Why should bottlepy stick to a file?(github.com)
github.com
Why should bottlepy stick to a file?
https://github.com/bottlepy/bottle/issues/1158
16 comments
The (admittedly biased) history of bottle and flask from the author was fascinating https://github.com/bottlepy/bottle/issues/1158#issuecomment-...
I think one of the most important technical statements in this comment is this: "even the same design errors (global request/response objects)".
(Note to outsiders: we're not talking globals in the usual sense here, they are very carefully crafted proxies on top of thread-locals that behave like globals, i.e. can be imported at module level, but behave correctly in terms of concurrency.)
As a long-time Flask developer, this is the biggest source of frustration when trying to do anything "frameworky" around Flask (see http://mitsuhiko.pocoo.org/AdvFlaskPatterns.pdf for some technical explanations).
There are a lot of emerging web frameworks in Python-land these last years, hopefully some of them will get the success they deserve.
To displace the current dominant players (arguably: Django, Flask and Pyramid), one needs a paradigm shift, just as Django displaced the dominant player (Zope) in 2005 when "MVC frameworks" became the dominant paradigm, or when Flask and Bottle rode the "microframeworks" wave in 2010.
This current paradigm shift is probably spelled ASGI. For more read https://dev.to/florimondmanca/introduction-to-asgi-emergence...
BTW: here's a fresh take from a friend who did exactly that last year: https://florimond.dev/blog/articles/2018/12/how-i-built-a-we... (with limited success, despite a lot of talent and dedication).
I'm curious to know which of the "new" (ASGI or not) Python frameworks are using these days.
(Note to outsiders: we're not talking globals in the usual sense here, they are very carefully crafted proxies on top of thread-locals that behave like globals, i.e. can be imported at module level, but behave correctly in terms of concurrency.)
As a long-time Flask developer, this is the biggest source of frustration when trying to do anything "frameworky" around Flask (see http://mitsuhiko.pocoo.org/AdvFlaskPatterns.pdf for some technical explanations).
There are a lot of emerging web frameworks in Python-land these last years, hopefully some of them will get the success they deserve.
To displace the current dominant players (arguably: Django, Flask and Pyramid), one needs a paradigm shift, just as Django displaced the dominant player (Zope) in 2005 when "MVC frameworks" became the dominant paradigm, or when Flask and Bottle rode the "microframeworks" wave in 2010.
This current paradigm shift is probably spelled ASGI. For more read https://dev.to/florimondmanca/introduction-to-asgi-emergence...
BTW: here's a fresh take from a friend who did exactly that last year: https://florimond.dev/blog/articles/2018/12/how-i-built-a-we... (with limited success, despite a lot of talent and dedication).
I'm curious to know which of the "new" (ASGI or not) Python frameworks are using these days.
It's a bit difficult to understand what's going on based only on the slides you've linked; do you have a link to a recording of the talk by any chance?
Starlette is very nice.
starlette.io/
starlette.io/
He says that Bottle lost and maybe it's true that flask is more popular, but I regard it as a superior framework because of its simplicity. I for one have completely replaced all my flask projects with bottle.
I don't quite get those arguments [1]:
> The single-file restriction is an effective save-guard against feature creep. [...] We are able reject pull requests simply because they would add too many lines or add too much complexity, and that's nice.
bottle.py is already 4425 lines of code [2].
So, despite they say they reject pull requests because they add too many lines, it's also can be said that they reject pull requests because file is too big already and merging PR makes it even harder to maintain.
> A single file is easier to debug and understand. [...] You can read the entirety of bottle.py in an hour or so. Most of it is pretty easy to understand. Try that with flask/pyramid/django [...]
Why not at least to just split it to 1 section per 1 file, which they emulate right now by using section-delimiting comments [3]? Will it make code much harder to understand? Or will it make code easier to understand?
Also, from a developer perspective, what about navigation to the correct section quickly. E.g. if developer wants to add a new Exception, how he can efficiently go to the specific section? If it wouldn't be single file, developer could jump to required file by typing couple of letters in fuzzy file opener it his editor of choice.
No one forces developers plunge into a feature creep apart from their own, and it actually doesn't matter is there a single file or not. Some prefer to write as many code as they can in one single line. I doubt that they do it to ease maintenance and to avoid feature creep.
[1]: https://github.com/bottlepy/bottle/issues/1158#issuecomment-... [2]: https://github.com/bottlepy/bottle/blob/master/bottle.py [3]: https://github.com/bottlepy/bottle/blob/332215b2b1b3de5a321b...
> The single-file restriction is an effective save-guard against feature creep. [...] We are able reject pull requests simply because they would add too many lines or add too much complexity, and that's nice.
bottle.py is already 4425 lines of code [2].
So, despite they say they reject pull requests because they add too many lines, it's also can be said that they reject pull requests because file is too big already and merging PR makes it even harder to maintain.
> A single file is easier to debug and understand. [...] You can read the entirety of bottle.py in an hour or so. Most of it is pretty easy to understand. Try that with flask/pyramid/django [...]
Why not at least to just split it to 1 section per 1 file, which they emulate right now by using section-delimiting comments [3]? Will it make code much harder to understand? Or will it make code easier to understand?
Also, from a developer perspective, what about navigation to the correct section quickly. E.g. if developer wants to add a new Exception, how he can efficiently go to the specific section? If it wouldn't be single file, developer could jump to required file by typing couple of letters in fuzzy file opener it his editor of choice.
No one forces developers plunge into a feature creep apart from their own, and it actually doesn't matter is there a single file or not. Some prefer to write as many code as they can in one single line. I doubt that they do it to ease maintenance and to avoid feature creep.
[1]: https://github.com/bottlepy/bottle/issues/1158#issuecomment-... [2]: https://github.com/bottlepy/bottle/blob/master/bottle.py [3]: https://github.com/bottlepy/bottle/blob/332215b2b1b3de5a321b...
> bottle.py is already 4425 lines of code
Have you noticed how bloated the other web frameworks are?
Bottle avoided feature creep very well and it even provides one of the fastest templating engines around.
Have you noticed how bloated the other web frameworks are?
Bottle avoided feature creep very well and it even provides one of the fastest templating engines around.
I recon it is similar to the desire for header-only C libs, package management isn’t great, so single files are easier to deal with. I wonder if anyone has made a Python package manager based on the idea of bundling up all the files in a Python project into a single .py file that can just be downloaded and copied around as-is? No requirements.txt, just vendored .py files in a libs dir.
Header only c libs also emerge from c and c++ realities that are less true elsewhere.
The biggest one is the optimizer does a better job when stuff is in the same compilation unit. (Inlining and stuff.) A good example to look at is sqlite, which is developed as multiple files but they concatenate them together for distribution.
Another reason is a C++ one, and C is not C++ but the community has enough overlap that I think it tends to spill over. And that is templates. Templates need to be in the header so a lot of c++ libraries have complicated, slow to compile headers.
I do think, for example when any given person on HN links to a header only C or C++ library, that a lot of these people are overdoing it. They have likely not measured it and found it to be better like the sqlite people. They are fighting, perhaps even misunderstanding the tooling, at the cost of compile times and large binary sizes. And for what? Makefiles are not really that hard to write.
The biggest one is the optimizer does a better job when stuff is in the same compilation unit. (Inlining and stuff.) A good example to look at is sqlite, which is developed as multiple files but they concatenate them together for distribution.
Another reason is a C++ one, and C is not C++ but the community has enough overlap that I think it tends to spill over. And that is templates. Templates need to be in the header so a lot of c++ libraries have complicated, slow to compile headers.
I do think, for example when any given person on HN links to a header only C or C++ library, that a lot of these people are overdoing it. They have likely not measured it and found it to be better like the sqlite people. They are fighting, perhaps even misunderstanding the tooling, at the cost of compile times and large binary sizes. And for what? Makefiles are not really that hard to write.
Title is confusing, maybe "keep to a single file"? Had to read a while before I realized it wasn't some kind of file-clinging code barnacle.
First I've heard about bottle, I'm intrigued - as a longtime Flask user who is occasionally frustrated with its OOP-zeal it might be up my alley.
First I've heard about bottle, I'm intrigued - as a longtime Flask user who is occasionally frustrated with its OOP-zeal it might be up my alley.
Back in college, Bottle was where I started to understand how web servers worked. I had written PHP but had zero idea how it worked (or how to write it) and eventually I tried Django, but couldn't configure it to send static files. Once I tried bottle though, it just clicked. It was so simple and I loved that there was little magic. It being a single file that I just dragged into a folder was a BIG part of the simplicity for me, too.
I mostly write nodejs these days (express sorta scratches the same itch), but I have fond memories of bottle. In fact, my old personal site (http://www.superftc.com/) is still a bottle app to this day.
Wow the comic is great I’ve just spent so long enjoying it. Thank you so much! How come you’ve stopped doing it?
Well thank you for the compliment, I'm glad you enjoyed it! I can't really recall why I stopped, but I think I just got tired of doing it and no one was reading it. It was just hard to make time for it I guess :/
I have a relatively popular single-Python-file, no-dependency-beyond-PSL project myself (slightly smaller than bottle.py, ~3.5k lines). It would certainly be shorter and easier to develop if I added a bunch of dependencies, but I’ve resisted the urge over time, and being able to just download a single file and use it with any Python installation that’s less than five years old is ultimately worth the effort IMO. Also got me closer to some lower level PSL modules.
Care to share what lib it is ?
Bottlepy is like one of those plucky little mammals living amongst giants. It is nimble and quick with small footprints.
I use it for quickly deploying endpoints for specific tasks such as a miniscule CDN and the obligatory blog. Great for home automation as well.
I use it for quickly deploying endpoints for specific tasks such as a miniscule CDN and the obligatory blog. Great for home automation as well.
I don't blame him. I made the same design choice for my python library[1]. I also don't use any non-native libraries in the single file. I'm sorry to say it's a much better experience for users if they don't have to deal with pip if they don't want to. I myself have been burned many times trying to install some python module for hours and failing because of some dependency issue.
[1] https://github.com/RPGillespie6/fastcov
[1] https://github.com/RPGillespie6/fastcov
Bottle was my 3rd python web framework that I tried and I really liked the idea.
But then I learned about Flask blueprints and now I can't live without them.
Imagine that I can make a lambda function that only loads selective blueprints, selective parts of my API, into AWS.
That's very powerful for off loading bottlenecks.
Is it possible to do blueprints with bottle? Because seeing as they've cemented the single file policy, it would make sense to have some sort of plugin system.
But then I learned about Flask blueprints and now I can't live without them.
Imagine that I can make a lambda function that only loads selective blueprints, selective parts of my API, into AWS.
That's very powerful for off loading bottlenecks.
Is it possible to do blueprints with bottle? Because seeing as they've cemented the single file policy, it would make sense to have some sort of plugin system.
Bottle doesn't have "blueprints" but I've used it for something like that in the past. IIRC you can "mount" another bottle app object into a URL path [1].
[1] https://bottlepy.org/docs/dev/api.html#bottle.Bottle.mount
[1] https://bottlepy.org/docs/dev/api.html#bottle.Bottle.mount
Hm:
> app – an instance of Bottle or a WSGI application.
So, that means a require to load the code, and the some kind of app = my app.default_app()?
Do you have an example of how you create an instance of a wsgi app? For bottle it looks like that's normally something bottle.run() does?
> app – an instance of Bottle or a WSGI application.
So, that means a require to load the code, and the some kind of app = my app.default_app()?
Do you have an example of how you create an instance of a wsgi app? For bottle it looks like that's normally something bottle.run() does?
wsgi is an interface; if you implement it then you are compatible. The simplest way is to just use another Bottle app, or maybe subclass the Bottle class like this `class MyClass (Bottle):`, otherwise you can create a new class by implementing the interface methods (many tutorials online for this). Then you "mount" that wsgi object into a specific path using the `.mount()` function of your main Bottle app.
Right, I get that part (sort of).
But looking at the quickstart, you typically don't subclass Bottle, you require it, and call run()? So your typical bottle app won't be a subclass of bottle? At which point you can't make an instance of it, to pass to mount()?
Looking at the documentation for mount() it looks like you should be able to mount the quick-start app - but it's not clear of you create an instance to pass to mount?
That is if I have example.py:
But looking at the quickstart, you typically don't subclass Bottle, you require it, and call run()? So your typical bottle app won't be a subclass of bottle? At which point you can't make an instance of it, to pass to mount()?
Looking at the documentation for mount() it looks like you should be able to mount the quick-start app - but it's not clear of you create an instance to pass to mount?
That is if I have example.py:
from bottle import route, run, template
@route('/hello/<name>')
def index(name):
return template('<b>Hello {{name}}</b>!', name=name)
run(host='localhost', port=8080)
test.py
=======
import bottle
@bottle.get('/')
def index():
return 'INDEX'
# a new wsgi app
hello = bottle.Bottle()
@hello.get('/<name>')
def greet(name):
return 'Hello, ' + name
bottle.app().mount('/hello', hello)
You should have 2 working routes, "/" and "/hello/<name>".
Thanks, that makes sense.
But I'm still not clear on how I require "example.py" and mount it (ie how do I make an wsgi instance from a python file that is a bottle app)?
But I'm still not clear on how I require "example.py" and mount it (ie how do I make an wsgi instance from a python file that is a bottle app)?
I love Bottle because it’s a single file and I can read through all of it to (re)understand it. It has all the essentials and nearly zero cruft, and I’ve been using it for ages - it runs my personal website, many API servers I wrote, and the odd minimal system daemon[0].
I rather like the single-file approach for small, focused software - imapbackup[1] and piku[2] were written to be easy to deploy and customize, and there certainly is a place for drop-in-and-forget libraries like bottle (I wish there were more like it, in other fields).
[0]: https://github.com/rcarmo/azure-k3s-cluster/blob/master/clou...
[1]: https://github.com/rcarmo/imapbackup
[2]: https://github.com/piku
I rather like the single-file approach for small, focused software - imapbackup[1] and piku[2] were written to be easy to deploy and customize, and there certainly is a place for drop-in-and-forget libraries like bottle (I wish there were more like it, in other fields).
[0]: https://github.com/rcarmo/azure-k3s-cluster/blob/master/clou...
[1]: https://github.com/rcarmo/imapbackup
[2]: https://github.com/piku
Kudos to Bottle for keeping their code simple and auditable. I've been writing small projects in Bottle for a while.
I never used Bottle and don't doubt the comments saying it's a great framework, but I'm more interested on the "single file" approach in general.
bottle.py may be a single file but it's 4,425 lines long. That doesn't strike me as particularly simple. Any project could potentially be "a single file" if it's a very long file.
Why would splitting it into a few separate files make it more complex? There's a reason why every modern programming language supports modules, and that's precisely to break down complexity into smaller, simpler units.
bottle.py may be a single file but it's 4,425 lines long. That doesn't strike me as particularly simple. Any project could potentially be "a single file" if it's a very long file.
Why would splitting it into a few separate files make it more complex? There's a reason why every modern programming language supports modules, and that's precisely to break down complexity into smaller, simpler units.
because it's extremely easy to "deploy" on old-fashioned cgi-hosts. You need a python-cgi configured, drop 2 files with ftp and your app is up and running. Also auditing is a lot easier (and you only have to trust a handful of maintainers).
Why does a single file mean you only have a trust a handful of maintainers? Each line of the ~4k lines could have been written by a different person and merged by a different maintainer.
because generally projects like these have some opinionated person who is maintaining it, which actively blocks "get the ssh-keys"-patches. compare that to the 100+ dependencies of your typical node-package.
Having many dependencies can be a problem. And the node ecosystem is a total joke.
But if the question was why not split it up into smaller files (which it was), then that's nothing to do with the number of dependencies or the number maintainers. My point is that auditing a program with no external dependences doesn't get any easier if the code is contained in one file or across ten.
But if the question was why not split it up into smaller files (which it was), then that's nothing to do with the number of dependencies or the number maintainers. My point is that auditing a program with no external dependences doesn't get any easier if the code is contained in one file or across ten.
Ok, ease of deployment in that specific context makes some sense. You could still have separate source files, and bundle them together for deployment, but that would add a build step so it's a trade off.
All the other reasons I heard, I don't really see what they have to do with having a single or multiple files.
All the other reasons I heard, I don't really see what they have to do with having a single or multiple files.
In a similar vein, the creator of nomadlist.com also runs a remote job site called remoteok.io. The site runs via a single PHP file called index.php (150kb) with 4.5k lines of code. No frameworks or libraries. Note: this tweet is from Dec 2017 so things may have changed.
https://twitter.com/levelsio/status/938707166508154880
https://twitter.com/levelsio/status/938707166508154880
Arguments like this are why I gave up on scripting languages. "go build" generates one statically-linked binary that you can send anywhere. You can build a Windows binary from your Raspberry Pi or vice-versa. It is how everything should work, because distributing (and cross-compiling) is something you should have to spend 0 time thinking about. Now I spend 0 time thinking about it.
The argument has pretty much nothing to do with dynamic languages.
Single file dependencies are getting more and more popular because integrating with build systems is a major pain, particularly in static languages like C and C++, but also in the mess that is modern Javascript.
Some languages do manage to do this better than others, but the type system has nothing to do with it.
Lastly, having one single statically-linked binary isn't the norm even for static languages.
Single file dependencies are getting more and more popular because integrating with build systems is a major pain, particularly in static languages like C and C++, but also in the mess that is modern Javascript.
Some languages do manage to do this better than others, but the type system has nothing to do with it.
Lastly, having one single statically-linked binary isn't the norm even for static languages.
I'm just saying that's why I stopped using them, not that it's a differentiating factor between static and dynamic languages. I wrote books about Perl, I taught classes on Perl. Having to install stuff from CPAN was probably the most common complaint. I wrote a new CPAN client at some point, and it didn't help. Enough was enough; the approach failed and I moved on to writing my code some other way (as did all of those students, probably). npm seems to have gotten something that mostly works; I use that and tolerate it because I have no other choice for client-side stuff, but I don't really do enough browser programming to call myself a Javascript user.
Yes, static languages have dependency hell. I've ./configure'd a million times and have wanted to die about 80% of the time. I've used buildroot and have wanted to die 100% of the time. I never seriously programmed in C or C++ to say that I've stopped using them, so while they're broken... you can't stop using something that you haven't started using. (I still cringe when I see that some C-based project I want to use doesn't have binaries. It's especially annoying because I mostly use Windows now. I want "dc" in Powershell because that is my preferred calculator app... but can't compile it, and apparently neither can anyone else. Wow.)
Python is a pretty great programming language... but it's just not worth the tradeoffs. It is nice that bottle is one file; that is going to increase its adoption. It is not a solution to any general problem, though; the next library you depend on is going to require virtualenv and 100 dependencies, and they will fail to install.
Yes, static languages have dependency hell. I've ./configure'd a million times and have wanted to die about 80% of the time. I've used buildroot and have wanted to die 100% of the time. I never seriously programmed in C or C++ to say that I've stopped using them, so while they're broken... you can't stop using something that you haven't started using. (I still cringe when I see that some C-based project I want to use doesn't have binaries. It's especially annoying because I mostly use Windows now. I want "dc" in Powershell because that is my preferred calculator app... but can't compile it, and apparently neither can anyone else. Wow.)
Python is a pretty great programming language... but it's just not worth the tradeoffs. It is nice that bottle is one file; that is going to increase its adoption. It is not a solution to any general problem, though; the next library you depend on is going to require virtualenv and 100 dependencies, and they will fail to install.
I guess we're on the same page then and I wouldn't have written the reply had you been more specific.
Having said that, I do find the situation with Python/PIP/PyPI today is far better than a few years ago, especially concerning binary packages and the Python2/Python3 divide.
I'm not experiencing a substantial quality difference between pip and npm, at the same time I prefer venv over node_modules (while detesting both).
Having said that, I do find the situation with Python/PIP/PyPI today is far better than a few years ago, especially concerning binary packages and the Python2/Python3 divide.
I'm not experiencing a substantial quality difference between pip and npm, at the same time I prefer venv over node_modules (while detesting both).
Huge fan of bottle. I use it for a lot of things.
The singlefile-ness of Bottle is no longer unique.
For example, the modern day Fastapi built on async ASGI (and ranks top in framework benchmarks) is also single file.
https://github.com/tiangolo/fastapi
or Fastapi for that matter - https://github.com/encode/starlette
However, it is a bit tricky to run Bottle with production grade servers like gunicorn/uwsgi.
https://github.com/tiangolo/fastapi
or Fastapi for that matter - https://github.com/encode/starlette
However, it is a bit tricky to run Bottle with production grade servers like gunicorn/uwsgi.
What? FastAPI is certainly not single file: https://github.com/tiangolo/fastapi/tree/master/fastapi
The issue is talking about Bottle the framework itself being a single file, bottle.py.
The issue is talking about Bottle the framework itself being a single file, bottle.py.
oh my bad. That's interesting, but im not sure about utility.
Because everyone literally uses package management to install and maintain bottle.
Bottle has code like this - https://github.com/bottlepy/bottle/blob/master/bottle.py#L81... - which does check for external dependencies (which predicate a package manager)
Because everyone literally uses package management to install and maintain bottle.
Bottle has code like this - https://github.com/bottlepy/bottle/blob/master/bottle.py#L81... - which does check for external dependencies (which predicate a package manager)
FastAPI and Starlette both have tons of files, and even subpackages. Bottle ships as only a single bottle.py file, with no external dependencies (only Python standard library). FastAPI depends on Starlette, and both have a set of optional external dependencies.
> However, it is a bit tricky to run Bottle with production grade servers like gunicorn/uwsgi.
Uh, no? Bottle has docs for deployment: https://bottlepy.org/docs/dev/deployment.html#switching-the-...
They have automated support for gunicorn. uWSGI requires adding one line (`app = bottle.default_app()`) and pointing uWSGI at that.
> However, it is a bit tricky to run Bottle with production grade servers like gunicorn/uwsgi.
Uh, no? Bottle has docs for deployment: https://bottlepy.org/docs/dev/deployment.html#switching-the-...
They have automated support for gunicorn. uWSGI requires adding one line (`app = bottle.default_app()`) and pointing uWSGI at that.
It’s indeed strange to call any WSGI-capable framework (all of them?) tricky to deploy with gunicorn/uwsgi. All you need to do is expose the WSGI object.
In Flask's case, it's indirect, as Armin admitted in the 2011 presentation (http://mitsuhiko.pocoo.org/flask-pycon-2011.pdf): "Wordplay on Bottle, probably a mistake".