You're presuming man pages are primarily meant to serve as a reference. But I rarely need man-pages as a reference†.
Most of the time, if I'm looking up a man-page for something, it's because I've just installed a new package that sounded like it would solve a problem and then did a dpkg-query(1) to find out what binaries came with it—or used apropos(1) to find a relevant binary already installed—and now I want to know what the uses of a given binary are and whether those uses include solving my particular problem.
† Well, except for the utilities with absolutely horrible command-line UX-design, like tar(1) or ps(1) or rsync(1), where I just memorize the options I need for my usual case, and then have to look in the man page to do anything novel.
Let me rephrase: you assume that it makes sense for manpages to continue serving primarily as a reference—that this is the primary use-case people have for the standardized program documentation that ships with their distro packages.
Shipping a reference to a binary with that binary may have made sense before the internet. But nowadays, it's the opposite.
• Complex programs with many options have (sometimes dozens of) websites documenting them thoroughly. (Try searching with any search-engine for "wget mirroring", for example; the number and complexity of the results is overwhelming.)
• Meanwhile, for the simple "corner-case" programs, you really hope that they shipped with docs—because seemingly nobody else out there on the web cares to bother documenting them. With a lot of these little programs, the only web doc you can find are, in fact, online mirrors of their man-page.
For the popular-and-complex programs, man-pages are just redundant, because everyone will document what they did to achieve whatever. But for the simple-but-weird programs—the ones for which man-pages aren't just redundancies—if the man-page doesn't give usage, then nothing is going to give usage.
Now, I can understand why man-pages for these little utilities are the way they are. These programs are usually created by a single author, so time spent writing docs is time not spent fixing bugs or scratching their itch or whatever else. And an options reference certainly is the "minimal normalized form" of documentation: it lets others brute-force combinatoric-search the space of invocations until they find some combination that Works For Them™. Basically, you can (through a lot of trial and error) generate a cookbook from an options reference. So the author probably doesn't feel a strong need to add anything beyond an options reference, because the people who really need to solve the problem their binary solves are willing to go to that effort.
But if you're a distro downstream packager, and it's your job to make your distro easy for people to use, you should have every incentive to submit upstream patches to said author, with manpage additions of cookbook example usages resulting from your trial-and-error experimentation with their program.
Annoyingly, you, as a distro packager, probably don't have time to do that trial-and-error experimentation, especially if the utility serves a niche use-case that you don't even understand. That—and not the fact that "manpages should be a reference and nothing more"—is most of the reason manpages continue to be the way they are.
Man pages are written (ideally) as the authoritative reference on your system, where you can go to find information. That sort of thing needs to exist somewhere, it has a clear use case.
An ideal man page is concise, informative, and complete. Learning to read them is like learning to read scientific literature: a pain, but once you figure it out, you're at a higher level.
>you assume that it makes sense for manpages to continue serving primarily as a reference
No, I don't. But that's what they are, and what they're meant to be. Complaining that they are is like complaining that Haskell is functional: It may not be ideal for your circumstances, and you're welcome to make that known, but there's no use complaining about it, because it's the entire point.
• The primary win for all of those utilities would be in separating their horribly-large arrays of top-level runtime switches into subcommands that each have a restricted, learnable set of runtime switches relevant only to that subcommand. (See: git, docker, lvm, ip, ufw).
• As well, for top-level switches that are really preferences—that is, switches that don't apply to a use-case, but rather to a user—read those from ~/.config/foo/foorc with defaults in /etc/foo/foorc. Don't expose them at all as runtime switches. If there are features that would be customized by different wrapper-client UIs, provide a --config-file=[file] switch that takes a config-file that sets those. And for special drivers, like Makefiles or CI setups, that have a matrix of different UI options they might want to ask for, you can provide FOO_OPTION env-vars. (See: curl, compiler toolchains, ffmpeg, apt). In ffmpeg's case, there are even config-file presets—basically little plug-ins of config options you can enable as a group. You can dump new files into the /etc/ffmpeg/presets or ~/.config/ffmpeg/presets to add them as presets.
• For switches that are exposed individually, but which are also aggregated into switch-group-setting switches, in most cases there's literally no use-case for setting the individual switches, only the switch-group-setting switch. Remove the individual switches. Just because the program could theoretically be configured to do X compatibility thing for old-arch Foo, but not Y other old compatibility thing for old-arch Foo (where any time you're talking to Foo you always need both), you don't have to expose both an X switch and a Y switch. Just expose a --foo-compat switch and be done with it. People trying to use your binary on other weird OSes shouldn't be doing so by combining tons of option-switches; they should sit down and port your program to that OS, by finding the place in your program where you've defined {arch -> shim flag set} mappings, adding a new mapping for their arch, and exposing it as a new external --bar-compat switch.
• Speaking of compat flags, try just making such behaviors always-on when you build targeting the relevant arch, rather than needing to specify them at runtime. The only real use-case for runtime compat switches is as an IPC/RPC client talking to a server that expects the weird behaviour. And even then, your client should first try to auto-detect the server's expectations, and you should only add the runtime switch if that auto-detection turns out to be unreliable. (Meanwhile, If you're the IPC/RPC server, the compat settings belong in the config file. See: Samba, NFS, Netatalk, ...)
• If your binary manipulates state (like, say, how git manipulates repos), and you change the way it does so in a backward-incompatible manner, you might be tempted to add a runtime switch to turn the old behaviour back on to allow collaboration with people using older versions. If your state-store is at all extensible, though, you should instead add a field for a schema-version and a flag for pinning the state-store to that schema-version, as well as a subcommand to pin/unpin a given state-store's schema-version. Now your binary will upgrade the schema of its state-store by default, but will respect "protected" state-stores and perform only backward-compatible operations on those. (Importantly, this guides the code architecture into failing by default instead of doing something backward-incompatible; whereas, with backcompat switches, you're always forcing a state-store schema-migration on people by default every time you add new code, unless/until you add a matching backcompat-switch.)
• quiet/verbose and interactive/batch are silly "mode switches" to have in modern programs, just like forking is a silly way to do daemons when you've got a modern init(8). Put interactively-useful information on stdout, and do interactive prompting, if-and-only-if isatty(STDIN); put information of all levels of usefulness on syslog with appropriate error-level tags attached. It's up to the thing running your program to provide a PTY or not, and to filter your output or not. Be friendly to expect(1).
These guidelines together should trim each subcommand to a reasonable "visible" option-set. Now just ensure that typing "man command subcommand" gets you a separate man-page written just for the subcommand, and add bash/zsh completion for each subcommand and for the subcommands list itself.
> • The primary win for all of those utilities would be in separating their horribly-large arrays of top-level runtime switches into subcommands that each have a restricted, learnable set of runtime switches relevant only to that subcommand. (See: git, docker, lvm, ip, ufw).
In principle I agree. In practice, all of the commands you listed are multi-tool commands. For example, git overall can be described as a "content tracker"; only its subcommands can be described as doing one thing, and even then often they do multiple duties (to a newbie, git reset apparently does three different things). In contrast, two of the commands you listed (ps and rsync) do only one thing; list processes and copy files. The third, tar, does technically have multiple modes, but most of the switches apply to multiple modes (-f, compression settings, etc). How would you break these up into subcommands?
> • As well, for top-level switches that are really preferences—that is, switches that don't apply to a use-case, but rather to a user—read those from ~/.config/foo/foorc with defaults in /etc/foo/foorc. Don't expose them at all as runtime switches. If there are features that would be customized by different wrapper-client UIs, provide a --config-file=[file] switch that takes a config-file that sets those. And for special drivers, like Makefiles or CI setups, that have a matrix of different UI options they might want to ask for, you can provide FOO_OPTION env-vars. (See: curl, compiler toolchains, ffmpeg, apt). In ffmpeg's case, there are even config-file presets—basically little plug-ins of config options you can enable as a group. You can dump new files into the /etc/ffmpeg/presets or ~/.config/ffmpeg/presets to add them as presets.
Some would argue that this is an anti-pattern and that such configuration should reside in your shell as an alias or function. Regardless, with the possible exception of ps, for the commands you listed, which options exactly would you move into a configuration file? It would likely not be a good idea, for example, to say that all tar -c commands should imply -J, since that would break lots of scripts.
> • For options that are aggregated into option-set flags, where there's literally no use for the options outside of their use in the option-set flag, remove the individual options. Just because the program could theoretically be configured to do X compatibility thing for old-arch Foo, but not Y other old compatibility thing for old-arch Foo, you don't have to support both an X and Y option. Just expose a --foo-compat option and be done with it. People trying to use your binary on other weird OSes shouldn't be doing so by combining tons of option-switches; they should sit down and port your program to that OS, by finding the place in your program where you've defined {arch -> shim flag set} mappings, adding a new mapping for their arch, and exposing it as a new external --bar-compat switch.
I don't understand what this means. Can you give a concrete example?
> • Speaking of compat flags, try just making such behaviors always-on when you build targeting the relevant arch, rather than needing to specify them at runtime. The only real use-case for runtime compat switches is as an IPC/RPC client talking to a server that expects the weird behaviour. And even then, your client should first try to auto-detect the server's expectations, and you should only add the runtime switch if that auto-detection turns out to be unreliable. (Meanwhile, If you're the IPC/RPC server, the compat settings belong in the config file. See: Samba, NFS, Netatalk, ...)
Again, I don't understand what this means.
> Those four things together should trim each subcommand to a reasonable "visible" option-set. Now just ensure that typing "man command subcommand" gets you a separate man-page written just for the subcommand, and add bash/zsh completion for each subcommand and for the subcommands list itself.
I agree with this conclusion given the premises you provided, but the issue is that the premises themselves appear to be incorrect.
Broadly speaking, most complaints about the commands you listed originate in poor understanding of their interfaces; everybody thinks they're hard, so nobody ever reads the man page and learns how to use them correctly. For example, complaints about not knowing tar -z from -j are mostly a result of not knowing that GNU and at least one BSD tar know how to automatically select compression format in at least some cases. A similar situation is the cause of "ps -aux" and such monstrosities as "rsync -arlv -e ssh dir host:dir".
1. They commands I listed seem like "inherently multi-tool" commands because they've been refactored into multi-tool commands. ip is (mostly) a refactoring of ifconfig(1), but also of ethtool(1) and arp(1) and a few other things. A proper multi-mode refactoring of ps(1) would integrate overlapping utilities like pgrep(1) and fuser(1)/lsof(1), and then split those back down along separate lines.
A proper multi-subcommand refactoring of tar, I would think, might involve an interestingly different usage—a "fluent OOP interface" involving either a series, or pipeline, of invocations, all involving tar subcommands. For example:
tar manifest create --new-root=/ . | tar archive assemble --extended-attributes --timestamps --compression=lzma > foo.tar.xz
or, equivalently:
tar manifest create --new-root=/ . > foo.manifest
tar archive new --extended-attributes --timestamps --compression=lzma > foo.tar.xz
tar archive insert foo.tar.xz foo.manifest
then, later:
tar manifest read foo.tar.xz | grep -E '...' | sed '{ ... }' | tar archive extract foo.tar.xz
Using some file-descriptor introspection trickery, none of the steps except for the last one would actually have to stream bytes through them.
2. One interesting thing about ps is that it takes two separate sets of switches—BSD switches and GNU switches. Given that any given user will only want to use one or the other, these two interfaces should be broken into two utilities (likely using argv[0] detection, like gzip(1) and zcat(1)), with separate man-pages. Setting which one plain "ps" means when ps is called TTY-interactively should be a configuration-file thing.
3. Look at the man-page for GNU coreutils ln(1) or cp(1). They're messes of compat-switches, and compat-switch aggregates, aimed at allowing you to reproduce the default behaviour of the OS you're used to on any-and-every OS you can compile coreutils for. This configuration doesn't belong in argv[]; it belongs in a ./configure script, or in a ~/.config/coreutils/ui-rc file. (Some of the switches are maybe for things you might want to do even in your initrd before you've got the rootfs mounted; those are what env-vars are for.)
4. rsync(1) has options for copying ACLs, copying Extended Attributes, "handling sparse files efficiently", and so on. rsync should just try to do these things automatically, and fall back to not doing them if the source or dest doesn't have support for them. There's no use-case where both the source and target support EAs but you don't want them copies. There's also no use-case where one or the other doesn't support EAs and so you want to fail the sync instead of copying. There's no need for these switches, just like there's no need for an FTP PASV switch. It can be detected.
---
As an aside:
> not knowing that GNU and at least one BSD tar know how to automatically select compression format in at least some cases
For extraction, sure. But why not for creation? RAR and Zip can pick the best compression they "know how to" do at the current moment, and offer compression speed/quality presets that actually involve picking different algorithms.
The real confusion over tar's compression switches is more fundamental; it comes from the fact that a tar archive is "wrapped in" an arbitrary compression-container file-format, without there being any standard for those to allow you to detect something as a "gzipped tar" rather than "gzipped opaque data", and thus the inability of tar, or your OS, to recognize an "[unknown compressor]ed tar" file.
The gordian knot of tar(1)'s switches is for tar—already having treated the compressor as a subprocess rather than a pipe-destination for a long time now—to just begin adding a "this is a tar archive with a compressed stream of archive files inside it" header to the outside of the compressed stream. Even though full backcompat is still possible after defaulting to such a change (user pref files! env vars! in-file hinting that the previous binary can recognize!), everybody's too lazy to want to make that sort of change.
> A proper multi-subcommand refactoring of tar, I would think, might involve an interestingly different usage—a "fluent OOP interface" involving either a series, or pipeline, of invocations, all involving tar subcommands. For example:
... it was my understanding that the goal was to make the interface easier, not three times as complicated.
> 2. One interesting thing about ps is that it takes two separate sets of switches—BSD switches and GNU switches. Given that any given user will only want to use one or the other, these two interfaces should be broken into two utilities (likely using argv[0] detection, like gzip(1) and zcat(1)), with separate man-pages. Setting which one plain "ps" means when ps is called TTY-interactively should be a configuration-file thing.
"bsdps" and "gnups" seem much worse than "ps" and "ps -".
> 3. Look at the man-page for GNU coreutils ln(1) or cp(1). They're messes of compat-switches, and compat-switch aggregates, aimed at allowing you to reproduce the default behaviour of the OS you're used to on any-and-every OS you can compile coreutils for. This configuration doesn't belong in argv[]; it belongs in a ./configure script, or in a ~/.config/coreutils/ui-rc file. (Some of the switches are maybe for things you might want to do even in your initrd before you've got the rootfs mounted; those are what env-vars are for.)
cp I'll kind of buy, but ln? It's only got 14 real options, none of which can be described as a "compat switch". Again, I'll have to ask you for a specific example. Here's an example of a specific example: "cp -Q does <weird thing> that HP-UX does and nobody else".
> There's no use-case where both the source and target support EAs but you don't want them copies.
Security?
> There's also no use-case where one or the other doesn't support EAs and so you want to fail the sync instead of copying.
Wait, what? "fail silently"?
> There's no need for these switches, just like there's no need for an FTP PASV switch. It can be detected.
copying ACLs isn't like PASV, it's more like ASCII vs binary mode. trying to auto-detect makes the situation worse.
> The gordian knot of tar(1)'s switches is for tar—already having treated the compressor as a subprocess rather than a pipe-destination for a long time now—to just begin adding a "this is a tar archive with a compressed stream of archive files inside it" header to the outside of the compressed stream. Even though full backcompat is still possible after defaulting to such a change (user pref files! env vars! in-file hinting that the previous binary can recognize!), everybody's too lazy to want to make that sort of change.
how about "updating the compression algorithm requires everybody to install a new (de)compressor and until they do they can't open your files"?
In practice --help output is mostly useless—it’s neither simple enough for a quick overview nor comprehensive enough to actually know what I can do. Tried “less --help” lately? It spits out ten pages of output!
Plus, every program does --help differently. Does it spit out short output or tons of output so that I need a pager? Does it fork and execute a pager itself, or do I have to run the program again and pipe it to less manually? And then it turns out some programs print --help output to stderr, not stdout, so the pager doesn’t display any output until I do “blahblah --help 2>&1 | less”.
A good manpage beats --help all hollow. It’s simultaneously more brief and more (usefully) verbose. The Synopsis section provides a quick overview of flags, if I just need a quick reminder to jog my memory. But if I want to know the details of a particular flag, I can just scroll down to it or search for it. Mandoc, the default manpage formatter in OpenBSD, even has semantic search built in. It uses more(1)'s built‐in tags functionality to take me straight to whatever flag I want. (You can get this functionality on Linux too by installing mandoc: http://mdocml.bsd.lv/)
Really, horrendous and arcane manpages are a trope that mostly exists in the Linux/GNU world. In BSDs, especially OpenBSD and FreeBSD, clear, simple documentation is a priority. No need for --help or “bropages” when the actual manpages are perfectly clear. For example, check out OpenBSD’s signify(1) manual (http://man.openbsd.org/signify.1), used for cryptographic signing and verification. Compare that to the monstrosity that is gpg(1)… (https://www.gnupg.org/documentation/manpage.en.html)
Most of the time, if I'm looking up a man-page for something, it's because I've just installed a new package that sounded like it would solve a problem and then did a dpkg-query(1) to find out what binaries came with it—or used apropos(1) to find a relevant binary already installed—and now I want to know what the uses of a given binary are and whether those uses include solving my particular problem.
† Well, except for the utilities with absolutely horrible command-line UX-design, like tar(1) or ps(1) or rsync(1), where I just memorize the options I need for my usual case, and then have to look in the man page to do anything novel.