Month: September 2018

AWS CloudFormation became a programming language

Ilya Sher (B)Leave a comment

… kind of.

Declarative has its advantages which are hyped all over the internet so I’ll skip that part. The painful downside of declarative approach is often the expressivity. Sample proofs:

Now you can have Python embedded in your CloudFormation file. That is part of the CloudFormation Macros which were introduced on 2018-09-06.

Happy coding, everyone!

 

What I did not steal from Perl 6

Ilya Sher (B)1 Comment

I’m curious about programming languages. Not because I’m creating one right now. I always was. This post is about ideas and features that I have seen in Perl 6 and found interesting. If you are curious about programming languages in general, you should take a look at these.

There are various reasons for not stealing the interesting ideas from Perl 6:

  1. I’m trying to keep number of concepts in NGS as small as possible. If I’m not seeing huge immediate value in a concept – I skip it.
  2. Not taking anything that I think can confuse me or other programmers. I’m not talking here because someone is a beginner. I’m talking about confusing concepts.
  3. Simply because I don’t have enough resources to implement it at the moment.

Here are the interesting Perl 6 features, in no particular order (except the first one). There are also my comments whether I would like the feature in NGS or why not.

  1. Syntax. Very expressive an terse. Perl6 has even more of it than Perl 5. Now that we got rid of the $ and friends in the room:
  2. Grammars. Would actually be nice to have something like that in NGS.
  3. Lots of operators. The most interesting concept is Metaoperators. I’m trying to keep the amount of syntax elements in NGS relatively low. There are already two syntaxes in NGS: commands and expressions. Not taking more syntax without serious need.
  4. How the “pointy block” syntax mixes with “for” syntax: for @list -> @element . NGS already has several syntaxes for Lambdas.
  5. Flow control
    1. “when” flow control. The closest NGS has is “cond” and friends, stolen from Lisp.
    2. repeat while / repeat until . It would be nice to have something like that in NGS.
    3. once . Not sure about this one. The functionality might be needed.
  6. Slips. The behaviour is frightening me: if it does expand, how do I pass a Slip if I just want to pass it, say as an item of an array? NGS uses syntax for slips: [1, 2, *myitems, 3, 4] which I think is cleaner. You know you can’t pass it because it’s syntax.
  7. .WHAT method. I stole something similar from Ruby: the inspect method.

As a special note, I have seen a welcome change from $arr[0] to @arr[0] . I think it removes confusion. (That was Perl 5 vs Perl 6).

Please don’t be offended if you are a Perl 6 hacker and you see that there is amazing feature that I have not mentioned. It could be that I’ve seen this in several other languages already or maybe I did not find it interesting or … maybe I just missed it. Don’t hesitate to leave a comment anyway.

Happy coding, in whatever language rocks your boat! Except for bash. Coding in bash will never be happy.

JQ is a symptom

Ilya Sher (B)Leave a comment

jq is a great tool. It does what bash can not – work with structured data. I use it. I would like not to use it.

In my opinion, working with structured data is such a basic thing that it makes much more sense to be handled by the language itself. I want my shell to be capable and I strongly disagree with the view that a shell “is not supposed to do that”. Shell is supposed to do whatever is needed to make my life easier. Handling structured data is one of these things.

If “shell is not supposed to do that”, by that logic, bash is not supposed to do anything except for running external commands and routing the data between them. Doesn’t it seem odd that bash does have builtin string manipulation then? Maybe bash shouldn’t have added associative arrays in version 4? … or arrays in version 2? How about if and while ? Maybe bash shouldn’t have them either?

woman-698943_640

jq is a symptom that bash can’t handle today’s reality: structured data. The world is increasingly more about APIs. APIs consume and return structured data. I do work with APIs from shell. Don’t you guys use AWS CLI or any other API that returns JSON?

The reality has changed. bash hasn’t. I’m working on bash alternative. Please help me with it. Or at least spread the word.

If you don’t like my project, join Elvish . Elvish is another shell that supports structured data.

Happy coding! Hope it’s not in bash.

Bash pitfall: if test, if [, if [[

Ilya Sher (B)2 Comments

I bet you’ve seen a lot of scripts with seemingly innocent if [ -e blah ];then ...; else ...; fi or something similar . What’s the problem? if has at most two branches while test , [ and [[ have three different exit codes. Oops.

If you make a syntax error (or any other error occurs) in the test , [ or [[ expression, it will return the exit code 2 (or above, according to man test​). if will take the else branch. If you are lucky, you will notice the error message from the test, [ or [[ commands. If not, the else branch will always be executed.

I don’t want to use bash. The pitfall above is one of the many reasons. Unfortunately, I do use bash because it’s still best tool for some tasks. I’m working on alternative to bash. It’s called NGS, the Next Generation Shell. In NGS, the situation above is solved as one would expect from a modern programming language: exit codes 2 and above throw exception.

If you also think that there should be a viable alternative to bash, you are welcome to help me working on it.

Update 2021-12-31: real world example is at https://github.com/awslabs/aws-lambda-cpp/issues/140

Happy coding! Hope it’s not in bash 🙂