"zero configuration" - but not respecting most package.json options?

[mindplay-dk]

I may be completely ignorant of something, so forgive me, but isn't "zero configuration" meant to imply that your package.json gets treated as the "source of truth"?

Meaning, the intent is for package.json to govern what gets generated, filenames, and so forth - directly ensuring that the default settings match what your package.json specifies?

Because it seems like microbundle more or less completely ignores my package.json, and I have to manually configure everything anyway?

For example, I tried this first:

{
  "name": "lol",
  "module": "./dist/lol.js",
  "scripts": {
    "build": "microbundle"
  },
  "files": [
    "dist"
  ]
}

Which resulted in builds for every format supported - I had to add the --format modern switch to disable the other formats.

By adding module and specifying a path, haven't I explicitly said "this is the only kind of format I need"?

It also does not seem to respect the path. The output filename is lol.modern.js, despite my specifying module as ./dist/lol.js, and you end up with a package that can't be installed at all.

I decided to try exports with default instead:

{
  "name": "lol",
  "module": "./dist/lol.js",
  "scripts": {
    "build": "microbundle --format modern"
  },
  "files": [
    "dist"
  ]
}

This similarly gets ignored - it doesn't use the filename, and it doesn't govern what gets generated; the filename, apparently, is always derived from name and the paths you specify are ignored.

(and yes, I noticed the --name switch, where I could explicitly specify this again - but having to configure the same thing first for npm and then for microbundle, doesn't that kind of clash with the idea of "zero configuration"?)

Likewise, this being a Typescript project, I tried adding types:

{
  "name": "lol",
  "module": "./dist/lol.js",
  "types": "./dist/lol.d.ts",
  "scripts": {
    "build": "microbundle --format modern"
  },
  "files": [
    "dist"
  ]
}

This gets even more confusing, as the presence of types does have an effect - it does trigger the creation of .d.ts files. However, the value appears to be ignored - the .d.ts files just get filenames derived from the input filenames. (It also does not emit a single .d.ts file, as I had hoped - but that's probably an unrelated issue or a non-feature.)

Things got more curious still, when I attempted multiple entry-points:

{
  "name": "lol",
  "exports": {
    ".": "./dist/lol.js",
    "./wat": "./dist/wat.js"
  },
  "scripts": {
    "build": "microbundle --format modern"
  },
  "files": [
    "dist"
  ]
}

As I expected at this point, this had no effect, and reporting.js just wasn't generated - however, when I added src/index.ts src/reporting.ts to the microbundle command, not only did it generate both files, but in addition, suddenly the filenames in exports were being applied:

{
  "name": "lol",
  "exports": {
    ".": "./dist/lol.js",
    "./wat": "./dist/wat.js"
  },
  "scripts": {
    "build": "microbundle --format modern src/index.ts src/wat.ts"
  },
  "files": [
    "dist"
  ]
}

(Side note: the contents of the output dist/wat.js file somehow magically is what I expected - it's not clear to me how microbundle gets from src/index.ts src/wat.ts back to either the ./wat name or the ./dist/wat.js path in the exports section and figures this out? Other than the word wat, there's no direct relationship between either the name or the path and the input file - I wasn't expecting this to work, I was just sort of trying things and got lucky...)

Thinking I had figured out the special sauce (explicitly specifying the entry file) I reverted back to this:

{
  "name": "lol",
  "exports": {
    "default": "./dist/lol.js"
  },
  "scripts": {
    "build": "microbundle --format modern src/index.ts"
  },
  "files": [
    "dist"
  ]
}

Solved, right? 😅

Wrong.

Apparently, specifying the filenames in exports only works when you specify more than one entry point.

I wanted to share this whole user journey to give you some idea of what the developer experience might be like for some people. I'm sure there's some way to find a "winning configuration", but after sinking another half-day into this, I'm just going to go back to rollup... again.

It's not the first time I've attempted to use microbundle - and please don't get me wrong, I don't mean any disrespect, and I see this working for a bunch of projects.

But as compilers/bundlers go, this is not the most intuitive or pleasant experience. The idea of "zero configuration" appeals to me, but I can't honestly say that this delivers.

I guess maybe it depends on what you expect or mean by "zero configuration"? Personally, I take that to mean "zero extra configuration" - as it's obviously not possible to do anything without some sort of configuration, somewhere. I thought the idea was you derive a sensible default configuration from what's there anyway, your package.json file, and perhaps with the option to override?

Maybe that's not it at all, but I got the impression from other tools that that's the idea?

Are my expectations way off track here?? 😄

[rschristian]

I'd argue that requiring an specific package.json setup is configuration, while Microbundle outputs a set of generally acceptable defaults with none. Outputting extra files poses no real issue, hence, "no configuration".

By adding module and specifying a path, haven't I explicitly said "this is the only kind of format I need"?

That's configuration to me, if that's what you expect. But people have different expectations, nothing wrong with that.

While I do agree with that sentiment, I could also see how that could be frustrating if you needed to configure both your Microbundle build script AND your package.json in order to output certain formats at all. It'd likely be a very common and confusing issue.

{
 "name": "lol",
 "module": "./dist/lol.js",
 "scripts": {
   "build": "microbundle"
 },
 "files": [
   "dist"
 ]
}

It also does not seem to respect the path. The output filename is lol.modern.js, despite my specifying module as ./dist/lol.js, and you end up with a package that can't be installed at all.

You might not have read the docs thoroughly enough. Modern mode is mapped to "exports", not "module". As modern is ES2017, it isn't necessarily safe to supply to "module", as that field has been in use for a lot longer and some tools may not support consuming modern output from it. Gating being "exports" is safe, however.

The esm format is what maps to "module", as the docs show.

I decided to try exports with default instead:

{
 "name": "lol",
 "module": "./dist/lol.js",
 "scripts": {
   "build": "microbundle --format modern"
 },
 "files": [
   "dist"
 ]
}

Did you copy/paste the wrong snippet here? No "exports" in that block.

This similarly gets ignored - it doesn't use the filename, and it doesn't govern what gets generated; the filename, apparently, is always derived from name and the paths you specify are ignored.

No, it's not:

microbundle/src/index.js

Lines 271 to 277 in fb0a437

	function walk(exports, includeDefault) {
	if (!exports) return null;
	if (typeof exports === 'string') return exports;
	let p = exports['.'] || exports.import || exports.module;
	if (!p && includeDefault) p = exports.default;
	return walk(p, includeDefault);
	}

I'm guessing you didn't include "type": "module", which is pretty important. The "default" key of "exports" cannot be ESM if you've not the set the package's type to module. That'd be invalid in Node. Therefore your config should be ignored, as it would only lead to errors if you actually tried to use it.

Our docs have a config for you to copy/paste. If you skip over that and aren't super familiar with Node's rules, you could run into issues. I'm not sure how we can address that better, certainly open to suggestions.

That being said, I do currently have a few issues with Microbundle's current name resolution setup (see #896) but that's not related to this. There are some rough spots, no doubt, but this seems to just be an issue of missing important steps in the docs.

However, the value appears to be ignored - the .d.ts files just get filenames derived from the input filenames

Hm, I believe it should output a main/entry .d.ts matching the name you provide. Will look into this, thanks.

(It also does not emit a single .d.ts file, as I had hoped - but that's probably an unrelated issue or a non-feature.)

The plethora of other .d.ts is due to TSC, I believe. We use rollup-plugin-typescript2 for this.

I believe I remember Jason putting together a setup to bundle them all together for some project, will have to look into that again. I think we all agree it's not ideal, though, for the moment anyways, it shouldn't cause any functional issues.

Things got more curious still, when I attempted multiple entry-points:

Yep, it's on the roadmap to (one day, hopefully) complete: https://github.com/developit/microbundle#-roadmap

I wanted to share this whole user journey to give you some idea of what the developer experience might be like for some people.

We really appreciate you writing this up! I think there's a couple issues you ran into that need to be fixed on your end in your config, but there's certainly a wealth of feedback here for things we can look into doing better in the future.

Sorry you ran into difficulties with this.

Personally, I take that to mean "zero extra configuration"

I suppose that just depends on what you call "extra". Microbundle outputs a set of sensible defaults, but I guess we differ from whether you'd call a package.json setup "extra configuration". I would, but maybe others wouldn't.

The package.json is in my mind the override where you can control output names and the like if you wish.

[rschristian]

Looked into the "types" issue you mentioned and unfortunately that's correct TS behavior, which is used upstream by rollup-plugin-typescript2. TS exposes no way to rename entries or combine them into a single output, so we're stuck with that limitation too unless we were to migrate away to something not TS-based (which creates a whole ton of other issues).

We cannot even rename the corresponding entry .d.ts file as they're often written directly to the filesystem, bypassing Rollup, due to the declarationDir option in a tsconfig.json not necessarily being under Rollup's output directory. See: https://github.com/ezolenko/rollup-plugin-typescript2/blob/ff8895103c8466694e7d8eeb734f51d2850670d8/src/index.ts#L358-L362

You can use npm-dts to merge declaration files (I haven't used it myself, but seems to be the go-to) if you need that.

[mindplay-dk]

TS exposes no way to rename entries or combine them into a single output, so we're stuck with that limitation too unless we were to migrate away to something not TS-based (which creates a whole ton of other issues).

Hmm, well, there is this package, which appears to do it via the Typescript API?

https://www.npmjs.com/package/rollup-plugin-dts

We cannot even rename the corresponding entry .d.ts file as they're often written directly to the filesystem, bypassing Rollup, due to the declarationDir option in a tsconfig.json not necessarily being under Rollup's output directory.

Right. Square peg in a round hole. It's the story of Javascript isn't it. 🙄

[rschristian]

Interesting, hadn't heard of that one.

Hmm, well, there is this package, which appears to do it via the Typescript API?

That project is using TS's API to read and write files, which is a bit different from using TS to compile (and what I was referring to). Not entirely sure how using TS differs from just using Node to read and write honestly, but regardless, that plugin is a post-TS compilation step to combine .d.ts outputs together.

Theoretically we might be able to add it on top, will need to look into it. Maybe due to using the TS API it can pick up on the TS transpilation plugin writing directly to the filesystem.

Right. Square peg in a round hole. It's the story of Javascript isn't it. 🙄

Yeah. Rollup and .d.ts are a bit of an awkward combination, unfortunately.

[mindplay-dk]

Honestly though, the biggest issue with microbundle for me right now is the configuration.

I mean, yeah, if you copy/paste the example from the README, you happen to get something that works - but as soon as you change anything (even the package name) everything goes wrong.

Returning to this:

Personally, I take that to mean "zero extra configuration"

I suppose that just depends on what you call "extra". Microbundle outputs a set of sensible defaults, but I guess we differ from whether you'd call a package.json setup "extra configuration". I would, but maybe others wouldn't.

Having thought about this, I don't agree.

The configuration in package.json isn't optional - it needs to be there no matter what.

To reverse the question: when would you want configuration that doesn't match your package.json?

Is there any useful case for deliberately configuring microbundle to produce something that can't be reached or isn't declared or defined in the package metadata?

package.json is native to the platform, so it's there no matter what you're trying to do, right?

Having to configure twice (first in package.json and then in microbundle options) doesn't make sense to me - and having to suppress microbundle defaults that don't respect what you've declared in the native package metadata, doesn't make sense to me either.

Why wouldn't you use the information that is inevitably there anyhow? If it's not "zero configuration", then at least it's "sensible defaults", derived from things you need to specify anyhow. Why not? 🙂

[rschristian]

I mean, yeah, if you copy/paste the example from the README, you happen to get something that works

No, if you copy/paste the example in the README you get the product of many hours of our time spent in the rabbit hole of creating output that is usable in every major bundler and Node. It's not pure chance, it's the result of a bit of rather esoteric knowledge on the subject that most people don't have or need. That's totally fine, and is why we provide an example over expecting everyone to be able to make their own.

I really don't understand this criticism, especially after running into the above error. It should be clear our examples don't just "happen" to work.

but as soon as you change anything (even the package name) everything goes wrong.

As soon as you change names (or extensions) you're expected to understand Node's package resolution mechanism, yes.

We're certainly open to suggestions on how to provide a better experience, but if you ignore our recommended setup, and write an invalid one yourself? It's a bit out of our hands.

To reverse the question: when would you want configuration that doesn't match your package.json?

Let's say we did match your package.json: upon attempting to use your module, you would've been met by Node immediately exiting and telling you that you cannot have import/export in a CJS module. Which would you prefer, that issue, or the output being incorrectly named and not able to be resolved?

That's not to say we can't provide warnings, but a potential issue is that package exports are a fairly complex system, and we cannot necessarily assume Microbundle is in control of the whole process. Preact, for example. actually has a post-build script to make duplicate copies of a bundle but change the file extension.

Ejecting from Microbundle is a fairly common need for a whole plethora of reasons. Providing a bunch of warnings might not be ideal as they could decrease the signal-to-noise ratio.

This is something I'll look into more heavily in the next few days/weeks, however. We certainly want to improve Microbundle where we can.

Is there any useful case for deliberately configuring microbundle to produce something that can't be reached or isn't declared or defined in the package metadata?

You're making the assumption that every consumer of Microbundle's output does so via packages where a package.json is relevant and used. This is not the case. You can use Microbundle to build widgets that are requested straight from a CDN. The package meta data is utterly irrelevant in such cases as the resource is served from a GET request alone, assuming the CDN isn't doing processing of their own.

package.json is native to the platform, so it's there no matter what you're trying to do, right?

No, it's native to Node. There are many ways to consume JS outside of Node.

then at least it's "sensible defaults", derived from things you need to specify anyhow. Why not? 🙂

It, for the most part, already is. You ran into issues providing invalid specifications and skipping the docs (ignoring the TS thing, as unfortunately we just can't control that bit, though I'd like to).

[mindplay-dk]

I mean, yeah, if you copy/paste the example from the README, you happen to get something that works

I really don't understand this criticism, especially after running into the above error. It should be clear our examples don't just "happen" to work.

Sorry, Ryan - I think I'm not explaining myself well. 😕

All I meant to point out with this, is that there's no link between the information you put in package.json and the configuration defaults and resulting output of the tool - if you happen to get both of these configurations exactly right, and/or assuming your required configuration exactly matches the defaults of the tool, then yes, you get something that works; if you deviate in your package.json from the tool's defaults, you have to make sure what you configured for npm in package.json exactly matches your configuration for the tool: you have to configure everything twice, once for npm, and again for the tool.

I think that's a problem. That's been my only point from the start.

Is there any useful case for deliberately configuring microbundle to produce something that can't be reached or isn't declared or defined in the package metadata?

You're making the assumption that every consumer of Microbundle's output does so via packages where a package.json is relevant and used.

Okay, so some people publish packages that aren't even valid npm packages. I didn't know that.

But I'm not trying to suggest you shouldn't be able to override the defaults - if somebody wants to use npm to publish things that aren't even installable packages, then so be it.

All I'm suggesting is maybe those aren't safe, helpful, meaningful defaults - I still don't understand why you wouldn't derive meaningful defaults from package.json, as most likely somebody wants to publish a valid npm package manifest that actually matches what's in the package.

If I'm wrong about that, there's something very fundamental I'm not grasping here.

then at least it's "sensible defaults", derived from things you need to specify anyhow. Why not? 🙂

It, for the most part, already is. You ran into issues providing invalid specifications and skipping the docs

I didn't skip the docs, I just read it with a different set of assumptions.

I understand now what's actually in package.json in the docs is boilerplate - I assumed it was just an example. Looking at it again, I think what mislead me here was the comments saying "where to generate", which I took to mean that I'd be able to specify where to generate the outputs - what this actually is, is "where it will be generated" regardless of what you actually put there.

I'm sorry, and maybe I'm an idiot, but I'm still not grasping why you couldn't use package.json to establish a meaningful default configuration. Parcel does, and I think that may have been where I first heard the term "zero configuration". As far as my understanding, that was what Parcel originally meant by "zero configuration" - of course there's a configuration for npm, or you wouldn't have a working package, but there's "zero configuration" for the bundler.

You're not configuring everything twice, unless you want to diverge from the defaults.

Am I making any sense here? If I'm completely wasting your time here, I really do apologize. Feel free to ignore me. I'll admit, the whole node/JS/TS ecosystem is baffling, confusing and completely overwhelming to me. No matter how many articles, tools, courses and tutorials I plow through, I only seem to get more confused and overwhelmed. (If you can point me to anything that's going to help me make sense of all this craziness, please share. I'm more than willing to learn. I too have spent "many hours in the rabbit hole", but the tunnels seem to go in circles and the exits never seem to lead where I wanted to go. 😅)

[rschristian]

All I meant to point out with this, is that there's no link between the information you put in package.json and the configuration defaults and resulting output of the tool

There is a link, but it's probably just not as "hard" of a link as you might like.

The default is just setup so that keys your provide rename the default formats when applicable. I do agree with you that a better default would be in fact to only output formats where matching keys are found, but that was set way before I knew the tool existed and I think it's unlikely that the behavior can really change now as it's matured.

Thinking about just the first-party usage alone, there'd probably be a fair bit of breakage/issues upon changing the defaults, much less for the wider community.

I'll certainly keep that in mind, maybe we can change that, but I think for now it's a bit unlikely, unfortunately.

if you happen to get both of these configurations exactly right, and/or assuming your required configuration exactly matches the defaults of the tool, then yes, you get something that works;

I'm a bit confused by this. Getting both configurations exactly right (matching output format to package.json keys) is an issue that's intrinsic to the ecosystem. This will be an issue with or without Microbundle. Using tsc would present this issue, for example.

if you deviate in your package.json from the tool's defaults, you have to make sure what you configured for npm in package.json exactly matches your configuration for the tool: you have to configure everything twice, once for npm, and again for the tool.

Microbundle doesn't set any expectations about what your package.json contains as a default. Not sure what you mean?

This is an issue but the other way around, in that if you configure Microbundle, you need to make sure your package.json matches its output (else you'll have keys referring to output formats that don't exist, for example).

But if you deviate the keys in your package.json (actually, if the keys exist at all) then we'll use them when valid. You wouldn't need to ensure your configuration for Microbundle matches.

All I'm suggesting is maybe those aren't safe, helpful, meaningful defaults - I still don't understand why you wouldn't derive meaningful defaults from package.json, as most likely somebody wants to publish a valid npm package manifest that actually matches what's in the package.

Well, the thing is that your criticism technically is a safe default. If you specify a "module" field and that alone, with no configuration of Microbundle's outputs, you'll get 3 outputs that are never used and 1 that matches up to your specified field. You can publish that and have no issues (beyond the issues of only publishing a package with "module", definitely don't do that). Extra files aren't actually a problem.

Again, I do agree, this isn't ideal, and I'd never (knowingly) publish a package with a bunch of unused files, but.... that's not necessarily uncommon in the ecosystem and it wouldn't be an "issue" per se. Sometimes this can even be intended, though I too would much prefer users passing the --format flag to Microbundle and being more explicit in such circumstances.

There's also times where we do need to ignore and not match what's in the package.json, like invalid extensions for package type, but maybe we just print out warnings instead on that. Not sure. I do want to improve that.

I didn't skip the docs, I just read it with a different set of assumptions.

I was referring to the mistaken mapping of --modern to "module" and using "exports": { "default": without setting the package to "type": "module". This is fairly esoteric stuff, we don't expect everyone to know it, but it's there for a very good reason.

The --modern to "exports" thing you did just miss, as that's Microbundle-specific. Exports issue is Node, and we don't even look at the key that you provided in that instance as "default" couldn't be used for ESM output. That's not a case of skipping over what you've provided, but not even checking if that exists as we know ahead of time that it wouldn't be correct for modern output. It CAN match up to the CJS though and be totally valid, so it's not necessarily an issue from our side if it does exist. In fact, you using the .js extension would indicate to us that you are trying to match it up to CJS, though that wasn't your intention.

It's tough. This stuff (package.json in general) isn't that explicit and has a lot of overlapping behaviors. What might seem like an obvious issue to you might be a totally valid config for someone else. It's really hard for us to create warning messages because we have to just assume everyone fully understands the implications of what they're writing (or are using our example config).

Looking at it again, I think what mislead me here was the comments saying "where to generate", which I took to mean that I'd be able to specify where to generate the outputs - what this actually is, is "where it will be generated" regardless of what you actually put there.

Not sure what you mean here. You are able to specify where to generate the outputs[1] when you provide valid entries.

I'm sorry, and maybe I'm an idiot, but I'm still not grasping why you couldn't use package.json to establish a meaningful default configuration. Parcel does, and I think that may have been where I first heard the term "zero configuration".

I think this is just going to be an issue of some disagreement over what counts as "configuration". I'd say adding keys to your package.json is config. Granted, if you want to upload your package to NPM it's necessary config, but I'd say config nonetheless.

Microbundle itself is "zero configuration" in the sense that you just run it and you have working output you can run with. The specific environment that you're targeting may require some config on top, but not necessarily. I have used rsync to upload some bundles up to a CDN of mine before. No package.json needed, as usage was just a script tag in HTML.

of course there's a configuration for npm, or you wouldn't have a working package, but there's "zero configuration" for the bundler.

I mean.. that's kind of what's done now. You specify keys you want in order to get a working package and you don't have to pass any flags to Microbundle. You'll get some extra output, but if you're using this with NPM, that's not necessarily a problem. Node and any bundler won't use files unless you provide paths to them in your package.json.

Am I making any sense here? If I'm completely wasting your time here, I really do apologize.

Gosh, certainly don't apologize for something like this. No need, and I'm filling up a notebook of changes to consider lol.

I'll admit, the whole node/JS/TS ecosystem is baffling, confusing and completely overwhelming to me. No matter how many articles, tools, courses and tutorials I plow through, I only seem to get more confused and overwhelmed.

I think I've lost a non-insignificant % of my life dealing with this sort of thing... kinda horrifying to think about 😅. It's confusing and overwhelming as there's a lot of ad-hoc standards and individual projects running with a config/implementation that's specific to them.

I've spent a fair bit of time on build tooling and libraries, and there's always some other tool that doesn't respect a standard you'd think is universal. It's a swell time /s.

(If you can point me to anything that's going to help me make sense of all this craziness, please share. I'm more than willing to learn. I too have spent "many hours in the rabbit hole", but the tunnels seem to go in circles and the exits never seem to lead where I wanted to go. sweat_smile)

Best resource that I know for some of the community-driven specs and their implementations is probably Webpack's package exports docs which covers a bunch of different fields, sub-fields, expected formats, and what tools support it. It's a real mix-n-match of horror, though getting better. It's a ton of very specific information though, so might be overkill to do anything more than skim. You probably don't need to know which tools support which subset of the "browser" field spec, for example.

Node on the other hand is much simpler: their docs on "type" and "exports" are quite good, make sure to follow the links in both sections. It goes into some common issues/hazards as well as recommended configurations.

---

[1]: there are some limitations there, like the TS entry name you ran into, as well as some issues where file paths aren't fully resolved from what's listed in the `package.json`. This is generally only an issue if you're trying to output different formats in different directories, though this will actually be supported when #896 is merged.

[mindplay-dk]

if you happen to get both of these configurations exactly right, and/or assuming your required configuration exactly matches the defaults of the tool, then yes, you get something that works;

I'm a bit confused by this. Getting both configurations exactly right (matching output format to package.json keys) is an issue that's intrinsic to the ecosystem. This will be an issue with or without Microbundle. Using tsc would present this issue, for example.

Sure, but I don't think tsc ever claimed to be "zero configuration"?

Microbundle itself is "zero configuration" in the sense that you just run it and you have working output you can run with.

To my understanding, that's just "default configuration".

I mean, sure, that could mean you have no configuration, if the default configuration happens to match your expectations and your package.json - but that seems more a matter of whether the tool has any defaults at all, and not whether those defaults make sense in the context of your project.

So look.

I decided to give Parcel 2.0 a shot, and this is what I thought "zero configuration" meant:

{
  "name": "inertion",
  "source": "src/index.ts",
  "main": "dist/index.js",
  "module": "dist/index.mjs",
  "types": "dist/index.d.ts",
  "engines": {
    "node": ">= 16"
  },
  "scripts": {
    ...
    "build": "parcel build"
  },
  ...
}

It's using main, module, types and engines, which I should be specifying anyway, and correctly configured the targets as CommonJS for the main and ES for module - the only thing I had to do was call parcel build. I didn't need any command line configuration overrides - that's what I thought "zero configuration" meant. (It even managed to bundle my .d.ts file, which wasn't a hard requirement for me, but, nice.)

The targets page of Parcel's docs explains for every setting how, why and whether a default is established based on your package.json - if I wanted to override something, I could, or else it just does the "most likely thing" based on configuration that's there anyway.

This is what I thought "zero configuration" meant - it's "plug and play" with the project configuration you should/already have, with the option to override.

That's quite different from just having some defaults for every option, I think.

Gosh, certainly don't apologize for something like this. No need, and I'm filling up a notebook of changes to consider lol.

Very cool. I'm glad you're not taking this the wrong way. 😄

[rschristian]

if you happen to get both of these configurations exactly right, and/or assuming your required configuration exactly matches the defaults of the tool, then yes, you get something that works;

I'm a bit confused by this. Getting both configurations exactly right (matching output format to package.json keys) is an issue that's intrinsic to the ecosystem. This will be an issue with or without Microbundle. Using tsc would present this issue, for example.

Sure, but I don't think tsc ever claimed to be "zero configuration"?

The "zero configuration" portion is entirely separate from this issue is my point. You can run into it anywhere.

To my understanding, that's just "default configuration".

Different interpretations is all. That's what this project is running with as "zero configuration"; you need not provide any config in order to get a tiny module built that works in a number of environments.

[mindplay-dk]

As to the meaning of "zero configuration", from my small sample size of 16 respondents, it's a tie. 🙂

https://twitter.com/mindplaydk/status/1510515106999898115

It seems the original meaning of the term "zero configuration" was in regards to networking - where you just connect to a network, and the computer auto-configures your IP and DNS etc. based on external factors.

So in it's original sense, I think the term "zero configuration" does imply "plug and play", meaning something auto-configures based on what's already there - in that sense, I think Parcel (which, to my knowledge, popularized the use of the term with regards to bundlers) got it right.

Either way, I would love it if Microbundle did more of that. 🙂

[rschristian]

I think Parcel (which, to my knowledge, popularized the use of the term with regards to bundlers)

Your timelines are off, just to be clear, though I'm really done talking about different interpretations of "zero configuration". That's not the definition in use here and you'll just have to accept that.

Microbundle was released and using "zero configuration" in it's ReadMe for 6 months before the first Parcel version was released. Nothing against Parcel, they can use whatever language they'd like, but continuing to hold it up as the de facto standard when it's much newer (it was one of the very last tools to the bundler party) is an odd choice.