Linus Torvalds Injects Tabs To Thwart Kconfig Parsers Not Correctly Handling Them
"if you can't parse tabs as whitespace, you should not be parsing the kernel Kconfig files." ~ Linus Torvalds
This is what we got after people sent him into PC training. The OG Linus would say something like "if you're a piece of s* that can't get over your a** to parse tabs as whitespace you should be ashamed to walk on this planet let alone parsing the kernel Kconfig files. What a f* waste of space."
And honestly, I find his phrasing today far more damming with its conciseness. It screams leadership.
The Gordon Ramsey of programing
I think your keyboard is broken. It's inserting random stars in your text.
Yours is too, its posting a silly link at the bottom which makes you look like you're a Facebook mom in 2003.
Dear Mark Zuckerberg
With this statement, I give notice to Facebook it is strictly forbidden to disclose, copy, distribute, or take any other action against me based on this profile and/or its contents. The content of this profile is private and confidential information. The violation of privacy can be punished by law (UCC 1-308- 1 1 308-103 and the Rome Statute. NOTE: Facebook is now a public entity. All members must post a note like this. If you prefer, you can copy and paste this version. If you do not publish a statement at least once it will be tacitly allowing the use of your photos, as well as the information contained in the profile status updates. FACEBOOK DOES NOT HAVE MY PERMISSION TO SHARE PHOTOS OR MESSAGES.
No-no-no. "Whoever can't handle tabs as whitespaces should heat their IQ, it is below freezing right now". Or classic "should be retroactively aborted".
Nice one. Maybe e can make a website "whatwouldlinussay.com"?
Complaining about PC but type the word "fuck"? Genius.
The associated forum post contains a much more endearing translation.
Why would anyone NOT parse a tab as whitespace? Like, python really wants you to use spaces but will still let you use tabs if you are consistent.
There's the .tsv format too.
Even then, unless for some reason you code on paper, there is no excuse to not have a formatter, that enforces one style, or the other
I'm honestly on Torvalds' side here.
Tabs are a necessary part of the tooling and configuration files. Any tool which doesn't properly handle files that are correctly formatted for other tools is... a broken tool.
I'll take you up on that. ChatGPT, translate this commit message to actual Linus style:
Subject: Regarding Commit d96c36004e31
So apparently, this commit thought it was a brilliant idea to strip away a tab because some third-rate kernel config parsing tool couldn't handle it. Seriously? We're bending over backward for these tools now?
Listen up, if your parsing tool throws a tantrum over a tab, it's time to rethink your life choices. Tabs are as fundamental to Kconfig files as oxygen is to humans. If you can't handle tabs, you have no business parsing kernel configurations.
And let's not pretend this is some isolated incident. The whole tab vs. space saga is like a bad sitcom rerun from the '70s. Sure, 'make' might have its quirks, but that doesn't mean we should stoop to that level of incompetence.
I'm tempted to flood our Kconfig files with tabs just to prove a point, but let's not make a mess where it's not needed. However, if this idiocy persists, don't be surprised if I start tabbing everything in sight.
Let's get our priorities straight, people. We're building a kernel here, not babysitting fragile parsing tools. So let's fix this mess and move on to more important things.
I tried it with Claude 3 Opus, even better:
Alright, listen up you little shits. Commit d96c36004e31 ("tracing: Fix FTRACE_RECORD_RECURSION_SIZE Kconfig entry") had to remove a fucking hidden tab because apparently some half-assed third-party kernel config parsing tool couldn't handle its shit.
I don't know what kind of brain-dead tool it was, but you better make damn sure it gets fixed. Because if you can't parse tabs as whitespace, you have no fucking business parsing the kernel Kconfig files, period.
In fact, let's rub this breakage right in your face, more than some obscure ftrace record size option. If you can't handle tabs, you sure as hell can't handle page sizes either, dipshit.
Yeah, I know the whole tab-vs-space clusterfuck is sadly a classic Unix thing, and 'make' is notorious for being a broken piece of crap in that regard. But guess what? That doesn't make it fucking okay, got it?
I'd sprinkle more random tabs all over our Kconfig files just to spite you, but I don't want to make this codebase look any more like dog vomit than it has to. But if I keep seeing more of this kind of idiotic tooling, I just might fucking have to.
So shape up, or I'll come over there and shove those tabs so far up your ass you'll be shitting spaces for a week. Fuck!
“ok, now add a metric shit ton of swearing and further belittle parsers who can’t deal with tabs.”
Listen up, if your parsing tool throws a tantrum over a tab, it's time to rethink your life choices. Tabs are as fundamental to Kconfig files as oxygen is to humans. If you can't handle tabs, you have no business parsing kernel configurations.
Heh this is so succinctly scathing, but also dead on ....
I'm tempted to flood our Kconfig files with tabs just to prove a point, but let's not make a mess where it's not needed. However, if this idiocy persists, don't be surprised if I start tabbing everything in sight.
This is genius.
I am more than sure that Linus wrote the original message as he would normally do, and then made it clean and pretty with an AI. Sometimes I resort to this option too.
The Robustness Principle may seem like little more than a suggestion, but it is the foundation on which many successful things are based.
To boil it down to meme-level old-school Torvaldsry: Assume everyone else is a f--king idiot who can barely do what they're supposed to and expect to parse their files / behaviour / trash accordingly.
If you do not do this, you are, without doubt, one of those f--king idiots everyone else is having to deal with. If you do do this, it does not guarantee that you are not a f--king idiot. Awareness is key.
Examples where this works: Web browser quirks mode; Driving a car; Measure twice, cut once. This latter one is special because it reveals that often, the f--king idiot you're trying to deal with is yourself.
Assume everyone else is worse.
Fun corollary: In altering his behaviour towards f--king idiots people who should know better, Linus has learned to apply the robustness principle to interpersonal communication.
using a rly bad word but pretending not to is kinda weird
Sometimes it works well as a stylistic choice. It's not pretending not to use a bad word, but rather drawing attention to the fact that you're deliberately being a little bit naughty with a wink to the reader. It's like the absurdity of what happens when you find a stranger in the Alps.
Robustness Principle
It's a principle for brittleness. First you get implementation-defined behaviour, then bug-compatible software.
Torvalds believes parsers unable to handle tabs shouldn’t be parsing kernel Kconfig files, aiming to force fixes.
Stern but just
Based Linus is based.
I do the same to people who refuse to follow specifications they agreed to follow.
There is a slight satisfaction to get back at them for continually delivering much lower quality than is required.
But it really is to cover me. Because, it always happens, later in the future that edge case comes up, and everything breaks. And management is ready to blame me. But then I show them that I tested the edge case before the conclusion of the project. And that programmer ignored my emails, and that I told management these edge cases weren't covered. But then management signed off on calling it complete. And suddenly management is no longer red with fury. And they usually won't allow me time to fix it. So the can gets kicked down the road until the next time that edge case fails.
Me: "This is the file format we agree on."
They: "Yep, that's what you'll get."
They: "Why is your script not working?"
Me: "Idk what was the last file you put through it?"
In their defense, they got the file from a third party that we both previously assumed competent enough to follow explicit written instructions. Guess there's a lesson in trust...
Unconventional way to take acid, but I support him.
WE DO NOT BREAK USERSPACE
Invalid, not an ABI change. Besides, this does not break, it only reveals existing breakage.
The Register did a good article covering the change.
Source files should be conservative with the standard they expect from the developer, and parsers should be liberal in what they expect from the source, ie. allow deviations from the standard.
Python for example supposedly only allows 4 spaces for indentation, but as long as the developer is consistent most if not all Python interpreters will accept any kind of indentation.
Python for example supposedly only allows 4 spaces for indentation, but as long as the developer is consistent most if not all Python interpreters will accept any kind of indentation.
That's a recipe for disaster as your syntax is under-specced. You're right-up inviting programmers to produce programs with implementation-defined semantics.
Haskell (which also uses layout syntax) never had that problem as tabs were simply defined to be eight spaces, but that then led to issues with people setting different tab-widths in their editors and a flurry of syntax errors when they did "tabs for indentation, spaces for alignment". Which is why Haskell then moved ahead to outlaw tabs, I think it's still in the "throw a warning" phase but at some point it's going to be a hard error.
That's not to say that kconfig should do the same -- presumably they used tabs for a good reason, and all those other programs are simply not following the spec. Essentially including unit tests in the actual production files is a good move when you're dealing with that kind of situation.
Having a language dependent on indentation is absurd on the face of it. It's a ridiculous idea that should have been ridiculed from the outset.
No. Having a language depend on semicolons even though there's ways to do without, ways that don't even include layout if you don't want to, is well not absurd on the face of it it's hysterical raisins.
Haskell has one of the most admired syntaxes out there, and it's layout. It's clean, predictable, very simple and most of all intuitive rules. It makes sure that semantic structure always follows visual structure, thus provides a single source of truth why Algol-likes (i.e. everything that looks at least vaguely like C) have two.
I don't indent my Rust, I let rustfmt do that. All that automation and I still get into lots of missing or mismatched braces situations which literally never happen in Haskell because the structure of the program is visually obvious, you don't have to look for tiny squiggles to figure out what it is.
The bot vs the human moderator
bell rings
FIGHT!
Bonebag wins
what a mad lad!
I am no kernel or parsing expert here but how hard would it be to convert tabs to spaces? Is it like very finicky and is weirdly platform dependent that it wouldn't just be one of the first things that you do if you are writing a parser for anything?
grep -oP ' *' oops no tabs
cut -d ' ' -f 3 oops no tabs
People who use spaces are not taping space space space space space space space space. Their tapping the tab key all the same.
Wait…. What? You mean I’ve….
Sorry I didn't realize I had replied with I deleted my comment. I understand some editors allow you to set tab and you can set actual spaces, like in vi. However, personally I feel like hitting tab gives me the whitespace I want for readability already.
For programmatic parsing it is simple because it's just looking for an HT.
Yeah, same, but I always set my editor to insert spaces instead when I hit Tab, so that the spacing won't be different in other environments.
I apologize. I didn't know I had replies when I deleted my post. Yeah I know you can set that behavior in some editors. And other than what I just replied with on another comment, I like tabs because I also don't have to worry about styling guides that some set down as '2 spaces' or '3 spaces' or '5 spaces' or whatever. It is basically just universally a horizontal tab.
Tabs are a dark pattern confirmed.
The war is over, long live spaces.
I will fight tooth and nail with anyone who agrees with this decision. Not because Torvalds did it, but because fuck you use tabs you goddamn neanderthals
Notice Linus doesn't use tabs; he just thinks the parser shouldn't die when it sees them.
Yeah and that makes sense. But I still choose to die on the hill of tabs or nothing. That's just how anal I am about my code.
And I realize there are niche scenarios where tabs don't make sense, but I could care less about those. Afaik, even assemblers support tabs. If you have tabs, I feel it is almost sacrilegious not to use them for indentation.
This isn't about using one or the other, it's about not exploding when you encounter a tab.
Your word choice reminds me of Linus before he went into this kind of rehab thing.
Yeah I liked him better then /s
Wow, super newsworthy. Even included some bs about "colorful language". Can't let a thing like that go by without trying to entice some internet drama.
"newsworthy"??
Who said news? The mailing list is my k-drama
Kernel-drama, my favorite.
I'm here for Linus zingers tho?
It's from Phoronix. This is kinda what they do. It's not worthy of the front page of the Wall Street Journal, but then, no one said it was....
They are kind of known for not being all that clickbaity. They just report on Linux and benchmark stuff
these days i think the only reason people want open source is to air the dirty laundry of software development