Why Emacs is a better editor – case study for JavaScript developer
blog.binchen.org
blog.binchen.org
It's a shame Emacs lacks the facility to properly capture the more complex requirements of languages today - especially files with mixed content, such as HTML, Javascript and templating tools. The few that do support this do it using a variety of clever hacks to work around this problem; and still, using regular expressions to capture the nuances and complexities of some languages make it an error-prone way of writing major modes in Emacs. Not to mention you lose out on the many advantages of having an AST in memory for code jumping, refactoring, syntax checking and what have you.
Emacs could do with a shake-up in this space. The proposed solution (Semantic Mode / CEDET) has never really caught on, even though it adds much of the same functionality to Emacs; my suspicion is that this is because Semantic/CEDET is ridiculously complex and lacking in documentation.
So, sooo much this. And it's a real shame, because semantic is quite a usable tool once you get to know it.
One thing I really wanted to have was a semantic index, a database of semantic symbols inside a project. I wanted to be able to write "Form" and have it display all the class names containing "Form". Grepping (even ag) is just much too slow and building my own index was too much work. I then saw "semanticdb" module.
A week, I'm not kidding, a week - instead of half an hour - later I finally managed to preload a custom tag database and to query it. And that was with Python code, which has Semantic parser already, which is well integrated into the rest of CEDET.
Now I'm fighting with CoffeeScript/LiveScript support. They both expose their parsers - I can get their AST with one line of code. After another week I still don't have working Semantic plugin for them.
When I last checked, the activity on a CEDET and Semantic mailgroup was almost non-existent. The home page for CEDET looks like it comes from last century and even info pages are not a big help.
It's a friggin shame, because when Semantic works, it does work well.
http://news.gmane.org/gmane.emacs.cedet
Sure, it's not LKML, but still, questions usually get answered.
> Now I'm fighting with CoffeeScript/LiveScript support. They both expose their parsers - I can get their AST with one line of code. After another week I still don't have working Semantic plugin for them.
I'm afraid an AST is just the beginning for properly supporting a language. If you need help, don't hesitate to ask on the mailing list.
> The home page for CEDET looks like it comes from last century and even info pages are not a big help.
Yes, unfortunately.
What Yegge's project shows is: that EMACS does not lack the facility to do what is needed [LISP is it's command language after all], and that implementing the task is non-trivial even for skilled programmers.
It's just that historically it just wasn't done. The rise of IDEs tied closely to the compiler/environment changed that equation and now it's Emacs that's behind the curve.
Parsers have been used to do semantic highlighting of symbols, e.g. paint identifiers in different colors depending on whether they're interfaces or classes, etc. But the big win for parsing isn't highlighting, it's completion. You can't do super useful completion without a fairly rich parse tree, and not just for the current file, but also for all the libraries.
And while it may be trivial against some other metric to deal with AST's in another language, the solution tends to be a one-off, that is dealing with the AST of C++ doesn't offer an advanced starting position for dealing with the AST of JavaScript - most of the work is in the details of a particular language not at the higher level abstractions of languages in general.
I'd also point to Ensime as an example of a language plugin that uses an external parser to check code.
(I use Emacs for all my coding, and Common Lisp is my favorite programming language, so, if anything, my biases would be in favor of Emacs. Sadly, it doesn't really deliver -- not that other editors get their interface right. Don't get me started on vim. ST is somewhat better, but doesn't expose powerful enough tools. If I may toot my own horn, I believe my own project, http://codemirror.net , does a lot of things right on the editor API level that popular editors get tragically wrong.)
(For me personally being able to run in a terminal is a huge feature, which means I'll probably be stuck w/ emacs forever -- no one making new editors seems to care about terminal support.)
JS would not be a better language. Guile could be.
I used to think a more semantically aware setup would be better. But the costs of semantic awareness are enormous, because a single mismatch between what your editor understands, and what the underlying language can represent, breaks everything. This makes things fragile; parser error recovery is not an easy thing to get right. Some artfully written regexes can go further, in a practical sense.
Part of what Emacs a great editor and Elisp a great tool for mutating -- if one can call it that -- is exactly the same things that make it "bad." Yes, it has dynamic scoping, and yes you can change variables outside your scope -- which is exactly the level of granularity your average Emacs hacker wants. He's not looking to write rock-solid code for the mars rover; he's trying to make M-x compile a bit less crap or a bit more specific to what he's doing.
Say what you want about side effects and advising and all the other "terrible" practices that, in any sane software project, would never pass peer review -- in Emacs it enables you to hot plug, duct tape and jury rig Emacs to get it to do what you want.
The last 2-3 years has seen a massive resurgence in Emacs packages and there's a lot of buzz in the community, I feel, in trying to push things in the right direction. But it takes time. And people. And we need both.
I guarantee you that if Emacs used a "real" extension language, with lots of OO and state hiding, you'd be yelling obscenities into the computer the first time you wanted to tweak some extension you had. Have you ever seen an Eclipse user tweak one of his extensions? Didn't think so.
Agreed the ease of modification of emacs is unparalleled (and tricks like dynamic scoping can be handy at times), but I'm not convinced that it wouldn't be possible to get something aprox. as flexible and easy to modify with a more robust culture and langauge. (But not a classic "real OO" language, which I agree would be very annoying -- I'm thinking something with a strong functional flavor and static structural typing...)
Whether this is really possible or not is an open question, though.
So out of curiosity, how far away do you think CodeMirror is from being usable as a real local editor - and by that I mean how far away would you be from using it instead of Emacs? It seems like it would be straightforward to embed in something like node-webkit[2]. What else then remains?
[1] http://eloquentjavascript.net/ [2] https://github.com/rogerwang/node-webkit
- The compiler really isn't very effective, so doing expensive things is often tricky and takes a lot of fiddly optimizing to get halfway acceptable performance (this should get better when/if guile becomes the Emacs compiler).
- But the killer is the APIs and conventions that exists around the language. It's neat with how little code you can do some things using direct actions on buffers, and the various hooks make a lot of things sort of possible, but I've found again and again that things just don't compose, scripts trample on each other's toes, and trying to deal with dynamic scope (even with lexical scoping, lots of interfaces depend on dynamically scoped variables and concepts, such as the current buffer) when you're doing something asynchronous is a total nightmare.
http://www.gnu.org/software/emacs/manual/html_node/cl/Struct...
http://www.gnu.org/software/emacs/manual/html_mono/eieio.htm...
I remember nicferrier and others discussing how to write lazy streams for elisp in order to have a more functional text processing ~std lib.
All in all I agree with you that the extensions are weirdly integrated and fragile, maybe it's time to build a side-project emacs with better abstraction layers.
ps: What's make me so sad it that elisp editing api is really not lispy in nature. Language doesn't dictate good design...
That said, an emacs rewrite in Clojure is a wet-dream of mine (emacs like goodness with the ability to use modern GUI libraries, ANTLR parsers etc... and speed!)
Although it seems to be stalled. Part of me wonders/hopes LightTable will get to the point that it is a viable replacement for emacs.
I'm not saying that makes vim a better editor than emacs. Just pointing out that vim has other options for extending. Also, I use vim and haven't found emacs to be more productive; in my college days I tried emacs, in earnest, for a few months. I came back to vim eventually, and haven't felt compelled to give emacs another go since.
Edit: it's obviously far from perfect, but it has stood the test of time and it lets you get things done. It's no Scheme or CL, but for what people use it for it's fine.
That said, I wish I could pin some of my hopes to the upcoming Guilemacs.
"Will it be open source?
I'm a firm believer in open source software and open source technologies. I can guarantee you that Light Table will be built on top of the technologies that are freely available to us today. As such, I believe it only fair that the core of Light Table be open sourced once it is launched. At some level, this is an experiment in how open source and business can mix - it will be educational for us all." - http://www.kickstarter.com/projects/ibdknox/light-table
In the meantime, humble projects are progressing along without all the young people JS-dev hype.... CodeMirror, emacs/vim editor plugins, etc
I am the founder of a browser based IDE software company, our product uses Codemirror exclusively and extensively..check my profile for the link if you want to check it out.
Beats the hell out of javascript.
One useful technique to make Emacs appear cleverer than it is vis-a-vis code searching and navigation is to navigate by syntax highlighting - most of which is assigned one of several types (constant, keyword, statement, etc.) -- and simply use that. It's hardly perfect, but it's good enough for 90% of the basic navigation you need to do.
How do you do that?
Emacs has swank-js for interacting with a browser engine, but it doesn't appear to capture the functionality of Netbeans' chrome extension.
Sublime Text: http://imgur.com/p27i5L7
The tool should support even the most hideous constructs; saying that "but what you wrote could be written better" is completely orthogonal to "the tool couldn't handle it".
Nope, the tool should choke on hideous constructs, and make you write them properly.
See how I provided just as many arguments as you?
If that's a huge problem to you, use said editor only for greenfield projects...
The only reliable exceptions I've found are Perl, which is unsurprising if you know anything about how that language parses, and the PHP modes, all of which bear stigmata of having been written by PHP developers, and each of which is broken in its own special way. Yegge's js2-mode is a notable exception of the opposite, and highly laudable, sort; when you screw something up using that mode, it knows what you did wrong, and uses the echo area to tell you about it.
https://github.com/romainl/dotvim/blob/master/vimrc#L302
https://github.com/romainl/dotvim/blob/master/autoload/funct...
Because anything you put time into to makes it automatically better for your use case is actually better.
It does not inherently understand require.js. Nor does ST. With Emacs's JS2-mode, thanks to its AST, it is possible to augment the library so that it does understand require.js -- and the author does so without resorting to piles-upon-piles of regexp hacks.
Your main point stands though: all major editors have the ability to add plugins to achieve this same functionality.
Tern is a stand-alone code-analysis engine for JavaScript. It is intended to be used with a code editor plugin to enhance the editor's support for intelligent JavaScript editing. Features provided are:
*Autocompletion on variables and properties
*Function argument hints
*Querying the type of an expression
*Finding the definition of something
*Automatic refactoring
*Emacs
*Vim
*Sublime Text
*Brackets (built in to the base editor)Yes, but the important thing is that it CAN have one, same as Emacs could.
For example, it has a Go plugin which uses the Go AST.
I was playing around with another interesting editor (Chocolat) which actually uses a Javascript engine to identify symbols in Javascript source files; but it does this far worse that can be done easily with regex (and in fact makes terrible errors doing so).
Indeed, better results than that shown in the article are achieved by fancy new editors (like Espresso) and crufty old editors (like BBEdit) using regex and brute force. Fundamentally you've got an anonymous function or a function with a common name, and you want some outer context to further identify it -- regex can do that.
I wanted to start learning lisp and was told emacs was the best choice for it. I read that Richard Stallman made it and Linus Torvalds uses it. That was enough to get me started and it's now my editor of choice.
But that's not to say that I don't know how to edit, save, search and (force) quit vi. :)
Over months I read many articles describing cool things you can do with emacs you can't do with vim, and eventually I switched.
So: one article, no. Many, yes.
http://i.imgur.com/rMdKYde.jpg
(this is in jest but no doubt I'll be downvoted...)
They made "line-move-visual" the default, which in the service of making Emacs behave more like Microsoft Word, it makes ^N and ^P across wrapped lines stay on the same line and move to the part of the same line that is wrapped above or below the cursor, totally breaking type-ahead predictability and keyboard macros. That totally pissed of the late and long time Emacs user Marc Crispin, and he explains why it's such a terrible idea that breaks keyboard macros: http://compgroups.net/comp.emacs/line-move-visual/274792
It totally breaks keyboard macros, one of the most important things about Emacs. It definitely should not have been made the default, since keyboard macros are a way of life for real emacs hackers. I totally agree with Marc's take on the problem, and I am terribly disappointed that RMS let somebody get away with changing it that way, and making that horrible mode the default.
They just can't stop trying to make it more like Microsoft Word, while failing to actually achieve that goal, and only succeeding in inflicting half-assed solutions with harmful unpredictable side-effects on the users.
Another example is how the region highlighting and the weird pending delete behavior terrorizes me that sometimes but not all the time according to rules I just can't figure out or tell by looking at the screen, when I type something that I intend to insert, a huge chunk of my buffer might just disappear, but sometimes it doesn't. So now Emacs has become unpredictable and malicious. I have to do a dance of "insert character, undo" to know that I have canceled the pending delete mode. You can't tell if you are in pending delete mode just by looking at the screen and seeing the obnoxious highlighting, because sometimes it highlights the region and isn't in pending delete mode, and sometimes it highlights the region and is in pending delete mode. I just wish it would stop highlighting the region, because any time I set a mark and move around, it highlights half of the screen at no use to me, and that just gives me a headache, and I have to insert a character an undo it to cancel the highlighting, and until I cancel the highlighting I am living in terror that my next keystroke will delete a huge unseen portion of the buffer.
And the thing that REALLY pisses me off is the lame-assed attempt to make ^A and ^E ignore the prompt in shell mode. There is an extremely simple reliable solution to the problem of separating the prompt from your input in shell mode so you can always get to the beginning of the line with ^A, and that is to have a newline at the end of your prompt, so every line you type in is on the whole line and does not have a prompt prefix, therefore no half-assed magic is necessary.
But the half-asses magic is terribly implemented and has bizarre side-effects that screw me all the time: I get these fucking invisible force fields inserted into the line between my characters whenever I yank some text onto the line that has a prompt or a command line on it, and I can't move past them with ^A or ^E (but sometimes I can -- they're not predictable which is even worse)! And I can get any number of these fucking invisible force fields on my line, so in order to get to the beginning of the line I have to go ^A, look at the screen to see if I made it, type ^B ^A if I didn't, again and again, until I drill past all the invisible force fields that are trying to make my day happier.
And then I have to mark the beginning of the line I want to edit and repeat, go to the end of line by drilling back some number of opposite facing invisible force fields (which may be in different places than the ones going the other direction), and then kill the region, go to the end of the buffer, then yank the entire line to get a copy of it without any fucking invisible force fields in it.
But if there is an invisible force field in the line, or I just want to edit a few characters of the line and that adds some invisible force fields that were not even there before, and then I hit return in the vain hope of re-entering the entire line but not the prompt, it just enters the tail of the line after the last invisible force field, making an error in the shell, or sometimes even executing a totally different command than I intended and totally fucking me over.
I would really like somebody to explain what the fuck the idea behind the invisible force fields are, and give me the address of whoever thought is was such a brilliant idea, so I can send a robot drone to firebomb their house, or worse yet send RMS to live with them for a few months. Why did they do something so terrible, that totally fucks up such common important operations, to solve a trivial problem that has a much better solution?
Emacs used to be totally predictable. I learned it at 300 baud over the ARPANET on ITS, and I could type ahead hundreds of characters to do all kinds of stuff, and then go take a piss or get a drink or take a bong hit, and come back later, and it would be in exactly the state I meant for it to end up in. Now, I can't do that even on a local display. And for some fucking reason, it waits for up to five seconds sometime before updating the display when I'm running it in a terminal (especially when I start an incremental search or query replace). What the fuck is that about??! Not only is it totally unpredictable and forces me to stop what I'm doing and wait for it to catch up so I can figure out what state it guessed its way into before going on, but it won't paint the screen for five seconds when running on a modern top-of-the-line computer!
Emacs has become a shrine to outrageous violations of the principle of least astonishment. I just can't figure out how they could have come up with so many ways to corrupt it, when it used to work so well. It's totally unpredictable now, and I can't type ahead any more, so I have to type a few characters, look at the screen to see how it misinterpreted my intentions, and then try to work around the misunderstanding.