In my opinion, clap is a textbook example of over-engineering for a single metric (UX) at the expense of all other considerations (compilation speed, runtime cost, binary size, auditability, and maintainability). It is an 18kloc command-line parser with an additional 125kloc of dependencies that takes nearly 6 seconds to compile (‘only’ 400ms for an incremental build) and which adds nearly 690KiB to an optimised release binary (‘only’ 430KiB if you strip out most of the sugar that only clap provides).
There are many other command-line parsers to choose from that do all the key things that clap does, with half or less the build cost, and most of them with 30x less binary overhead[0]. argh is under 4kloc. gumdrop is under 2kloc. pico-args is under 700loc. What is the value of that extra 10kloc? A 10% better parser?
I am not saying there is no room for a library like clap—it is, at least, a triumphant clown car of features that can handle practically any edge-case anyone ever thought of—but if I got a nickel every time I spent 15 minutes replacing a trivial use of clap with pico-args and thus reduced the binary size and compile time of some project by at least 80%, I would have at least three nickels.
Just to try to pre-empt arguments like “disk space is cheap”, “compiler time is cheaper than human time”, etc.: there are no golden bullets in engineering, only trade-offs. Why would you default to the biggest, slowest option? This is the “every web site must be able to scale like Facebook” type logic. You don’t even have to do more work to use argh or gumdrop. If clap ends up having some magic feature that no other parser has that you absolutely need, you can switch, but I’ve yet to ever encounter such a thing. Its inertia and popularity carry it forward, but it is perhaps the last choice you should pick for a new project—not the first.
According to that link you posted, many of the other argument parsers don't even generate help, only one other offers multiple interfaces, and none of the others are noted as having color or suggested fixes.
These aren't exactly esoteric features, and you're not going to get them for free. I'm happy to pay in space and compile time for clap to gain those features.
This isn't a case of the commandline app needing to be facebook, but rather putting the exponential gains we've made in storage space to good use providing features which should be table stakes at this point.
You’re right that there are only trade-offs in engineering. But the key to evaluating trade-offs is evaluating impact, and how long my dependencies take to compile when I first check out a repo or whether it takes 1ms or 2ms to parse my command line (if we’re even talking about something above microseconds) have no discernible impact for approximately all use-cases. If you’re making some odd CLI tool that has to run on an microcontroller with 1MB of RAM or something, fine, agonize about whether your command line parser is parsimonious enough. Otherwise you’ve abjectly failed to evaluate one of the most important trade-offs in engineering: whether something is even worth your time to think about.
> Otherwise you’ve abjectly failed to evaluate one of the most important trade-offs in engineering: whether something is even worth your time to think about.
Phew, several folks have replied to me about how it’s not worth the time thinking about these impacts at all, thus creating a paradox whereby more time has been spent thinking about writing about whether to think about it than has been spent in not thinking about it and just accepting that I wrote a reply on HN about how I feel there are more suitable command-line parsers than clap for most Rust projects! :-)
I agree that much of high-level engineering is knowing whether something is worth thinking about; in this case, I did the thinking already, and I am sharing what I know so others can benefit from (or ignore) my thinking and not have to do so much of their own. If my own personal anecdote of significantly reducing compile times (and binary sizes) by taking a few minutes to replace clap is insufficient, and if the aggregate of other problems I identified don’t matter to others, that’s alright. If reading my comment doesn’t make someone go “huh, I didn’t know {argh|gumdrop|pico-args} existed, clap does seem a little excessive now that you mention it, I will try one of these instead on my next project and see how it goes”, then I suppose they were not the target audience.
I don’t really want to keep engaging on this—as almost everyone (including me) seems to agree, command-line parser minutiae just aren’t that important—but I guess I will just conclude by saying that I believe that anchoring effects have led many programmers to consider any dependency smaller than, say, Electron to be not a big deal (and many think Electron’s fine too, obviously), whereas my experience has been that the difference between good and bad products usually hinges on many such ‘insignificant’ choices combining in aggregate.
Assuming whichever command-line parser one uses operates above a certain baseline—and I believe all of the argparse libraries in that benchmark do—it seems particularly silly to make wasteful choices here because this is such a small part of an application. Choosing wastefulness because it’s technically possible, then rationalising the waste by claiming it increases velocity/usability/scalability/whatever without actually verifying that claim because it’s ‘not worth thinking about’, seems more problematic to me than any spectre of premature or ‘unnecessary’ optimisation. I hope to find better ways to communicate this in future.
If someone cannot distinguish between the impact of choosing Electron over QT (or GTK or whatever) and the impact of choosing clap over argh, they were never going to make good engineering decisions to begin with. There’s no slippery slope here.
Hmm isn't optimizing to save 690KiB for an optimised release binary and getting incremental builds to be significantly less than 400ms actually much closer to the after-mentioned "every web site must be able to scale like Facebook” type logic" ?
The “every website must scale like Facebook” mindset is premature optimization driven by hypothetical future needs exactly what YAGNI advises against. But in your case, you’re investing time upfront to avoid a heavier dependency that already works and has no clear downside for the majority of users.
If you don’t actually need ultra-small binaries or sub-200ms compile times, then replacing Clap just in case seems like a textbook case of violating YAGNI rather than applying it.
> But in your case, you’re investing time upfront to avoid a heavier dependency
This is very confusing to me. What of this API[0], or this one[1], requires “investing time upfront”? With argh, you already know how to use all the basic features before you even start scrolling. These crates are all practically interchangeable already with how similarly they work.
It is only now that I look at clap’s documentation that I feel like I might understand this category of reply to my post. Why does clap need two tutorials and two references and a cookbook and an FAQ and major version migration guidelines? Are you just assuming that all other command-line parsers are as complicated and hard to use as clap?
Neither of those libraries provide cross-shell completions, or coloured output, or "did you mean" suggestions, or even just command aliases, all of which I would consider basic features in a modern CLI. So you need to invest more time to provide those features, whereas they just exist in clap.
That's not to say that clap is always better, but it is significantly more full-featured than the alternatives, and for a larger project, those features are likely to be important.
For a smaller project, like something you're just making for yourself, I can see why you'd go for a less full-featured option, but given there's not much difference between clap and, say, argh that I feel like I'd get much benefit out of argh. If you're really looking for something simple, just use lexopt or something like that, and write the help text by hand.
In engineering, principles exist to guide you toward a preferred outcome. That’s what ultimately matters, not how many times you get to write YAGNI in your commit messages.
And you disregard user experience and other developer experience with your own custom parsing code.
Acts as if there's no trade-off whatsoever in your own decision and my way is holier than thou in engineering is beyond sad.
I can understand this, if we were talking about JavaScript CLIs that requires GBs of dependencies. But 690KiB for modern computing is a drop in the ocean. It is not something you should base or make a consideration of unless you were doing embedded programming.
690KiB is a far compromise if Clap provided, for example, better performance or better code readability and organization. The benchmarks you provided shows the performance is practically the same which is close to no library usage.
I did do a bit of CLI work (I try to maintain https://github.com/rust-starter/rust-starter) and will always pick up clap. It just happens that even for the simplest CLIs out there, things can get hairy really fast. A good type interface that let me define my args as types and have extras on the fly (ie: Shell completion code) is worth the less than 1MiB overhead.
That 690KB savings is 1/97000th of the RAM on the machine I develop and run most of my Rust software on.
If I ever encounter a single howto or blog post or Stack Overflow answer that tells me how to use Clap to do something 5 minutes more quickly than with an alternative, it’s paid for itself.
Amdahl’s Law says you can’t optimize a system by tweaking a single component and get more than that component’s total usage back. If Clap takes 1% of a program’s resources, optimizing that down to 0 will still use 99% of the original resources.
At this point you're just flexing that you have 96GiB machine. (Average developer machines are more like 16GiB)
But that's not the point. If every dependency follows same philosophy, costs (compiler time, binary size, dependency supply chain) will add up very quickly.
Not to mention, in big organizations, you have to track each 3rd party and transitive dependency you add to the codebase (for very good reasons).
I can write and have written hand-tuned assembly when every byte is sacred. That’s valuable in the right context. But that’s not the common case. In most situations, I’d rather spend those resources on code ergonomics, a flexible and heavily documented command line, and a widely used standard that other devs know how to use and contribute to.
And by proportion, that library would add an extra .7 bytes to a Commodore 64 program. I would have cheerfully “wasted” that much space for something 100th as nice as Clap.
I’ve worked in big organizations and been the one responsible for tracking dependencies, their licenses, and their vulnerable versions. No one does that by hand after a certain size. Snyk is as happy to track 1000 dependencies as 10.
96? It sounds more like 64 to me, which is probably above average but not exactly crazy. I've had 64 GB in my personal desktop for years, and most laptops I've used in the past 5 years or so for work have had 32 GB. If it takes up 1/4700 of memory, I don't think it changes things much. Plus, argument parsing tends to be done right at the beginning of the program and completely unused again by the time anything else happens, so even if the parsing itself is inefficient, it seems like maybe the least worrisome place I could imagine to optimize for developer efficiency over performance.
> I got a nickel every time I spent 15 minutes replacing a trivial use of clap with pico-args and thus reduced the binary size and compile time of some project by at least 80%, I would have at least three nickels.
Hahaha, awesome. Thanks for the pico-args recommendation.
It supports the bare minimum.
I sure would like deriving-style parsing and --help auto-generation.
I think deriving-style unavoidably causes build time and complexity.
But it could be done without the dependency explosion.
Among the ones you recommend, argh supports deriving, auto-generates --help and optimizes for code size. And its syntax is very comparable to clap, so migrating is very easy. gumdrop seems very similar in its feature set (specifying help strings a little differently), but I can't find a defining feature for it.
> Why would you default to the biggest, slowest option?
Because it's not very big, nor very slow. Why wouldn't you default to the most full-featured option when its performance and space usage is adequate for the overwhelming majority of cases?
> Why wouldn't you default to the most full-featured option when its performance and space usage is adequate for the overwhelming majority of cases?
This is the logic of buying a Ford F-150 to drive your kids to school and to commute to the office because you might someday need to maybe haul some wood from the home improvement store once. The compact sedan is the obviously practical choice, but it can’t haul the wood, and you can afford the giant truck, so why not?
> This is the logic of buying a Ford F-150 to drive your kids to school and to commute to the office because you might someday need to maybe haul some wood from the home improvement store once.
No, it's like buying the standard off the shelf $5 backpack instead of the special handmade tiny backpack that you can just barely squeeze your current netbook into. Yes, maybe it's a little bigger than you need, maybe you're wasting some space. But it's really not worth the time worrying about it.
If using clap would take up a significant fraction of your memory/disk/whatever budget then of course investigate alternatives. But sacrificing usability to switch to something that takes up 0.000000001% of available disk space instead of 0.0000001% is a false economy, the opposite of practical; it feels like a sister phenomenon to https://paulgraham.com/selfindulgence.html .
Well you hit the nail on the proverbial head. The compact will handle 99% of people's use-cases, the truck will handle 100%. People don't want the hassle of renting something or paying for help for the 1% of cases their compact wouldn't handle.
Believe it or not, I'm with you; I live somewhere where it's sunny all year round, so I get around with a motorcycle as my primary transportation year-round and evangelize them as the cheap alternative to people struggling with car-related payments. But no, my motorcycle isn't going to carry a 2x4. Someone who cares about supporting that, even if they only need to do so exceptionally rarely, is gonna buy a truck. And then they won't have the money to buy a motorcycle on the side.
Not sure why you’re being downvoted. I also don’t like oversized motor vehicles but I think the parable is sound;
If the effort of switching out when you need the last 1% is higher than whatever premium you will pay (compilation time/fuel cost) - especially as a small ongoing cost, people will likely choose it.
I’m not saying this as if its wisdom into the future, only in that we can observe it today with at least a handful of examples.
These are not even remotely equivalent scenarios. If I want to remove clap as a library, I just remove it. If I buy an F150 I now have spent a lot of money and it's mostly gone so replacing it is significantly more expensive. It also burns more fuel.
Maybe. If your build is too slow, fix it. But pre-emptively microoptimising your build time is as bad as pre-emptively microoptimising anything else. Set a build time budget and don't worry about it unless and until you see a risk of exceeding that budget.
This is an incredible exaggeration. The vast majority of these projects don't even approach it. I use clap in a number of projects and they compile in just seconds.
I use clap everywhere for this, but I'm not sure that I agree that it has a "great" experience. It does the job and I've no need to reach for another tool. But it can be frustrating sometimes.
There are many other command-line parsers to choose from that do all the key things that clap does, with half or less the build cost, and most of them with 30x less binary overhead[0]. argh is under 4kloc. gumdrop is under 2kloc. pico-args is under 700loc. What is the value of that extra 10kloc? A 10% better parser?
I am not saying there is no room for a library like clap—it is, at least, a triumphant clown car of features that can handle practically any edge-case anyone ever thought of—but if I got a nickel every time I spent 15 minutes replacing a trivial use of clap with pico-args and thus reduced the binary size and compile time of some project by at least 80%, I would have at least three nickels.
Just to try to pre-empt arguments like “disk space is cheap”, “compiler time is cheaper than human time”, etc.: there are no golden bullets in engineering, only trade-offs. Why would you default to the biggest, slowest option? This is the “every web site must be able to scale like Facebook” type logic. You don’t even have to do more work to use argh or gumdrop. If clap ends up having some magic feature that no other parser has that you absolutely need, you can switch, but I’ve yet to ever encounter such a thing. Its inertia and popularity carry it forward, but it is perhaps the last choice you should pick for a new project—not the first.
[0] https://github.com/rosetta-rs/argparse-rosetta-rs