[flavio81]

>That section made me chuckle. Admirable if true

Mostly true for very high level languages like Lisp/Scheme, or ML/OCaml/F#/Haskell, when faced against not-so-high-level languages like C, C++, Java.

Against Racket, i wouldn't be so sure. Nor against Ruby.

Python and Javascript are high level languages but they are crippled by some bad design decisions.

[Zaak]

I know that Python and Javascript have their warts (as do all languages in my experience), but what decisions in particular are you thinking of?

[fnord123]

Lambdas are limited to a single expression. Threads don't work the way anyone would ever want threads to work because o the GIL.

That said, homoiconicity (s-expressions) is not a feature I want. It makes code like reading a wall of text compared to a nicely laid out magazine.

[kazinator]

Since Python lambdas also are expressions, they cannot contain statements, due to Python adhering to the Algol-like statement/expression syntactic paradigm. If a lambda expression could contain statements, that would mean that almost any other kind of Python expression could also contain statements by containing a lambda expression.

But, here above, I'm writing more from the angle of supporting multi-line lambdas that contain statements. Strictly speaking, you are only bemoaning the lack of multiple expression support, not multi-line lambdas.

Python could adopt something similar to the C comma operator. It almost has that in the form of list constructors, except that these return a list, instead of the rightmost value:

  [foo(), bar(), xyzzy()] #   foo, bar and xyzzy are called

Idea: a dummy function called progn could be used for this:

  >>> def progn(*rest):
  ...    if len(rest) > 0:
  ...       return rest[-1]
  ...    return None
  ... 
  >>> progn(1, 2, 3)
  3
  >>> progn(1)
  1
  >>> progn()

So now we can do:

  >>> x = lambda arg: progn(print(arg), print(arg), "done")
  >>> x(42)
  42
  42
  "done"
  >>>

There you go. Lambdas are (effectively) not limited to a single expression. If progn is too long, call it pg (Paul Graham) or pn (Peter Norvig).

Always have your Lisp hat on, even if you find yourself in Python land.

Maybe this is a common trick? I don't use Python; I hardly know anything about it. I wrote one Python program before which garbage-collects unreferenced files from a Linux "initramfs" image, making it smaller (thus reducing a kernel image size). This was in a Yocto environment, which is written in Python 3, so that choice of language made sense.

BTW does Python require left-to-right evaluation order for arguments? I would sure hope so; it would probably be "un-Pythonic" to plant such a bomb into the language as unspecified eval order.

BTW looks like a more idiomatic definition for progn is:

  >>> def progn(*rest):
  ...    return None if len(rest) == 0 else rest[-1]

[klibertp]

> Maybe this is a common trick?

No. It's obvious and trivial, but you'd be on a verge of being called names if you tried to use it in a Python codebase. Lambdas in Python are limited to a single expression by convention - which in Python-land is scarily rigid and specific - rather than just by the language spec.

Before `... if ... else ...` was added to the language as an expression (I think around 2.5), people had to make do with some workarounds. The fact that `True` and `False` get automatically casted to ints and back allowed for writing something like `[val_if_false, val_if_true][condition]`. Or you could use `and`/`or` combination as per usual. The official stance at the time was to never do this and use an `if` statement instead, but people still sometimes resorted to it. Then, the `if` expression was introduced specifically to combat the use of such workarounds. Now you'd be lynched if you tried to use one of them.

"There should be one - and preferably only one - obvious way to do it" - from the Zen of Python[1].

In general, despite a lot of effort to eliminate them, there are still some creative ways to use the language. It will always be the case, obviously, as you demonstrate. However, that creativity is 100% rejected by the community, to the point that even mentioning inadequacy of some construct for some use case is frowned upon - because it could lead to people inventing creative workarounds. If you try to complain about something in the language, the general attitude is "write a PEP or GTFO". More often than not it results in the latter.

The saddest part of it all is that this apparently is one of the major factors that made Python as popular as it is. There are valid reasons and a lot of advantages to this strategy. Go is similar as far as I can tell. Among the dynamic languages with rich syntax, Python codebases tend to be stylistically very close to each other, and not because there is a lack of ways this rich syntax could be (ab)used, but because doing so is unpythonic.

Haaah, now I said it... I hope not many Python programmers read this thread; I can already see torches and pitchforks on the horizon...

Source: I've been writing Python for the last 12 years for pay.

[1] https://en.wikipedia.org/wiki/Zen_of_Python

[maddyboo]

My experience with Go is that the syntax is more rigid as compared to Python but the community is much less idealistic.

As long as your code is linted with gofmt and it compiles (and doesn’t abuse reflection), the community tolerates more creative uses of the syntax to get around some of the pitfalls of the language – but there is of course less opportunity for creative syntax than in a dynamic language like Python.

[fnord123]

>Strictly speaking, you are only bemoaning the lack of multiple expression support, not multi-line lambdas.

Oh, I'm not bemoaning. I'm just answering GP about the likely limitations that the previous poster was complaining about.

>Maybe this is a common trick? I don't use Python; I hardly know anything about it.

I work around lambdas by naming internal functions. It's easier to read intent if I tell you what I'm trying to do.

[Zaak]

Gotcha.

I do agree about homoiconicity. It's great for writing macros, and terrible for everything else. Of course, go too far in the other direction and you get perl, so... yeah.

[klibertp]

I don't want to reopen this 50+ years old can of worms, but I have 2 questions:

1. Is it about homoiconicity in general or specifically s-expressions [EDIT: I see GP writes about s-exps specifically, missed it at first]? Prolog, Erlang, TCL, and Rebol (just some examples) are homoiconic, but not s-exps based. What do you think about them?

2. Do you often read code without syntax highlighting and proper indentation? Assuming the code is properly indented and colorized, what makes it so hard to read in your eyes? Take a look for example at snippets in: https://docs.racket-lang.org/quick/index.html#(part._.Local_... - what do you feel is wrong with them? Is it only the placement of parens, or is there something else?

[Zaak]

1. It's homoiconicity in general. It's most obvious in arithmetic expressions, but homoiconicity (as far as I've seen) eschews semantic indicators besides function name and argument position. I find that keywords and symbols are almost indispensable for smooth eye-parsing of code.

2. No, I use syntax highlighting and indentation all the time. Trying to put words to my vague thoughts, I think the main issue is that, for example, in a let block the only indication of the meaning of all the items is the single function/macro name at the beginning of the block. Whereas in an algol-type language you have assignment operators in each item to indicate what they mean.

[klibertp]

> I find that keywords and symbols are almost indispensable for smooth eye-parsing of code.

Sure, but that's a totally separate issue from homoiconicity. All the examples I mentioned are infix languages (although TCL requires you to opt-in for that):

https://github.com/adambard/learnxinyminutes-docs/blame/mast... (I couldn't find Rebol, but Red is very similar)

https://github.com/adambard/learnxinyminutes-docs/blame/mast...

https://github.com/adambard/learnxinyminutes-docs/blame/mast...

https://github.com/adambard/learnxinyminutes-docs/blame/mast...

(sorry for the links to blames, but it's impossible to link to a specific line otherwise)

Prolog lets you define your own operators and you have control over associativity and precedence. TCL `expr` works like `$(( 2 + 3 ))` in BASH (which also doesn't support mathematical operators normally) and by reimplementing it you can also add your own operators. Erlang also allows you to define new operators, although it's much more hassle there, as you need to write a parse transform (which is only possible to do so easily because of homoiconicity of the language). I'm not sure about Rebol/Red - I had only passing contact with it long ago.

In other words, homoiconicity itself has little to do with where the "keywords and symbols" are placed in the code, or whether they are used at all.

Accidentally, Lisps also support infix syntax just like TCL: for most Lisps you can grab a lib/package/source of the `infix` macro, which allows you to write infix arithmetic (and code in general) without problems.

> I think the main issue is that, for example, in a let block the only indication of the meaning of all the items is the single function/macro name at the beginning of the block. Whereas in an algol-type language you have assignment operators in each item to indicate what they mean.

Ok, that's one way of looking at this. On the other hand, one can also say that the repeated use of the operator where the meaning of each line is already determined (by the fact that it's inside `let` block) is useless cruft, which actually hinders comprehension by forcing you to mentally parse one element more on every single line of the assignment block. In Lisp, it's actually trivial to extend the language to include this form of `let`:

  (let
      ((a = some_expr)         ;; could be `:=` `<-` `is` instead of `=` 
       (b = some_other_expr))
    some_statements
    ...)

In Scheme, the fundamental conditional construct called `cond` even has an operator-like construct built-in

    ;; in Racket
    (cond
      [some_condition => a_function_called_on_the_result_of_condition_expr_if_its_not_false])

Yet, including this kind of additional syntax is not widespread in the community. Unless the additional syntactic element actually changes the meaning of the code, like in `cond` case, it is seen as superfluous and not needed.

Both opinions have some merit to them. How you feel about both styles is largely determined by what you're familiar with and what you're used to. In the first style, you have to learn to ignore some token in some places as they don't add any meaning to the code. In the second, you need to be careful not to mistake one kind of block for another, and you need to learn what is the relation between subexpressions in each kind of block.

Both approaches require learning. The difference is that you already learned the first one, while you're not familiar with the other. But - and that's what I really wanted to say - there's no difference between readability of the two approaches once you learn them. In other words, Lisp code is as readable for Lisp programmers as JavaScript is for JS programmers. Further, all Lisps are equally readable to (a specific) Lisp programmers, just like all Algol-like languages are readable for JS programmers.

With a bit - and I mean a bit, like a few days; if you want a language you won't find readable after half a year, go for J - of practice you could read Lisp-like code without problems. There's no inherent unreadability to either Lisp- or Algol-like syntaxes - is what I'd like to convince you to :)

[fnord123]

>I see GP writes about s-exps specifically, missed it at first

My experience is with s-expressions being a wall of text. I haven't used prolog (end erlang uses the same syntax) enough to have a beef with them. Perhaps I was being over general.

[klibertp]

> My experience is with s-expressions being a wall of text.

That's sad, but unfortunately not that rare. Making a Lisp readable is not as easy as doing so in Python.

In my experience, Lisp code written by an experienced lisper who cares about readability can often be much more readable than well-written Python (assuming equal levels of proficiency in their respective languages). On the other hand, the number of ways you could destroy the readability of Lisp source is endless, they are always close by, and are even context-dependent. Beginners or people without the focus on readability use those ways rather liberally.

The situation is the exact opposite: Python defines the lower bound on the code readability ("Readability counts" -> PEP-8 -> linters -> (lately) `black`) while Lisps, in general, don't even have an authoritative PEP-8 equivalent.

On the other hand, Python also has an upper bound on how readable it can be: its syntax is rich, but if you find yourself in a place where it's not rich enough, you're on your own. You could just use the expressive power of the language to hijack the syntax and beat it into shape better suited for your problem, but it will be almost certainly seen as un-Pythonic.

On this point, Lisps have a huge advantage, because you can change the language parser on the fly easily, and you can add whatever syntax sugar you need in a couple of lines of a macro. In other words, Lisps give programmers tools for making their code as readable as they want (and are able to) while at the same time allowing them to write "walls of text" (and honestly, that'd be a very polite way of describing some of the Lisp code I've seen) which - in readability - could be one of the worst among many languages I've seen.

So, what I want to say here is that it's possible - and not that hard - to write readable Lisp code. Unfortunately, a programmer has to both think of readability when writing and have skills to make their ideas on readability into reality.

In effect, yes, there's a lot of Lisp code which is hard to ingest. Some style guides are there or are being created, paredit helps a lot, I'm not aware of any linters yet, but they should start appearing at some point. On the other hand, Lisp code skillfully crafted for readability is rivals (and sometimes surpasses) Python at its best.

I'm not sure what Lisp code you've seen, but I assume it was all of the former kind. This is unfortunate. Without knowing what was it you were reading/working with it's hard to recommend anything, but I found examples in "How to Design Programs" quite readable: https://htdp.org/2019-02-24/part_five.html and there are also other books and Open Source projects with code worth reading, but I'd have to dig through my bookmarks, which I don't have the time for right now, sorry :(

TLDR: Lisps - Schemes, Racket, CL, PicoLisp, Emacs and TXR Lisps to name a few - can be used to write astonishingly readable and to-the-point code, but the languages do absolutely nothing to discourage using them to write the most unreadable mess under the heavens. As for the reasons for this - I've honestly no idea at all.

[flavio81]

>I do agree about homoiconicity. It's great for writing macros, and terrible for everything else.

I don't know; i've written code in C, C++, C#, Java, Python, Ruby, Pascal, Delphi, Assembler x86, TCL, Javascript and Common Lisp. Lisp codebases are the cleanest and clearest i've seen by far, although ReasonML/SML/OCaml might be as clean too.

[dataangel]

Racket also has the GIL though. It's a problem in every dynamic language I've looked at. Last I checked it was true for Ruby too.

[flavio81]

Common Lisp implementations often have no GIL, and, btw, are able to reach C speeds in certain cases.

[srean]

Tcl does not have GIL and is possibly one of the most dynamic languages that has seen nontrivial use. Guile does not have GIL. Running multiple interpreterd in different threads is a common notion in Tcl.

[blacksqr]

Running multiple sub-interpreters in different threads is easy in Tcl, but it doesn't stop there. If you want, you can run multiple threads in a single interpreter, using locks mutexes, etc., if you're inclined to take on the attendant complexity and risks.

Ability to run a separate interpreter per thread was designed as a handy simplification, so you could sidestep the usual complexity of thread programming if you wanted.

[klibertp]

That's true about Racket, but it has a quite original way of dealing with it with their futures[1].

To be honest, I'm not sure about the details, but it should let other threads run truly in parallel as long as it's "safe" to do on the VM implementation level. So, if the code inside the future doesn't perform any "future unsafe" operations, it can execute within a separate OS thread without worrying about the main thread.

Examples of "future unsafe" actions were given as memory allocation, and JIT compilation. Further, it's mentioned that some simple (for the language users, at least) operations may be too complex internally to be "future safe". An example of this is using a generic number comparison operators - `<`, `>`, etc. Apparently, these have to handle the full numeric tower of Racket and in the process perform some future unsafe operations.

In the Mandelbrot function given as an example in the guide, simply replacing the generic comparisons with the ones specialized for work on floats specifically (and assuming that contract is not broken, which would immediately stop the future) allows the future to execute fully in parallel.

What is important to note here is that `set!` and friends, and so mutation of shared memory, is considered "future safe", ie. it's permitted to use them! (although then it's you who deals with the usual problems that brings).

I think it's worth mentioning here, because it's a novel strategy that seems to be between the two usual solutions (1. we've got GIL, live with it; 2. spawn more processes and get them to work - well, now you have many GILs...) and is showing some promising results. Plus they have a neat visualization tool!

Currently, it's limited and works best for purely numerical computations (which is also where you'd need it 99% of the time), but in some cases, it appears to work: the programmers of the language (not the implementation of the language) are given a tool to work outside the GIL in a structured manner plus a tool for closely inspecting low-level operations that happen in their code which would suspend or stop the future.

I'm not aware of any other dynamic or not language which has both the GIL and a nice, language-level tool for freeing it and running in parallel. Because what is considered "future unsafe" depends on the details of the implementation, I'm full of hopes for Racket-on-Chez, although I think I read somewhere that work on futures is not a priority at this time.

Also, to confirm the sibling comment, SBCL is happy to spawn truly parallel threads. There are other Scheme implementations (I think Chicken at least, but not sure right now) who allow the same.

[1] https://docs.racket-lang.org/guide/parallelism.html#%28part....

https://docs.racket-lang.org/reference/futures.html

[BoorishBears]

null and undefined being separate values