Does anyone know on a technical level why the Monaco editor feels so much faster than the Atom editor? Is there any mechanism Microsoft is employing that Atom could adopt, or are the two editors that fundamentally different?
I'm working on the editor since almost 5 years now. Phew, time flies.
There is no silver bullet, we mostly try to keep all computations limited to the viewport size (if you have 20 lines visible, then typing, colorizing, painting a frame, etc. should all end up being computed with loops covering those 20 lines and not the entire buffer size).
We also use extensively the profilers, and most recently (last month) I learned about this great tool called IR Hydra[1]. The gains from eliminating bailouts in the hot code paths are probably too small to notice (5-10% per render), but I like to think that everything adds up.
We use translate3d for scrolling (except in Firefox which has a known bug[2]) and that brings down the browser painting times considerably when scrolling.
I've also found insertAdjacentHTML to be the fastest way to create dom nodes (from big fat strings) across all browsers.
Sort of silly to mention, but we use binary search a lot :).
[1] https://top.fse.guru/nodejs-a-quick-optimization-advice-7353...
In what kind of world is that a sensible metric to decide if a function can be inlined?
> Graph generation: Crankshaft builds the Hydrogen control flow graph using the AST, scope info, and type feedback data extracted from the full-compiled code. Inlining also happens at this stage. Hydrogen is Crankshaft's high-level architecture-independent intermediate representation.
> Inlining: Performed during the graph generation phase. The inlining heuristic is very simple: basically, if the function being called is known, and it is safe to inline, the function will be inlined. Large functions (>600 source characters including whitespace or 196 AST nodes) will not be inlined. Cumulatively, a total of 196 AST nodes can be inlined in a single function.
So they do use AST nodes as a heuristic; I don't really understand why they would also use "source characters including whitespace" though.
Guess I better keep my comments outside the function body from now on when possible.
I mean, why would they fix it with high priority if the engine already has reasonable performance and the benchmarks are happy as well?
It looks like you're adding and removing line divs, did you run benchmarks on trying to reuse the line divs and change the contents of them?
By using translate3D you're not using the browser's native scrolling correct? Are you using an open source library to replace the scrollbar? What was the actual performance benefit of using translate3D vs native scrolling?
I've actually never tried to compute a diff and apply it inside a line, maybe I'll try it tomorrow :). A line is basically a list of spans, each having a certain class name and a certain text content. I've just always assumed that iterating over the previous spans, adjusting their class names, adjusting their text content, appending new spans or removing extra left overs would be slower than a big fat innerHTML call (given that each dom read/write access leaves the JS VM and I always thought there's a certain penalty associated with each dom call). But I will definitely try it out!
Here is what we do now - https://github.com/Microsoft/vscode/blob/master/src/vs/edito...
[It might not be the best method, but it was guided by measuring]:
* if there is no overlap between frames (e.g. you jump to a completely different location), the whole thing does a single innerHTML call
* otherwise:
* all the old lines that leave the viewport are removed via multiple domNode.removeChild
* all the new lines that enter the viewport are added via a single domNode.insertAdjacentHTML
* all the old lines that have changed are parsed in one single off-dom innerHTML and then cherry picked via multiple domNode.replaceChild
That is what I could come up with in my attempts to minimize the count of dom calls and not pay for reparsing/repainting the entire viewport on each frame. Maybe there are better ways?
If I remember correctly, we ended up not using native browser scrolling for multiple reasons:
* [if we would not set scrollTop ourselves] the browser would just happily jump to a certain scrollTop, painting nothing (white), then we'd get an `onscroll` and we'd be able to paint the lines. But you'd always get this white flash.
* if we would set scrollTop ourselves:
* AFAIK setting the scrollTop causes a stop the world sort of synchronous layout - I don't know why
* We wanted to have an overview ruler that sits inside the scrollbar and highlights things (find matches, diffs, etc.)
* IE has or had a limit around 2.5M px. That meant we would have had to do something special anyways around 80.000 lines @ 19px line height
The scrollbars are custom implemented (https://github.com/Microsoft/vscode/tree/master/src/vs/base/...). Quick tip: do not implement custom scrollbars.
PS:
Some anecdotal evidence I got that making less calls with larger chunks of data might be better was when I was investigating why creating an editor buffer was slow for very large files (100k+ lines). One of the first things the model (buffer) code did was to split the text into an array of lines.
I implemented this as any sane person would, with a nice for loop, iterating over the string, grabbing the character code at each offset and checking if it was \r or \n or a \r followed by a \n. I would then remember the last cut off index and do a simple substring to extract each line, building an array of lines. I thought that must be the best way one could possibly do this (I don't know a better way than a for loop even in C++).
If I remember correctly, that was taking 50ms in some browser for a pretty large string. I replaced that simple for loop with a lame split with a regex! - /\r\n|\r|\n/ - and the time dropped to 3ms. I can only think that looping in C++ must be a lot better than looping in JS [here's the code today - https://github.com/Microsoft/vscode/blob/master/src/vs/edito...]
I'm maintaining a web spreadsheet that supports 10s of thousands of rows and have spent a lot of time on optimization. We have to handle some richer content (i.e. contact pictures) so simple spans aren't always sufficient.
The way we're doing the rendering each cell is its own div and we reuse divs as they scroll out of the viewport (changing their top position). Previously I was using innerHTML on each div but I found that constructing the dom nodes manually (document.createTextNode, etc) and then doing a dom.appendChild turned out to he slightly faster (full "wipe" = 16% lower render time). I then cached those prebuilt DOM nodes and then doing a full wipe ended up being 3x faster.
So there was a small speedup on initial scroll and then when you're scrolling around and seeing rows/cells that you've seen before there's a large speedup. Not sure if that's helpful, but maybe worth investigating.
And yes, I know what you mean about scroll events not getting called synchronously. There seems to be a difference in how some browsers handle scrolling vs painting. I actually filed a bug with Chrome (https://bugs.chromium.org/p/chromium/issues/detail?id=619796...) as they introduced an issue in Jan 2015 that I just discovered.
And yes, I really don't want to implement custom scrollbars so I'm hoping to get optimized enough to not need them... we'll see though.
In our framework, we have an option to yank complex components out of the DOM and replace them with empty containers. And when they scroll back into view, we put them back (and activate them if they are newly rendered).
And now we're painting our own scroll bars and simulate mouse wheel interactions (which then probably still invoke some legacy scroll bar based facility internally in the OS):
https://github.com/Microsoft/vscode/blob/master/src/vs/base/...
Wasn't this part of the motivation for Java? I seem to recall reading something about the heinousness of debugging a half dozen different platform-specific scrollbar implementations...
Thanks for all the info, this is awesome. :)
If you're curious to see the difference, compare the running time of `wc` vs. `wc -l` on large / many files (you might have to force the locale to C for wc -l to hit the fast path).
And obviously creating issues on the github repo for things you're missing is a big help for me. I try to have a fast turnaround on these things.
I remember reading that the Atom Vim authors believed separating the two was a mistake; they were too intertwined. As much as I'd like to have it be a separate project and not think about it, apparently that doesn't work so well...
"Why all these web workers and why should I care?"
"A: Language services create web workers to compute heavy stuff outside the UI thread. They cost hardly anything in terms of resource overhead and you shouldn't worry too much about them, as long as you get them to work (see above the cross-domain case)."
Maybe that's part of the reason?
To be fair I haven't caught up with their current versions (I'm a vimmer, I just tried them) but I think I tried Atom after the React move.
It isn't pleasant to use on my personal laptop (MBA with 4GB RAM) where I frequently switch projects.
While I am a happy Atom user, I am disappointed with performance; I think there's an order-of-magnitude jump in overall speed that the editor could really use.
I understand that Atom is doing a lot, but if you need 16 GB of RAM to run your text editor, this simply means that the developers have not been using the right data structures to manage the state of the application or maybe the new code editors have an implementation of an AI coding for you so you can outsource yourself.
Atom is at a fundamental disadvantage - it's built on a much more complicated and abstracted stack of technologies. Javascript engines do spectacular things these days but they have a harder job to do than executing compiled lisp, and that's before you look at the whole of the rest of the stack involved.
Editor for Middle Aged Computer Scientists
Escape-Meta-Alt-Control-Shift
Tt's still my favourite, though.
I don't think that's a fair assessment given the what is now minimum expectations around syntax highlighting, syntax/grammar validation, autocomplete, etc. We're no longer dealing with "plain text".
e.g. live libclang-based code completion for Emacs: https://github.com/Sarcasm/irony-mode
My Visual Studio doesn't need 16 GB of ram... I have enough with 4 on Windows ;)
I had no issues whatsoever.
Interestingly enough, Atom seems to really struggle on my Windows laptop, it's often incredibly choppy. It actually runs better in a Linux VM than natively on Windows.
Speaking about syntax/text highlighting in HTML ...
In Sciter I've added an option [1] to mark character runs without creating tons of heavy weight DOM elements (The Monaco uses <span>'s for that). Plus an option to style those run marks in CSS:
plaintext > text::mark(keyword) { color: blue; }
plaintext > text::mark(symbol) { color: brown; }
<plaintext>
<text>first line</text>
<text>first line</text>
...
</plaintext>
[1] Tokenizer + ::mark() = syntax colorizer : http://sciter.com/tokenizer-mark-syntax-colorizer/
It could also be that mobile CSS has its own quirks so different implementation can be used, again, not a DOM expert here.