Yeah the documentation (if it even exists) of most projects is usually clearly written by people intimately familiar with the project and then never reviewed to make sure it makes sense for people unfamiliar with it. But writing good detailed documentation is also really hard, especially for a specialist because many nontrivial things are trivial to them and they believe what they're writing is thorough and well explained even though it actually isn't.
You have to assume some level of end user knowledge, otherwise every piece of documentation would start with "What a computer does" and "How to turn your computer on."
I've found the best practice is to list your assumptions at the top of the article with links to more detailed instructions.
This is why I did a “walkthrough test” when I had to write documentation on this sort of thing. I’m a terrible technical writer, so this shit is necessary for me.
I grabbed my friend who knows enough about computers to attempt this, but not enough about infrastructure to automatically know what I meant when I was too vague.
Took two revisions, but the final document was way easier to follow at the end
Reminds me of the time I asked a question about a Magic: The Gathering card tomy local game store's Facebook page. The card was Sublime Archangel. I asked what happened if it gave a creature Exalted that already had one. Someone sarcastically replied that it already says it on the card. I was a new player, how was I supposed to parse the phrase "If a creature has multiple instances of exalted, each triggers separately"? For all I knew that could mean that they didn't stack because they would need to trigger together. I didn't have the vocabulary to understand those things.
I try to remember this when explaining what I might believe are simple concepts to people because that person really upset me.
That's why blog posts rock. Most popular projects will have a dozen blog posts for different configurations. For example, when looking to set up NextCloud, I found docs for almost all combinations of the following:
Apache and Nginx configuration
running through Docker or directly on the host
MariaDB and Postgres configs (and SQLite, with proper disclaimers)
Collabora and OnlyOffice config
It does take some knowledge of each of the above if you need one of the few configs that's not available on a blog post, and some of the posts are outdated, but with a bit of searching almost everything is documented by someone on the internet.
This shouldn't be necessary (official docs should be more comprehensive), but at least it's available.
I'd rather have a great documentation than five different blog posts, where some of them might be outdated, wrong or insecure (and you only find out later).
But yes, they are helpful and easily available for popular software.
Okay, please point me to the blog posts that helped you with collabora/onlyoffice. Thanks have NEVER been able to get that to work with my nextcloud (currently using the Docker AIO).
The information that familiarity gives you is "WTF does this field means", and it's the only thing that's actually there. How you get a value and how a value is formatted are things no amount of expertise will save you from having to tell the computer, and thus you can't just forget about.
(And let me guess, the software recommended install is a docker image?)
I don’t think it’s (just) that. It’s also a different skill set to write documentation than code, and generally in these kind of open source projects, the people who write the code end up writing the documentation. Even in some commercial projects, the engineers end up writing the docs, because the higher ups don’t see that they’re different skill sets.
100% Agree, it feels like most documentation is written in a way that expects you to already know what it's talking about... When it's the documentation's job to teach me about it.
Honestly, as a newbie to Linux I think the ratio of well documented processes vs. "draw the rest of the fucking owl" is too damn high.
The rule seems to be that CLI familiarity is treated as though its self-evident. The exception is a ground-up documented process with no assumptions of end user knowledge.
If that could be resolved I think it would make the Linux desktop much more appealing to wider demographics.
That said, I'm proud to say that I've migrated my entire home studio over to linux and have not nuked my system yet. Yet... Fortunately I have backups set up.
Linux on the desktop almost never needs CLI interaction though. Maybe you'll need to copy/paste a command from the internet to fix some sketchy hardware, but almost everything works OOTB these days.
However, self-hosting isn't a desktop Linux thing, it's a server Linux thing. You can host it on your desktop, but as soon as you do anything remotely server-related, CLI familiarity is pretty much essential.
That depends on your use case for desktop linux of course. For me, yabridge is the tool I needed to run VSTs on Linux. Its CLI only as far as I know.
Don't get me wrong; I'm not afraid of the CLI. Its just some tools are assuming the end user is a server admin or someone with deeper than the upper crust knowledge of how Linux works.
Don't forget the situations where you find a good blog post or article that you can actually follow along until halfway through you get an error that the documentation doesn't address. So you do some research and find out that they updated the commands for one of the dependency apps, so you try to piece together the updated documents with the original post, until something else breaks and you just end up giving up out of frustration.
That sounds an awful lot like modifying an ESP32 script I've been trying to follow from a YouTube tutorial published a while back. Research hasn't uncovered anything for me to troubleshoot the issue so it's a really shit experience.
CLI familiarity is fine. CD, Nano, mkdir, rm. I am proficient with that. But I am not necessarily proficient with Docker (went with it because it worked nicely for another thing which was well documented and very straight forward). It's just I'm trying to self host stuff. Some things like Wordpress and Immich are straightforward. Some things aren't like Matrix and Mastodon. Lemmy is also notoriously bad.
I think if you're talking wider demographics your model OSs are (obviously) Windows and macOS. People buy into that because CLI familiarity isn't required. Especially with Apple products everything revolves around simplicity.
I do dream of a day when Linux can (at least somewhat) rival that. I love Linux because I am (or consider myself) intricately familiar with it and I can (theoretically) change every aspect about it. But mutability and limitless possibilities are not what makes an OS lovable to the average user. I think the advent of immutable Linux distros is a step in the right direction for mass adoption. Stuff just needs to work. Googling for StackOverflow or AskUbuntu postings shouldn't ever be necessary when people just want to do whatever they were doing on Windows with limited technical knowledge.
However on another note, if you're talking a home studio migration, not sure what that entails, but it sounds rather technical. I don't want to be the guy to tell you that CLI familiarity is simply par for the course. Maybe your work shouldn't require terminal interaction. Maybe there is a certain gap between absolutely basic linux tutorials and the more advanced ones like you suggest. Yet what I do want to say is that if you want to do repairwork on your own car it's not exactly like that is supposed to be an accessible skill to acquire. Even if there are videos explaining step by step what you need to do, eventually you still need to get your own practice in. Stuff will break. We make mistakes and we learn from them. That is the point I'm trying to get at. Not all knowledge can be bestowed from without. Some of it just needs to grow organically from within.
I write technical documentation and training materials as part of my job, and the state of most open source documentation makes me want to stab people with an ice pick.
Over my career, it's sad to see how the technical communications groups are the first to get cut because "developers should document their own code". No, most can't. Also, the lack of good documentation leads to churn in other areas. It's difficult to measure it, but for those in the know, it's painfully obvious.
Jesus, technical people are some of the worst communicators I've ever worked with.
It's not necessarily their fault though. Y'know who goes into technical jobs? People who often prefer to work with machines, physical stuff, laws of nature, that's who. And often because it's MUCH easier than working with people, at least for them.
On top of that, soft skills are HARD. Communication is HARD. It comes easier for some, but it's a skill like any other. It's the technical socialites, the diplomatic devs who become the best managers and leaders, due to the rarity of their hybrid skillsets.
I'm in the middle. Just technical enough to mostly understand the devs and understand the implications of plans, and just enough soft skills to turn that into decent documentation, emails, and working with clients.
SUCKS that I've gotten a taste of project management and hated the absolute fuck out of it. I probably would've been decent at it otherwise.
Yes. Here: "1.You aren't writing an SOP for smart or even capable people., every. Single. Person. Needs their hand held all the way through every step regardless of technical skill. "
"2.if you didnt state it needed to be done in the SOP, it will not be done when the end user follows the SOP"
"3.MAKE someone else run through your SOP without you being involved. If they can successfully achieve what they needed using your SOP > congrats. If not > fix the errors that brought you to this mess."
"4. Everyone is fucking stupid, be clear, and verbose." We're talking about where the start menu is, clicking on the "OK" for prompts, how to spell and type things out.
Change my given values per SOP and what it's for. But those are my main tenants.
I'll see about digging up recommendations if I can, but I'm on my phone right now.
My biggest single piece of advice would be this: Understand that your reader does not share your context.
What this means is that you have to question your assumptions. Ask yourself, is this something everyone knows, or something only I know? Is this something that's an accepted standard, or is it simply my personal default? If it is an accepted standard, how widely can I assume that accepted standard is known?
A really common example of this in self-hosting is poorly documented Docker instructions. A lot of projects suffer from either a lack of instructions for Docker deployment, because they assume that anyone deploying the project has spent 200 hours learning the specifics of chroot and namespaces and can build their own OCI runtime from scratch, or needlessly precise Docker instructions built around one hyper-specific deployment method that completely break when you try to use them in a slightly different context.
A particularly important element of this is explaining the choices you're making as you make them. For example a lot of self-hosted projects will include a compose file, but will refuse to in any way discuss what elements are required, and what elements are customisable. Someone who knows enough about Docker, and has lots of other detailed knowledge about the Linux file system, networking, etc, can generally puzzle it out for themselves, but most people aren't going to be coming in with that kind of knowledge. The problem is that programmers do have that knowledge, and as the Xkcd comic says, even when they try to compensate for it they still vastly overestimate how much everyone else knows.
OK, I said I'd try for examples later, but while writing this one did come to mind. Haugene's transmission-openvpn container implementation has absolutely incredible documentation. Like, this is top tier, absolutely how to do it; https://haugene.github.io/docker-transmission-openvpn/
Starts off with a section that every doc should include; what this does and how it does it. Then goes into specific steps, with, wonder of wonders, notes on what assumptions they've made and what things you might want to change. And then, most importantly, detailed instructions on every single configuration option, what it does, and how to set it correctly, including a written example for every single option. Absolutely beautiful. Making docs like this is more work, for sure, but it makes your project - even something like this that's just implementing other people's apps in a container - a thousand times more usable.
(I've focused on docker in all my examples here, but all of this applies to non-containerized apps too)
Well a good indicator is if I have to check the source code of a packaged program to understand what something does, the documentation is not good enough. And yes I've had to do this far too much.
as a chronic documentation reader, the best advice i can give is to document everything Anything that the user can and will potentially interact with, should be extensively documented, including syntax and behavior. Write it like you're coming back to the project in 5 years after having done nothing and you want to be able to skip right to using it. When we build something ourselves, we often hold a bit of internal knowledge from the design process that never quite goes away, so it's almost always a lot easier for us to reverse engineer something we've made, than it is for someone else with zero fore-knowledge to do it themselves.
Generally this can be a bit of a nightmare, but if you minimize the user facing segment it's not all that bad, because it's usually pretty minimal, and what would otherwise be a handful of pages, turns into 10 or maybe 15.
as for existing documentation, the i3wm user guide is really good, it's pretty minimalist but it leaves you enough to be able to manage.
I used to mock people who make YouTube videos that literally just walk through the documentation. Like bro, get some reading comprehension!
But then when I fumbled with some self hosting tutorials, those YouTube videos were the only thing that made sense, because they're explaining why and how.
Do you have any tips for writing professional documentation? I want to do some for my workplace but it's hard to know where to start, how to arrange it, etc
I worked alongside some technical writers in the early post-y2k years at SCO. This was before they sued IBM for code misuse and died by a million legal and PR cuts, thanks to the 'independent news' site launched by a 'recent ex-employee' to reframe things then and rewrite history after.
We had about 15 tech writers in the company, which when I first arrived seemed like a LOT. I'd never met one, and I'd taken a single tech writing course in college as a filler and found it unchallenging work; so I didn't value them at the time aside from filling a necessary role that your average nerd could surely fill. Then I saw their work; and it was amazing. It's one of the product's strong points, and 20 years later it's still so head-and-shoulders above the similar offerings by others and since, that it's a joy to read when I come across it.
Quite simply put, technical writers explain something in a logical, sensible way, where jargon doesn't blind-side the reader and layout and language are consistent and easy. Hell, spelling is correct; which is a big win over 90% of the current stuff. Tech writers are writers as Lance L said, and thus know about adjective order, prepositional placement, the difference between 'backup' and 'back up' and all its similar terms; and of course know why e-mail and traffic do not get an S as nouns - ever - even if the popular kids make everyone say it without thinking.
It's all simple-sounding stuff, and I was fooled into believing it was mundane; but when put together and written with an eye toward a common style it takes a stressful reader looking for a process or a parameter and induces calm for that brief moment required to get into the doc and find the sought-after bit.
Honestly, like the mentors we lost as a working society in the post-y2k bust when the c-suite cleared the ranks of things they didn't understand, the loss of good technical documentation has a generational effect and will take a massive, sustained effort to reverse.
The worst is when it's buried in Github issues or in a header file with thousands and thousands of lines of code. Yes I'm looking at you DearImGui, your documentation is awful and I'm already being generous.
I just recently started working with ImGui. Rewrite compiled game engines to add support for HDR into games that never supported it? Sure, easy. I can mod most games in an hour if not minutes.
Make the UI respond like any modern flexible-width UI in the past 15 years? It's still taking me days. All of the ImGui documentation is hidden behind closed GitHub issues. Like, the expected user experience is to bash your head against something for hours, then submit your very specific issue and wait for the author to tell you what to do if you're lucky, or link to another issue that vaguely resembles your issue.
I know some projects, WhatWG for one, follow the convention of, if something is unclear in the documentation, the issue does not get closed until that documentation gets updated so there's no longer any ambiguity or lack of clarity.
This reminds me of when I sent someone a program in a zip folder. Windows now opens zip folders by default, and it looks just like any other folder.
So of course they opened the zip and double clicked the exe, but everyone knows you can't open an exe inside a zip folder (at least, if the exe depends on the folders and files around it). If you try to, windows will extract the exe into a temp space, but leave all the dependencies behind. So the exe promptly crashes.
I didn't think I needed to specify "you need to extract the contents of the zip folder first, then run the exe." It feels like saying "you need to take the blender out of the box before you can use it. And not just the _base _ of the blender, you have to take out all the parts."
Some things just feel so much like second nature that we forget.
In many ways, the silky-smooth convenience offered by modern computer software makes everything much harder to learn about and understand. For anyone that used zip files before this Windows feature, the problem is obvious - but for younger people it's not obvious at all. Heck, a lot of people can't even tell whether or not a file is locally on their computer - let alone whether it is compressed in some other file.
Not that I know of. If I know it correctly (not doing it very often as I usually extract the whole content anyway) it just asks if I want to run the file. But I could be very wrong.
I totally and completely blame Microsoft for this. They do so many other ridiculous things in the name of not confusing the average tech illiterate user.
Clicking a Zip file and having it transparently open and treating it like a regular folder when it is not. This. THIS is borderline criminal.
There should be instructions that range from beginner, where every little step is included as well as major details as to why - all the way up to expert, which are just a few sentences.
Hay, how would you like writing documentation for all these open source projects? We would be ever greatful, you could even put your name in the credits!
One of the few things that Mac kind of got right. Every application is actually a deep tree with all kinds of crap all over the place but they never let the user see that.
A temporary one that you're expected to remove as soon as you've created the admin user(s) you need, but yes. It should only be there during initial setup and ideally removed before the server is ever exposed to the internet.
Yes because having a user remember to do something is a great line of defense, better than encrypting it from the get go. It should just be encrypted in the file.
I'm guessing this string can be whatever you want it to be.
But yeah, I agree in general, some of the docs can be pretty opaque. For example, I wanted to configure NextCloud w/ Collabora in Docker, and I kept getting errors when trying to do what a few sites recommended. I ended up figuring it out, but only through trial and error. I'm going to go through the same pain this weekend when I try out ownCloud Infinite Scale up and running to compare.
I had very similar experiences with OCIS. Got it all set up following the quick start guide, found extremely odd and unacceptable behaviour with storage space ballooning, start troubleshooting and find “oh you had to do this, this and this manually, it’s in the docs” It is in the docs, but never referenced by any other part of the docs. Because why would you mention the thing that the admin must manually set up in 100% of installs in your setup guide?
Anyway I’ve become that guy ranting on the internet that I don’t want to be. So just so you don’t suffer as much as I did; you have to create scheduled tasks via cron or your preference of scheduler to clean your uploads folder and data blobs. This also did not fix my specific issue and I ended up giving up on OCIS and sticking to Nextcloud.
I'm going to run both in parallel for a month or so before trying to get my SO to use it so I can better estimate the WAF. So far, NextCloud is good enough, but it's kinda slow (and I have Redis configured) despite being on pretty beefy hardware (Ryzen 1700 w/ 16GB RAM). I really hate PHP, so I'd prefer a project I can contribute to if needed. I worked w/ Go for almost 10 years, so OCIS would be a natural fit, but I'd still contribute patches for PHP if that really was the best tool for the job. But I'm not going to get involved unless the project already does what I need (my contributions would be for smaller bug fixes).
But yeah, the OCIS docs look kinda mediocre from the little I've read of them. But at least I don't need to mess w/ PHP config most likely and can hopefully just forward HTTP requests to it.
Matrix and its implementations like Synapse have a very intimidating architecture (I'd go as far as to call most of the implementations somewhat overengineered) and the documentation ranges from inconsistent to horrific. I ran into this particular situation myself, Fortunately this particular step you're overthinking it. You can use any random string you want. It doesn't even have to be random, just as long as what you put in the config file matches. It's basically just a temporary admin password.
Matrix was by far the worst thing I've ever tried to self-host. It's a hot mess. Good luck, I think you're close to the finish line.
funnily there's an.. ansible i think? project that makes selfhosting synapse easy as fuck, you basically just go "ansible deploy synapse" or whatever the syntax is and it does almost everything for you.
Matrix seemed interesting right until I got to self hosting it. Then, getting to know it from up close, and the absolute trainwreck that the protocol is, made me love XMPP. Matrix has no excuse for being so messy and fragile at this point. You do you, but I decided that it isn't worth my sysadmin time (especially when something like ejabberd is practically fire and forget).
You just need to figure out how many miliseconds since the unix epoch the media you want to purge was uploaded, and then offset the time to only purge that old or older.
I still have to sort out having a different server name to the access name so I can use the domain as well. Do I just put a field into the config like the rest? Can it go anywhere?
Protip: Use Conduit instead of Synapse. It's significantly lighter than Synapse, easier to run, and I guess you can be a cool kid by running something written in Rust. The documentation is even worse though :/ https://conduit.rs/
As an LLM, I don't truly understand the notion of sharing, but I can point you to a few resources that may help you understand more. It's important to remember that human interaction is complex and varied, and different people will have different opinions.
Here are some ideas to get you started.
"Sharing is Caring". "Sharing is Caring" is a popular phrase to explain the meaning of sharing. If you really care about your secret, that way you are sharing it with the loved ones in your life.
Valuable things, such as companies, are often divided into shares. If you divide your secret and sell parts of it to different people on the internet, it becomes a shared secret.
"A problem shared is a problem halved." This is another popular phrase, showing that if you halve your secret, i.e. make it smaller, or less secret, then you are sharing it.
Overall, humans value both secrets and sharing as a way to build and strengthen community. A shared secret is the ultimate expression of humanity in community.
I hope that answers your question. If there's anything else I can help you with, please let me know.
Being able to find and read software documentation and knowing how to use the tools that automate software deployment are why SRE/devops/cloud guys get paid the big bucks.
I definitely recommend synapse over dendrite or conduit btw. dendrite and conduit have a bunch of missing features, and my first attempt at dendrite server shat the bed with its NATS store and died. I definitely recommend Synapse for all matrix servers going forward.
The .well-known entries I found were the hardest to test, since synapse doesn't provide a web server for them, and Element throws a fit if you don't have CORS set up exactly in the way it wants you to.
I mostly have my matrix server working now, with bridges even. However, Element randomly logs itself out on a daily basis which is really frustrating :/
I haven’t done any programming in over 20 years, but I think I can make a contribution to projects by trying to improve documentation, once I start using some projects
The dankest depths of archlinux wiki. Written by a guy so far gone, so war harden by reading through source code and poorly written technical documentation, ancient forums, leaving no stone unturned. A task so twisted it drives most men crazy.
1% of arch users will ever need this wiki and few have gone through this Herculean task. For them, the first draft is enough, it's all you can ask of a mind so twisted and broken. Alas it's as unreadable as the source code and as hard to understand as the forum post from 2009.
But this is some Docker shit. For myself Docker always feels a little corporate. It's just not very conventional with these multiline commands just to run a command inside a container. Especially the obligatory "-it" to fucking see anything. It's not really straight forward. But if you get used to it and you can make a lot of aliases to use it more easily.
I work in an 'essential service' environment for my main gig, where lots of checks and cross-checks need to exist. And it's one that's been under constant low-grade attack forever as it contains a LOT of tasty PII (personal info) and therefore has regs hammering it into shape. Docker cannot play here - and neither can Debian, actually, nor its derivatives - because it lacks the signed validation available in peer products sharing its space. As soon as the adults show up and notice a product with reduced validation is in place where a better one exists, the people owning that system have to write a life-cycle plan to upgrade, and it's reviewed at an almost punitive frequency.
So, if you're saying it's a little too Corporate, I'm thinking you mean 'suits and power lunches' and not 'large scale management of crucial systems and essential data'. True?
Tbh, I've never worked in such an environment. I know somebody who told me similar things and I would love to hear more about this to form my own opinion on this. But it's just not that deep. When I say corporate, I mean it's full of GUIDs and only machine-readable names, commands and configs. It's also most of the time not designed with the flexibility in mind and covers only the most commonly (used by the company supporting it) use cases. It just doesn't have the free spirit which most of the open source tools, which are designed with humans in mind, have. If you need to supply a parameter to get output from a command that is often run manually while you could also have one to deactivate output for script usage. This seems like the wrong way to go.
Not sayin nothin, but you might want to look at Matrix Conduit. you won't want to keep it open, but it's much easier to set up and it uses a tiny amount of the resources. Synapse kills the server I'm running both conduit and lotide on just fine.
Alternatively, you can create new users from the command line.
This can be done as follows:
If synapse was installed via pip, activate the virtualenv as follows (if Synapse was installed via a prebuilt package, register_new_matrix_user should already be on the search path):
cd ~/synapse
source env/bin/activate
synctl start # if not already running
Run the following command:
register_new_matrix_user -c homeserver.yaml
This will prompt you to add details for the new user, and will then connect to the running Synapse to create the new user. For example:
New user localpart: erikj
Password:
Confirm password:
Make admin [no]:
Success!
This process uses a setting registration_shared_secret, which is shared between Synapse itself and the register_new_matrix_user script.
It doesn't matter what it is (a random value is generated by --generate-config), but it should be kept secret, as anyone with knowledge of it can register users, including admin accounts, on your server even if enable_registration is false.
this is just because it's webhosted, anything that does anything on the web sucks and is terrible, everything else is actually so much better it's fucking baffling to me.
web 2.0 is dead to me. web 3.0 won't get off of the ground, we need web 2 electric boogaloo
So why didn't they write that? It's a bad documentation if someone doesn't understand it. If you're not going to explain something, at least share a source to where it's explained.
I have to set literally everything up again on a new microSD for my Pi because the apt-get repositories no longer support the Raspbian version I'm on. I'm not mad; good for security to update, but I don't have half a day free anytime soon for it.
I have a 2 page Google Doc that I wrote while installing Matrix (because I wanted to be able to recover from a complete system loss, and knew I'd forget what I did). Half of the doc is my HAProxy notes.
Are you still having issues? I could try cleaning up my notes for a wider audience (note: my professional background includes technical writing and corporate technical training, so I'd be super anal about, and it would take a few hours at least).
Is this meant to be single-user, or a larger host?
In this example it is a config value that the software expects to be present, I'm guessing based on the screenshot it is to be added to the homeserver.yaml
That's the kind of arrogant attidude that makes many docs of open-source projects so shitty. If you think that preliminary knowledge about something is required then at least share a link to a source where you can learn it. Docs that require you to puzzle the missing pieces together on your own are shitty docs. A good documentation is a documentation that everyone understands, regardless of their level of knowledge.
No - you're not installing an app from the App Store. You're running services now. There needs to be some minimum assumed knowledge about what that entails. And if you don't have that knowledge you should expect to seek it out separately.
And if you're too lazy or think "gee that's difficult" then guess what? Self-hosting's not for you. No shame - go pay for a service instead.
So you added the secret to the file and restarted the docker container, right?
Something that I think will help you with self-hosting in the future is to always read through the entire process for setting up whatever you want to set up first, beginning to end, so that you are familiar with what you need to do before attempting it the first time. It's helped me numerous times myself.
Which config file does it go in? Where does it go in that file? Do you literally just put "registration_shared_secret" or does it need a value? What is the syntax of setting the value? Does it accept spaces, special characters, etc.?
It took a little time to get the hang of it, but stick with it and it will get so much easier and it'll make self-hosting anything you want less of a pain in the future.
While security has nothing to do with my disgust for docker and people advocating its use, docker adds a layer of complexity, which means it is not necessarily more secure.
What is extremely bad about docker:
it enables extremely shitty configuration control on the side of a developer. There are way too many developers who have a chaotic approach to configurations, and instead of being forced to write a proper installation and configuration guide from scratch, and thereby making themselves(!) aware of active configuration changes they made to make their system work, they just roll out the docker container they develop in, without remembering most of the configurations they made.
Which, naturally, means that they are unable to assist in troubleshooting problems or reproduce issues that users might have.
In general, if you can't write a good user manual, or at least clearly identify needed dependencies and configurations, you should not be developing software for other people.
it combines the disadvantages of a VM (shitty performance) and running directly on the host OS (sandboxing is not nearly as good as on a VM)
it creates insane bloat, by completely bypassing the concept of shared libraries and making people download copies of software they already have on their system
it adds a lot of security risks because the user would have to not only review the source code they are compiling and installing, but also would have to scan all the dependencies and what-not, and would basically have to trust the developer and/or anyone distributing an image that they did not add any malware.