4 ms·

> It's also more wasteful as a transport. This is most certainly true, however with GZip thrown into the mix, it's not quite as bad as one might imagine: https

> It's also more wasteful as a transport. This is most certainly true, however with GZip thrown into the mix, it's not quite as bad as one might imagine: https://www.codeproject.com/articles/604720/json-vs-xml-some-hard-numbers-about-verbosity https://www.codeproject.com/articles/604720/json-vs-xml-some... It compresses pretty decently and doesn't have too much of an overhead, in the example it being around 10% larger than JSON when compressed. I'd argue that if one were to swap out JSON for XML within all the requests that an average webpage needs for some unholy reason, the overall increase in page size would be much less than that, because huge amounts of modern sites are images, as well as bits of JS that won't be executed but also won't be removed because our tree shaking isn't perfect. Edit: as someone who writes a good deal of Java in their dayjob, i feel like commenting about the verbosity of XML might be unwelcome. I'll only say that in some cases it can be useful to have elements that have been structured and described in verbose ways, especially when you don't have the slightest idea about what API or data you're looking at when seeing it for the first time (the same way how WSDL files for SOAP could provide discoverability). However, it all goes downhill due to everything looking like a nail once you have a hammer - most of the negative connotations with XML in my mind actually come from Java EE et al and how it tried doing dynamic code loading through XML configuration (e.g. web.xml, context.xml, server.xml and bean configuration), which was unpleasant. On an unrelated note, XSD is the one truly redeeming factor of XML, the equivalent of which for JSON took a while to get there (JSON Schema). Similarly, WSDL was a good attempt, whereas for JSON there first was WADL which didn't gain popularity, though at least now OpenAPI seems to have a pretty stable place, even if the tooling will still take a while to get there (e.g. automatically generating method stubs for a web API with a language's HTTP client).

BerislavLopac 5y ago
You mean something like https://pyotr.readthedocs.io https://pyotr.readthedocs.io
- KronisLV 5y ago
  Thanks for the link, but not necessarily. How WSDL and the code generation around it worked, was that you'd have a specification of the web API (much like OpenAPI attempts to do), which you could feed into any number of code generators, to get output code which has no coupling to the actual generator at runtime, whereas Pyotr is geared more towards validation and goes into the opposite direction: https://pyotr.readthedocs.io/en/latest/client/ https://pyotr.readthedocs.io/en/latest/client/ The best analogy that i can think of is how you can also do schema first application development - you do your SQL migrations (ideally in an automated way as well) and then just run a command locally to generate all of the data access classes and/or models for your database tables within your application. That way, you save your time for 80% of the boring and repetitive stuff while minimizing the risks of human error and inconsistencies, with nothing preventing you from altering the generated code if you have specific needs (outside of needing to make it non overrideable, for example, a child class of a generated class). Of course, there's no reason why this can't be applied to server code either - write the spec first and generate stubs for endpoints that you'll just fill out. Similarly there shouldn't be a need for a special client to generate stubs for OpenAPI, the closest that Python in particular has for now is this https://github.com/openapi-generators/openapi-python-client https://github.com/openapi-generators/openapi-python-client However, for some reason, model driven development never really took off, outside of niche frameworks, like JHipster: https://www.jhipster.tech/ https://www.jhipster.tech/ Furthermore, for whatever reason formal specs for REST APIs also never really got popular and aren't regarded as the standard, which to me seems silly: every bit of client code that you write will need a specific version to work against, which should be formalized.
  - BerislavLopac 5y ago
    Well, in Python, code generation is an anti-pattern.
    - KronisLV 5y ago
      > Well, in Python, code generation is an anti-pattern. Hmm, i don't think that i've ever heard of this. Would you care to provide any sources, since that sounds like an interesting stance to take? So far, it seems like frameworks like Django don't have an issue with CLI tools to generate bits of code, i.e. https://docs.djangoproject.com/en/3.2/intro/tutorial01/ https://docs.djangoproject.com/en/3.2/intro/tutorial01/ If this is your first time using Django, you’ll have to take care of some initial setup. Namely, you’ll need to auto-generate some code that establishes a Django project – a collection of settings for an instance of Django, including database configuration, Django-specific options and application-specific settings. $ django-admin startproject mysite Similarly, PyCharm doesn't seem to have an issue with offering to generate methods for classes (ALT + INSERT), such as override methods (__class__, __init__, __new__, __setattr__, __eq__, __ne__, __str__, __repr__, __hash__, __format__, __getattribute__, __delattr__, __sizeof__, __reduce__, __reduce_ex__, __dir__, __init__), implementing methods, generating tests and copyright information. I don't see why CLI tools would be treated any differently or why code generation should be considered an anti-pattern since it's additive in nature and is entirely optional, hence asking to learn more.
      - BerislavLopac 5y ago
        First of all, just because a tool or project uses a pattern, it doesn't mean that it's a good idea. Second, code generation as part of IDE or one-time setup is something else. I need to clarify: when I say that "code generation" is an anti-pattern, I'm talking about the traditional, two-step process where you generate some code in one process, and then execute it in another. But Python works really well with a different type of "code generation". Someone once said that the only thing missing from Python is a macro language; but that is not true - Python has its own macro language, and it's called Python. Python is dynamically evaluated and executed, so there is no reason why we need two separate steps when generating code dynamically; in Python, the right way is not to dynamically construct the textual representation of code, but rather to dynamically construct runtime entities (classes, functions etc), and then use them straight away, in the same process. Unless you're dynamically building hundreds of such constructs (and if you do you have a bigger problem), any performance impact is negligible.

BerislavLopac 5y ago

You mean something like https://pyotr.readthedocs.io https://pyotr.readthedocs.io

KronisLV 5y ago
Thanks for the link, but not necessarily. How WSDL and the code generation around it worked, was that you'd have a specification of the web API (much like OpenAPI attempts to do), which you could feed into any number of code generators, to get output code which has no coupling to the actual generator at runtime, whereas Pyotr is geared more towards validation and goes into the opposite direction: https://pyotr.readthedocs.io/en/latest/client/ https://pyotr.readthedocs.io/en/latest/client/ The best analogy that i can think of is how you can also do schema first application development - you do your SQL migrations (ideally in an automated way as well) and then just run a command locally to generate all of the data access classes and/or models for your database tables within your application. That way, you save your time for 80% of the boring and repetitive stuff while minimizing the risks of human error and inconsistencies, with nothing preventing you from altering the generated code if you have specific needs (outside of needing to make it non overrideable, for example, a child class of a generated class). Of course, there's no reason why this can't be applied to server code either - write the spec first and generate stubs for endpoints that you'll just fill out. Similarly there shouldn't be a need for a special client to generate stubs for OpenAPI, the closest that Python in particular has for now is this https://github.com/openapi-generators/openapi-python-client https://github.com/openapi-generators/openapi-python-client However, for some reason, model driven development never really took off, outside of niche frameworks, like JHipster: https://www.jhipster.tech/ https://www.jhipster.tech/ Furthermore, for whatever reason formal specs for REST APIs also never really got popular and aren't regarded as the standard, which to me seems silly: every bit of client code that you write will need a specific version to work against, which should be formalized.
- BerislavLopac 5y ago
  Well, in Python, code generation is an anti-pattern.
  - KronisLV 5y ago
    > Well, in Python, code generation is an anti-pattern. Hmm, i don't think that i've ever heard of this. Would you care to provide any sources, since that sounds like an interesting stance to take? So far, it seems like frameworks like Django don't have an issue with CLI tools to generate bits of code, i.e. https://docs.djangoproject.com/en/3.2/intro/tutorial01/ https://docs.djangoproject.com/en/3.2/intro/tutorial01/ If this is your first time using Django, you’ll have to take care of some initial setup. Namely, you’ll need to auto-generate some code that establishes a Django project – a collection of settings for an instance of Django, including database configuration, Django-specific options and application-specific settings. $ django-admin startproject mysite Similarly, PyCharm doesn't seem to have an issue with offering to generate methods for classes (ALT + INSERT), such as override methods (__class__, __init__, __new__, __setattr__, __eq__, __ne__, __str__, __repr__, __hash__, __format__, __getattribute__, __delattr__, __sizeof__, __reduce__, __reduce_ex__, __dir__, __init__), implementing methods, generating tests and copyright information. I don't see why CLI tools would be treated any differently or why code generation should be considered an anti-pattern since it's additive in nature and is entirely optional, hence asking to learn more.
    - BerislavLopac 5y ago
      First of all, just because a tool or project uses a pattern, it doesn't mean that it's a good idea. Second, code generation as part of IDE or one-time setup is something else. I need to clarify: when I say that "code generation" is an anti-pattern, I'm talking about the traditional, two-step process where you generate some code in one process, and then execute it in another. But Python works really well with a different type of "code generation". Someone once said that the only thing missing from Python is a macro language; but that is not true - Python has its own macro language, and it's called Python. Python is dynamically evaluated and executed, so there is no reason why we need two separate steps when generating code dynamically; in Python, the right way is not to dynamically construct the textual representation of code, but rather to dynamically construct runtime entities (classes, functions etc), and then use them straight away, in the same process. Unless you're dynamically building hundreds of such constructs (and if you do you have a bigger problem), any performance impact is negligible.