed concerned about is to support programming styles
that suit the simpler meaning of 'begin' that you seem to want.

I am interested to know of programming languages that have no symbols
with meanings that can be expressed such that two different effects can
result from its use in two different contexts, and I don't mean things
that they refer to as symbols, but things which are actually symbols in
general.

I suspect there's no suitable language that satisfies both concerns, in
particular I think elimination of actions must allow the complex
specification of 'begin' in order to also be generally useful, yet
allowing only actions so that symbols can only have a simpler meaning
than has 'begin' will make a language too limited to serve scheme
programmers' purposes.

Perhaps it is the existence of an explicit syntax-checking threshold and
that (begin ...) causes either syntax pass or syntax fail depending on
context? This is common, and often especially desirable, where humans
will use the language.

Personally I don't see why begin shouldn't have always required either
no action and its return value been *unspecified* or always required at
least one action even if it were the nil action, *unspecified*. The
justification for requiring at least one action given in the guile help
is outside the philosophy that allows (if a b) to exist instead of (if a
b *unspecified*), which is that the language is not pure functional so
expressions can have some kind of *unspecified* atom for their results.

On Sat, 2 Mar 2024 13:18:01 +0000, Tristan Wibberley wrote:

> The term 'lambda' as it applies to logic was described by a group of
> logicians ...

λ-calculus was first created by Alonzo Church. This was around the same
time Alan Turing formulated his “Turing machine” to represent computation.
The two theories turned out to be equally capable in this regard.

λ-calculus also offers an interesting way to deal with paradoxes, of the
kind that scare mathematicians so much: it represents them as an endless
loop. This is something we are well familiar with in computer science, and
it’s not something that scares us.