I totally disagree on the parser-unfriendly syntax point (no. 1). C++ is a particularly bad example in that code cannot be definitely parsed without semantic information, however I think limiting yourself to a language which can be expressed as LALR/LL(1) is pretty brain dead - you limit yourself to what a given algorithm can express clearly rather than what is clearest to the programmer. It's really actually pretty hard to make a language "naturally" LALR anyway, at least without hacks of some kind. C# definitely isn't LALR, there are many features which get in the way of LALR-ness, such as generics which are, however, useful.
The progress in programming language design is going to come from eliminating human unfriendly syntax. The closer you can get to something that let's someone express their intentions without restrictions, the better. The computer is, uh, servant of a human.
There are many barriers in the way of programming languages or computers in general adapting to human functioning. But whenever someone takes the line that says humans must-adopt-to-computers, it is a fail. And that fail is going to be swept away by the next technology which just works.
The progress in programming language design is going to come from eliminating human unfriendly syntax. The closer you can get to something that let's someone express their intentions without restrictions, the better. The computer is, uh, servant of a human.
And a servant isn't very useful if he can't understand its master. Ignoring machine-friendliness in language design just leads to worse and more buggy tools (including compilers).
whenever someone takes the line that says humans must-adopt-to-computers, it is a fail. And that fail is going to be swept away by the next technology which just works.
Except the "just works" is far more likely to be actually achieved if it's easy to make it work as well as easy to understand.
Besides, "human-friendly vs. computer-friendly" is a false dichtomy. There are certainly things that are good for one and bad for the other, but language features can also be good for both, or very good for one while incurring very small disadvantages to the other.
But don't forget that we don't have to limit ourselves to LALR or LL(k). Approaches like GLR allow the user to be happy as well as the tool writers.
And again, don't forget, the tool writers have a harder time if they have to mangle their grammar to fit LALR, then write a bunch of hacks around it to get stuff to work.
Working around an algorithm which can't express usable languages properly is a bit silly I think. It's better to improve the algorithm in the first place and keep everybody happy!!
I do not know a single editor which does syntax highlighting by actually parsing (including semantic analysis) the code. Most of them use regular expressions at most.
Programming languages force programmers to think clearly and express ideas unambiguously. It's not a bad thing we should avoid; it helps us write robust software.
Both points are valid. Programming languages should be designed for people to use, and only secondarily for machines to execute. However, they should be designed so as to encourage clear expression of ideas when practical.
I think whether a general-purpose language should discourage or prevent unclear thinking is an open question. I don't think I've seen an example of a language doing so without being somehow crippled, and that's a Bad Thing.
I don't think LALR/LL(k) implies automated refactoring is easy and anything else does not!!
Obviously non-context-free grammars are a problem, but a GLR or PEG grammar is still perfectly refactorable.
You forget that refactoring an intuitive grammar to LALR involves totally mangling that grammar and making it more difficult for everyone.
If the language is expressed as a GLR or PEG grammar then tool makers have it easier I reckon - easier to understand grammar.
The way to really ensure it's easy for everyone is for the compiler write to provide some abstract yet intuitive representation of the underlying grammar.
It really isn't an either/or, it's a case of both at once, I think!
I think you hit the nail on the head there. "Context-free" is more of a requirement than LALR or LL. The author clearly isn't a parser writer. I am, but I've yet to try a PEG grammar because ANTLR does such a great job.
But this doesn't change his point (or mine): languages should be created that are easy to write parsers for, so that its easier to write good tools.
And I think this applies to the grammar too: ANTLR's .g files are a good grammar language because its easy to write tools for - hence AntlrWorks.
I'm the author and i did write some parsers. However i'm more in the write-parser-by-hand camp and context-free is not so hard here. ANTLR needs strange hacks to parse Python with its whitespace indentation. A handwritten parser just needs some context within the lexer.
In my eyes a parser generator is the solution, if you want to trade efficiency for cross-language-portability. ANTLR is probably the best solution at this point, though the C backend is quite wierd.
I like to get shit working first, and then downcode if the PG does a horrible job. Nothing better than a hand-written anything, but I like the options of tools. Stange hacks for python supports your original argument I believe.
There's no language of higher order than machine code that allows you to specify computer behavior unambiguously (even that's doubtful).
In fact, any entity abstract enough to be called "an idea" is going to be ambiguously specified. The question is what sort of ambiguity is most useful - ambiguity about about which object shows itself on the screen is great in an object-orient window manager. Ambiguity about which memory location to read when dereferencing an uninitialized pointer is less useful.
I completely agree with the parser friendly syntax (and disagree with you).
The issue is that tools are important. C++ is famously tool unfriendly, when it in fact, has the most need for tools to verify it.
From my understanding of the generics problem, generics do not get in the way of LALR requirements. You just have to implement them slightly differently than C# does to make the tokens for the generic types more identifiable. So you can do generics, just they can't look like C++ makes them look.
That is not the only problem I had with writing a LALR C# grammar. Anything even slightly touching anything else that is used for other stuff caused endless shift/reduce reduce/reduce conflicts. Not to mention I had to totally mangle the intuitive representation of a C# grammar to fit LALR.
Forcing tool writers to have to do the sort of shit I did, even without generics, is not the way forward.
The alternative of a fully LALR-able language is not necessarily desirable either - 'I want to add a feature which makes my language far more usable, oh hang on, that screws the LALR. Oh well, let's scrap that then'.
It would be interesting to see how often code is written for the language vs how often a parser is written for the language before tuning the syntax for one or the other.
Also I suspect that a sufficiently well designed parser api, even if it internally uses quirky non LALR parsing techniques, will get a lot of tools going without the need for rewriting a parser each time.
C++ is a pertinent example of terrible syntax design. Not only does the complexity make it a nightmare to write a standard compliant compiler, it also slows down development because you need much non-contextual information to read code.
Sure, it would be great if you could just describe what it is you are after, except for two problems:
1. Programmers out of a job.
2. How good are you at describing what you want? In English? You think an AI will be able to do it better than a lawyer?
What has this got to do with your point? Well, until that time, your productivity is going to be determined by the quality of your tools. You are going to need to iterate, test and refactor. To do that you need good tools, and good tools are hard to write for non-easily parsed languages. Compare the refactoring tools for C++ (limited) and Objective-C (ha ha) vs C# or Java. My productivity is double, tripple in Java and C# because I'm not spending my time moving characters around files - I'm programming.
You're comparing four relatively similar C-family languages — it's not too surprising that tool support is the biggest difference here. Throw in Ruby or F# here if you really want to compare language quality vs. tool support.
I do agree on every other point, however!