Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

> exa is written in Rust, so it’s small

I suspected this would be total bullshit, and it is. Its small binary is a mere 3.4 megabytes. I wonder if I misinterpreted the "small" part.



I wouldn't say it's bullshit, just a difference of opinion.

38 kilobytes of executable isn't small, it's tiny; there's no way exa could get to that level without compromising its featureset or development, and even if you did, you'd just have another ls, and we already have ls.

If it's not small, what would you call it? Medium-sized? Something of that size means "download tens of megabytes of runtime and scatter files all over your computer" to me, and exa's smaller than that.


If it's not small, what would you call it?

Self-contained, portable, featureful, etc. There are plenty of other adjectives to say it's not huge or "download tens of megabytes of runtime and scatter files all over your computer".

For me, <64KB is small and <4KB is tiny. But then again, being aware of the demoscene, I've seen what can be done in those size categories... ;-)


> 38 kilobytes of executable isn't small, it's tiny; there's no way exa could get to that level without compromising its featureset or development

I'd be willing to bet dynamic linking would get it much closer to that ballpark and wouldn't compromise the feature set or development.


Dynamic linking to what, though? The features you're looking for have to available in a library for you to be able to link to it.


The standard library for one; many C programs have libc dynamically linked. Rust's stdlib isn't.


Dynamic linking to the standard library + the projects dependencies, from the cargo file:

ansi_term = "0.8.0"

datetime = "0.4.3"

getopts = "0.2.14"

glob = "0.2"

lazy_static = "0.2"

libc = "0.2.9"

locale = "0.2.1"

natord = "1.0.7"

num_cpus = "1.3.0"

number_prefix = "0.2.3"

scoped_threadpool = "0.1.*"

term_grid = "0.1.2"

unicode-width = "0.1.4"

users = "0.5.2"


As usual, you have to consider static vs dynamic linking, all sorts of other stuff.

Good overview and HN discussion of this: https://news.ycombinator.com/item?id=11823949


Compared to JavaScript, one would guess. Rust is capable of generating extremely small binaries but none of the compiler's defaults are amenable to it


After strip it is 1.1M.


For comparison, ls is 38k on macOS (edit: and coreutils’ ls is 138k).


Rust links all its runtime statically by default. Your /bin/ls is probably dynamically linked to libc and a few others.

That being said that the argument of size is a bit silly to begin with, especially for a 3MB binary.


> That being said that the argument of size is a bit silly to begin with, especially for a 3MB binary.

For a single app, sure it's no problem, but what about when the thousands of other programs in bin introduce similar levels of bloat? Another one is ripgrep, weighing in at 5MB compared to greps 204kb.

And this is just for fairly simple command line tools, Imagine what life will be like when we have UI toolkits like QT statically compiled into everything.


> Another one is ripgrep, weighing in at 5MB compared to greps 204kb.

Did you strip both binaries? If I strip rg after compiling it, it shrinks to 2.7MB on Linux. If I strip the distributed binary on Github (which is statically linked with musl), then it shrinks to 2.4MB.

> Imagine what life will be like when we have UI toolkits like QT statically compiled into everything

I imagine Qt will be dynamically linked, so I'm not sure why you'd say that.

Of course, if you want to statically link something like Qt because it makes distribution easier for $reasons, then that's a different story. But that has nothing to do with Rust.


> Did you strip both binaries?

No, should I have to? I'm just looking at the published binary sizes.

> I imagine Qt will be dynamically linked, so I'm not sure why you'd say that.

I said like Qt, as in something similar written in rust. The Qt5 GUI module weighs in at about 10MB, if every gui program included it plus all the other libs then the bloat quickly adds up.

> But that has nothing to do with Rust.

Until rust has a stable ABI that doesn't require everything be statically linked it's a rust problem.

Love the brigading the rust community always does btw. Apparently someone was triggered enough to go through my comment history downvoting.


> No, should I have to?

You don't have to do anything, including publishing misleading comparisons.

> Until rust has a stable ABI that doesn't require everything be statically linked it's a rust problem.

Qt can be dynamically linked into a Rust application. For example, the default is to dynamically link glibc on Linux x86_64.


I'm comparing the file size of one tool to the published sized of a similar tool, how is that misleading? If the size can be reduced then why isn't it done when you publish it? Considering one is on disk size and the other is zipped I think I'm being more than fair.

>Qt can be dynamically linked into a Rust application. For example, the default is to dynamically link glibc on Linux x86_64

Again, it was an example. Show me a similar rust library that can be dynamically linked to like this? A rust equivalent will have to either use a c interface or be statically compiled.


> how is that misleading?

I explained how. Take it or leave it.

> If the size can be reduced then why isn't it done when you publish it?

Because it's not a big deal? The only people who complain about this are people debating holy wars on HN, as if the static vs dynamic trade offs haven't been well understood by now.

> Again, it was an example. Show me a similar rust library that can be dynamically linked to like this?

Rust's standard library.


The performance difference between grep and ripgrep is stark, ripgrep is blazingly fast, isn't that worth some "bloat"?


Small correction: the performance difference between ripgrep and GNU grep on single large files isn't that great for common use cases. For some more complex cases involving Unicode (and specifically, predominantly non-ASCII files), ripgrep can do a lot better.

(That statement changes a lot if you're looking at another grep like BSD grep, or if you're directly comparing `grep -r pattern` with `rg pattern`, which is always subject to caveats, because the latter is doing some guesswork and parallelism where the former isn't.)


Interesting; du reports on mine at 20k.

    % du -h /bin/ls
     20K	/bin/ls

Although as you mention...

    % ls -laF /bin/ls
    -rwxr-xr-x  1 root  wheel  38624 Jul 15 00:29 /bin/ls*


my work macbook's ls is 16k. huh


It's a lot bigger than conventional ls, but compared to a node.js implementation or Java program the description "small, fast" seems justified.


Well, I have to admit, it is smaller than a docker container of alpine Linux containing the ls binary! Still doesn't mean it's monstrously large for no good reason.


Why would I want to compare apples to oranges? Though admittedly in this crazy day and age someone has probably already reimplemented coreutils in node.js just because.


I have seen my fair share of javascript-based console applications. My guess is that the author tries to assure us that exa is no such thing but behaves reasonably for a console application (not gigantic, not incredibly slow) and makes his statements in that light.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: