Data-driven Scala Programming
blog.buildo.ioI don't mean to sound rough but do you really need to analyze compilation logs to arrive to the conclusion that you should use an IDE and that types are helpful ?
There is also no real hint into why getting this error: "Too many arguments for method parameters" is inherently bad. I assume that it is not informative enough and that you'd prefer getting a type error instead ? In that case it is not the quantity of error but the quality of error that is an issue.
I have always wondered why anyone would not use an IDE. I have tried to start using editors like vim and sublime. But I just think it makes me type endlessly and wait for compile to see my typos and so on, when in intellij or netbeans you type a few letters and get code completion and automatic imports. This blogpost, however, is the first concrete evidence that I havent really missed anything, and I should not spend a lot of time trying to free myself from the ide.
There is some benefits to not have automatic imports is that it incentivize to keep everything tidy under package object such that you only need a few meaningful imports. It is easy to notice automatic import because the header is an import hell.
But sublime, emacs and vim all support autocompletion (through ensime) so it is not really an advantage of intellij or netbeans.
Those are only "benefits" if you don't have automatic imports, though. In an IDE the neatness of your imports is a non-factor in deciding package structure, and who cares about the size or structure of your import block? You never need to look at it and your IDE probably folds up that block anyway.
I was a vim purist for a long time, but even in Python, where IDE functionality is much more limited, I can't honestly compare the two experiences and say I'm not much more productive in PyCharm w/vim plugin. I can't fathom how anyone writes Java/Scala in plain vim, and Ensime is a cumbersome mess.
Speaking re. python:
IDEA also sorts imports alphabetically, separates system modules from non-system modules and removes the names that are no longer used if configured properly. How much time does someone spend doing this by hand (e.g. just simply reading through the list of them)?
IDEA also allows to automatically make the code PIP8-ready by using Ctrl+Shift+L.
I can refactor the name of the function with a basic shortcut, and I am 100% sure, even if it touches >5 projects at the same time - IDEA will replace them correctly. That could be done by a simple find-and-replace, but then I would need to keep my attention on whether that was a correct replacement. How many minutes a day do you spend trying to keep yourself aware of that?
There are literally so many details in working with moderately-complex Python code (e.g. type hints or lack of them; function arguments; unresolved names) that IDEA basically equals to a constant number of hours I save every day.
I am not going to argue - most of what I mention could be solved by automated cli-based tools; and that's exactly what IDEA does under the hood. By it keeps it all together in one place on my screen right now; here, rather than in a set of disjoint interfaces.
I literally do not get why people still use vim/sublime/notepad++. That's like cooking with a blunt knife. You may and will cook as good. But is it not justifying your laziness to go out and buy a proper knife?
I find Java unusable w/o an IDE. However there certainly are languages that are easier to work with w/o one. Go being one example. Depends on design and available tooling. OT: For automatic imports (and formatting) run `goimports` or configure to run automatically on file save. Congratulations, here's also your error stack traces. (EDIT: I mean syntax errors) An IDE, in that case, isn't at all an obvious improvement. Quite to the contrary I find.
Well, I dunno; writing Elm means I get a compiler warning about unused imports, which a lot of compilers can do. Automatic handling of imports coupled with easy pruning of unused ones seems like a complete win to me, what's the downside?
You can configure IDEA to import only the things you absolutely need to import. So I don't know what you mean.
One practical effect I've seen with this in Scala is that IDE users seem to be more averse to typeclasses and other implicit heavy stuff. These things can't be autoimported so you have to do it manually. People who don't use IDEs are used to importing things by hand anyways
reasonable given one should be averse to them even with all the IDE aids (at least in the sense to be judicious when using them). the scala plugin simply cannot handle many parts of the language anyway
If typeclasses are well-designed, they can be auto-imported. Sadly, it's not a common design in practice.
Because IDEs are bloated and buggy
You are right, the conclusions might seem obvious, but nevertheless it is a great step forward that finally somebody in IT has taken the effort to make a statement about developer productivity that is based on empirical data.
Too many discussions on this subject are just based on intuitions and gut feelings, really amazing if you consider how much of an exact science IT is otherwise. I hope there will be more of this kind of fact based research in the future. I can think of a few more subjects, e.g. productivity gains by switching from an imperative to a functional language.
> do you really need to analyze compilation logs to arrive to the conclusion that you should use an IDE and that types are helpful ?
Yes you do. That's the point of the article - we should apply basic scientific method to programming (that's what distinguishes us from magicians).
> There is also no real hint into why getting this error...
This part of the story is better explained here https://blog.buildo.io/http-routes-at-buildo-1424250c41d3
A little more regarding that error from the same site:
"Spray-routing uses a cool and crazy DSL to write routes in Scala, based on the magnet pattern. Despite it being all nice and flexible, it’s also extremely frustrating to debug."
From, "What's Wrong With Routes?" https://blog.buildo.io/http-routes-at-buildo-1424250c41d3
> That makes sense even if you never experienced how horrible writing JavaScript code is
Please don't do this sort of thing; all it will do is cause people to dig their heels in and fuel the idea that FP is too "dogmatic" or such.
I know you're joking, but I hear this refrain far too often while reading "persuasive" FP articles. This sort of jest just comes off as smug when it's not clear that you're "in the trenches" with JS (or other non-FP coders).
Im curious as to how he's using vim + ensime. I've been trying off and on for the last 2 years to figure out a good workflow using ensime but i typically end up just turning it off after a few days. EnType only works for me when its completely obvious. EnImport seems kind of useless. After these 2, i just give up, tbh.
Still, I wont give up my text editor :) Any emacs folk like to comment on its usefulness? Its seems more fully implemented there.
I've been using vim + ensime for a while.
- EnType is slow and only works when obvious
- EnImport ... never worked for me, but I'd _love_ it! Perhaps there's something I'm missing. How do you add imports?
- The automatic typechecker sometimes works sometimes doesn't. Shows unused imports which is somewhat neat.
However, ensime's killer feature for me is fully contextual completion with <C-x><C-o> - works great and is supremely useful.
Seeing type errors logged is only "good" if they are catching actual bugs. A large amount of Scala type errors aren't bugs I would have had in Python; they're hoops I wouldn't have needed to jump through like worrying about whether a parameter should be a "Long" or an "Int".
Actually, those are the type errors I _like_ having (and don't always have: Scala will implicitly convert `Int` to `Long`), although I concede this is a matter of opinion. However, I am definitely less impressed by the unhelpful type errors one gets when implicits somehow fail...
This is a neat idea.
I wonder if this can be extended to analyse unit test failures.
I'd guess one can get some interesting information from what tests fail more often. I'm thinking of being able to spot bad tests because they're too brittle or spotting bad code because it's too fragile.
Stopped reading right here:
> The second error, “Type mismatch”, appeared 1771 times in the logs. That’s good news. It means the type system is working well: it catches type errors pretty often. Cool! I can finally honestly claim that I use Scala for a good practical reason.
That’s good news? Why? How? "Pretty often"? Comparing to what? Maybe it is, maybe it isn't. I don't know. 1771. Ok then.
> "Pretty often"?
20 times a day in my case and 45 in case of another guy. If we suppose you code actively 6 hours a day it's like being interrupted 7.5 times a hour. That's "often" when compared to other errors.
> "Comparing to what?"
To other kinds of compilation errors (I think this point was clear in the article)
> "That’s good news"
If the compiler doesn't catch errors, most of them will go in production (you're usually not unit-testing all the possible type errors). I'm appealing to common sense here - It would be interesting to see if data confirm this.
This is a blog post, not a scientific paper. This is far from complete - it's just meant to be a good starting point. Factchecking everything is stated here would require an entire research team.
Why not?
I think it's fair to assume that programmers starts the compilation either when they believe their code is correct or when they do want to get an "hint" from the compiler.
Either way, they get a valuable feedback from the compiler and that happens multiple times per day.
Static type systems sometimes refuse to compile otherwise valid code. Type error is compiler telling "I can't prove this code correct with regard to some class of problems I'm supposed to catch". It does not necessarily prove the code is incorrect and would fail at runtime.
Therefore a high number of type errors may also mean the type system is very strict and getting in the way very often.
BTW I prefer the type system to be a bit too strict rather than not catching obvious bugs and then having to struggle with a debugger.