Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

The code published to npm is minified #4065

Open
Raynos opened this issue Jul 9, 2023 · 26 comments
Open

The code published to npm is minified #4065

Raynos opened this issue Jul 9, 2023 · 26 comments
Labels

Comments

@Raynos
Copy link

Raynos commented Jul 9, 2023

Since the code in the npm tarball is minified by default ( https://socket.dev/npm/package/preact/files/10.15.1/dist/preact.mjs )

It's nearly impossible to debug or profile the preact framework when using it to build my app.

Please do not publish minified code and let me run my own minifier.

@JoviDeCroock
Copy link
Member

JoviDeCroock commented Jul 9, 2023

The main goal of us doing so has been to

  • ensure the lowest size, we know what properties are safe to mangle
  • consistent performance, no minifier/transpiler would insert a snippet that performs poorly

I do see what you are saying about traces becoming unreadable while investigating bugs/performance. It's been one of my main motivators to fight for "readable" dist bundles as well.

However doing so would most likely warrant a major as we do ship properties that we need mangled for our plugin ecosystem, the reconciler is pluggable so devtools/signals/fast-refresh rely on that being consistent.

If you are facing a bug and don't know how to reproduce due to the bundle being unreadable I am open to figure this out together on a GitHub discussion/...

@Raynos
Copy link
Author

Raynos commented Jul 9, 2023

For my personal use case I am likely to copy paste the source code of preact into my application and use it directly so that I can have the unminified version.

I ran into the same issue with react, and was fighting with nextjs in that it does not allow me to set NODE_ENV

It would help me if you committed a version of preact.single-file.js into the git repo so I could copy a single file instead of the entire source code.

The way i used to copy jquery.js into my vendor/jquery.js codebase and also copy vendor/jquery.min.js and control what I load when and where for development / staging / production.

@Raynos
Copy link
Author

Raynos commented Jul 9, 2023

I guess having dist/preact.raw.js next to dist/preact.min.js would also help me, in that I can quickly and easily copy the raw single file source code or import it with require('preact/dist/preact.raw.js')

This would be a non breaking change, just an extra file into the npm publish tarball.

@rschristian
Copy link
Member

For my personal use case I am likely to copy paste the source code of preact into my application and use it directly so that I can have the unminified version.

As Jovi mentioned, this could be an issue as the plugin system relies upon mangled properties for hooking into the reconciler. If you're only using Preact core, you shouldn't run into any issues, but you are locking yourself off from devtools/prefresh/etc.

Unfortunately I'm not sure a mangled, but unminified, distribution file would end up being much more readable... _children still needs to become __k, etc.

@Raynos
Copy link
Author

Raynos commented Jul 10, 2023

I appreciate the technical explanation of why this wouldn't work with the ecosystem and the devtools.

I am however completely blown away that this technical limitation exists. it boggles my mind.

A mangled & unminified build would not really be helpful. I guess all the plugin ecosystem tools would have to support both mangled & unmangled copy of preact with some kind of config setting.

@rschristian
Copy link
Member

rschristian commented Jul 11, 2023

Well, I'm not quite sure what other alternatives exist. The problem is that properties need to remain consistent across versions and across every distribution method. A few CDNs mangle themselves (whether they should is another issue) and bundlers may need some config to ensure both Preact & the add-on have consistently mangled properties.

Mangling ahead of time certainly isn't ideal, but I don't know of any better options that doesn't offload the problem onto end users.

@Raynos
Copy link
Author

Raynos commented Jul 11, 2023

The consistent mangling for ecosystem I can understand.

Would be nice if the ecosystem respects both the unmangled and mangled property names, so I could use either the min or raw version.

@MicahZoltu
Copy link
Contributor

I'm also frustrated by this right now as I have some code that is getting called too frequently and what is happening inside preact is quite opaque. The included TypeScript definition files help a lot with reading the code, but unfortunately Firefox (at least) doesn't unmangle variables in the debugger so this is what I see:
image

My position on this is that minification/mangling isn't worth the costs, especially given that basically every file served over the internet is gzipped. A given word/token (e.g., _children) is essentially one lookup table entry in a gzipped file plus maybe a byte or two of data per reference, thus mangling gives very minimal benefits. Meanwhile, the cost is that it takes significantly longer to debug applications than if the code was un-mangled.

I would encourage comparing just gzipping vs gzipping+mangling to make sure you are actually benefiting in a significant way at least. My guess is that mangling isn't actually saving much. Even if it is, there should still be a way to run without mangling for people who don't want/need mangling.

@dliebner
Copy link

Is there a way to import the source version?

I'm currently using:
import * as preactSignals from '@preact/signals-core';

I tried this:
import * as preactSignals from '@preact/signals-core/src/index.ts';
... but rollup complained about it not being JavaScript even though I have the typescript plugin loaded, and other typescript imports in my bundle are working fine.

@rschristian
Copy link
Member

Is there a way to import the source version?

As mentioned, no.

rollup complained about it not being JavaScript even though I have the typescript plugin loaded, and other typescript imports in my bundle are working fine.

You'd need to configure your TS plugin to compile node_modules, that's not standard behavior.

@matthewmueller
Copy link

I was able to get a PoC working with esbuild by:

Tweaking preact/package.json:

  1. Add preact/src as a valid export
"./src": {
  "types": "./src/index.d.ts",
  "browser": "./src/index.js",
  "import": "./src/index.js"
},
  1. Tweak preact/jsx-dev-runtime to point to src:
"./jsx-dev-runtime": {
  "types": "./jsx-runtime/src/index.d.ts",
  "browser": "./jsx-runtime/src/index.js",
  "import": "./jsx-runtime/src/index.js"
},

Currently jsx-dev-runtime points to the same thing as jsx-runtime, so unless that's for historical reasons, might make sense to tweak for dev. Otherwise an ./jsx-runtime/src export could be added.

Then you adjust the alias settings in esbuild for development:

Alias: map[string]string{
	"preact":             "preact/src",
	"preact/jsx-runtime": "preact/jsx-dev-runtime",
},

I'd be open to submitting a PR if this is something the Preact team would consider!

@johrstrom
Copy link

johrstrom commented Jul 9, 2024

  1. Add preact/src as a valid export

I'd be open to submitting a PR if this is something the Preact team would consider!

I would +1 this as somewhere in between 10.20.1 and 10.22.1 an exported function started being compiled as $2 that's conflicts with jQuery/Ruby on Rails javascript.

I'm currently exploring something like this in esbuild to force it to resolve to the src instead of the dist directories.

const preactSrcResolvePlugin = {
  name: 'preactSrcResolve',
  setup(build) {
    build.onResolve({ filter: /preact/ }, args => {
      const basePath = `${__dirname}/node_modules/preact`;
      if (args.path == 'preact') {
        return { path: `${basePath}/src/index.js` }
      } else if(args.path == 'preact/hooks') {
        return { path: `${basePath}/hooks/src/index.js` }
      }
    })
  },
}

@lilnasy
Copy link
Contributor

lilnasy commented Nov 26, 2024

Seeing how unminified code is published on npm as well, I ended up with the following patch, and another one in prefresh to unmangle the fields. It added 253 bytes to the brotli'd javascript size.

preact.patch
diff --git a/package.json b/package.json
index 166b927ce89d1ff7ecc259912cfc6ff26fbe5f13..27baf4ea22b3e84464ff4fb8749f1530ef3d5594 100644
--- a/package.json
+++ b/package.json
@@ -20,56 +20,56 @@
         "types": "./src/index-5.d.ts"
       },
       "types": "./src/index.d.ts",
-      "browser": "./dist/preact.module.js",
+      "browser": "./src/index.js",
       "umd": "./dist/preact.umd.js",
       "import": "./dist/preact.mjs",
       "require": "./dist/preact.js"
     },
     "./compat": {
       "types": "./compat/src/index.d.ts",
-      "browser": "./compat/dist/compat.module.js",
+      "browser": "./compat/src/index.js",
       "umd": "./compat/dist/compat.umd.js",
       "import": "./compat/dist/compat.mjs",
       "require": "./compat/dist/compat.js"
     },
     "./debug": {
       "types": "./debug/src/index.d.ts",
-      "browser": "./debug/dist/debug.module.js",
+      "browser": "./debug/src/index.js",
       "umd": "./debug/dist/debug.umd.js",
       "import": "./debug/dist/debug.mjs",
       "require": "./debug/dist/debug.js"
     },
     "./devtools": {
       "types": "./devtools/src/index.d.ts",
-      "browser": "./devtools/dist/devtools.module.js",
+      "browser": "./devtools/src/index.js",
       "umd": "./devtools/dist/devtools.umd.js",
       "import": "./devtools/dist/devtools.mjs",
       "require": "./devtools/dist/devtools.js"
     },
     "./hooks": {
       "types": "./hooks/src/index.d.ts",
-      "browser": "./hooks/dist/hooks.module.js",
+      "browser": "./hooks/src/index.js",
       "umd": "./hooks/dist/hooks.umd.js",
       "import": "./hooks/dist/hooks.mjs",
       "require": "./hooks/dist/hooks.js"
     },
     "./test-utils": {
       "types": "./test-utils/src/index.d.ts",
-      "browser": "./test-utils/dist/testUtils.module.js",
+      "browser": "./test-utils/src/index.js",
       "umd": "./test-utils/dist/testUtils.umd.js",
       "import": "./test-utils/dist/testUtils.mjs",
       "require": "./test-utils/dist/testUtils.js"
     },
     "./jsx-runtime": {
       "types": "./jsx-runtime/src/index.d.ts",
-      "browser": "./jsx-runtime/dist/jsxRuntime.module.js",
+      "browser": "./jsx-runtime/src/index.js",
       "umd": "./jsx-runtime/dist/jsxRuntime.umd.js",
       "import": "./jsx-runtime/dist/jsxRuntime.mjs",
       "require": "./jsx-runtime/dist/jsxRuntime.js"
     },
     "./jsx-dev-runtime": {
       "types": "./jsx-runtime/src/index.d.ts",
-      "browser": "./jsx-runtime/dist/jsxRuntime.module.js",
+      "browser": "./jsx-runtime/src/index.js",
       "umd": "./jsx-runtime/dist/jsxRuntime.umd.js",
       "import": "./jsx-runtime/dist/jsxRuntime.mjs",
       "require": "./jsx-runtime/dist/jsxRuntime.js"
@prefresh/core.patch
diff --git a/src/constants.js b/src/constants.js
index b39d40371ec6e156678fb9ca2a2719aa94bb1b1b..5260b84312057693b76002dd7b79d036d70e7e85 100644
--- a/src/constants.js
+++ b/src/constants.js
@@ -1,13 +1,13 @@
-export const VNODE_COMPONENT = '__c';
+export const VNODE_COMPONENT = '_component';
 export const NAMESPACE = '__PREFRESH__';
-export const COMPONENT_HOOKS = '__H';
-export const HOOKS_LIST = '__';
-export const EFFECTS_LIST = '__h';
-export const RERENDER_COUNT = '__r';
-export const CATCH_ERROR_OPTION = '__e';
-export const COMPONENT_DIRTY = '__d';
-export const VNODE_DOM = '__e';
-export const VNODE_CHILDREN = '__k';
-export const HOOK_VALUE = '__';
-export const HOOK_ARGS = '__H';
-export const HOOK_CLEANUP = '__c';
+export const COMPONENT_HOOKS = '__hooks';
+export const HOOKS_LIST = '_list';
+export const EFFECTS_LIST = '_pendingEffects';
+export const RERENDER_COUNT = '_rerenderCount';
+export const CATCH_ERROR_OPTION = '_catchError';
+export const COMPONENT_DIRTY = '_dirty';
+export const VNODE_DOM = '_dom';
+export const VNODE_CHILDREN = '_children';
+export const HOOK_VALUE = '_value';
+export const HOOK_ARGS = '_args';
+export const HOOK_CLEANUP = '_cleanup';
diff --git a/src/index.js b/src/index.js
index 6c80ef4ed9ca406299c2ed0271c87eb19e6011d5..0ca44d43722191e9adfd31e6bb331daf9b9c6341 100644
--- a/src/index.js
+++ b/src/index.js
@@ -54,7 +54,7 @@ function replaceComponent(OldType, NewType, resetHookState) {
   pendingUpdates = pendingUpdates.filter(p => p[0] !== OldType);
 
   vnodes.forEach(vnode => {
-    if (!vnode.__c || !vnode.__c.__P) return;
+    if (!vnode._component || !vnode._component._parentDom) return;
     // update the type in-place to reference the new component
     vnode.type = NewType;

@lilnasy
Copy link
Contributor

lilnasy commented Dec 14, 2024

Just ran into a pesky hydration bug. Preact kept remounting dom elements, rerunning the @starting-style transitions, and causing a flash. The html elements preact was hydrating were identical to what it rendered; logically I had done everything right. I had no idea what it wanted from me!

It turned out there was some whitespace around the elements and preact was diffing against blank TextNodes. There's no way I would have found that out if I wasn't stepping through the unminified preact code in the debugger.

@rschristian
Copy link
Member

It turned out there was some whitespace around the elements and preact was diffing against blank TextNodes. There's no way I would have found that out if I wasn't stepping through the unminified preact code in the debugger.

Were you using preact/debug? We have warnings for hydration mismatches now.

If you were, and you didn't see any warnings, please open an issue.

@JoviDeCroock
Copy link
Member

JoviDeCroock commented Dec 14, 2024

Oh, that sound pesky, I guess the issue was that your jsx is structured like hello, {name} which makes it so that JSX will produce 2 VNodes but in HTML they are just 1 text node. Or was it that you had pretty html rendering which makes it so that the HTML parser will inject an empty text node at every new line?

I guess animation state is lost when an element gets inserted before the animating element or what was the cause here?

Trying to understand the issue as it's not necessarily related to minified code but more to these potential footguns of Virtual VS real HTML. If you have the time, mind informing us further and maybe in a separate issue or discussion?

As Ryan is saying, we want to know about these scenario's and account for them so please do open the issue.

@lilnasy
Copy link
Contributor

lilnasy commented Dec 14, 2024

I guess the root cause in my particular case was server side rendering. I am using astro, which is whitespace sensitive unlike jsx.

@lilnasy
Copy link
Contributor

lilnasy commented Dec 14, 2024

Yes, I did have preact/debug enabled and it did report that there's a mismatch. It did not help with the specifics as to where what the mismatch was.

@rschristian
Copy link
Member

Hm, to my knowledge Astro shouldn't be an issue there, as I thought all islands had to be in a JS framework? i.e., nothing from .astro files end up in your islands? Is that not the case?

It did not help with the specifics as to where the mismatch was.

Would love a reproduction then, we'd definitely want to fix that if we can.

@lilnasy
Copy link
Contributor

lilnasy commented Dec 14, 2024

I should clarify, it did say where the mismatch was but not that it was trying to match against a TextNode which was the important bit. For all I knew, it was matching against the element itself which came after the server side rendered space.

And the specifics of my bug aside, what I meant to get across by sharing the anecdote was that dependencies play a role in the debugging process.

@JoviDeCroock
Copy link
Member

JoviDeCroock commented Dec 14, 2024

I mean; if you don't want to create an issue I get it but... our code explicitly does not match element and text nodes because they are different nodeTypes so the cause is different than this assumption.

It would mean that there are in fact more element nodes than vnodes or the other way around. Which maybe could be caused by some adjacent node to the root of your island. My guesses here are shots in the dark ofcouese, but I'd prefer to surface these issues without folks having to debug library code.

Most people won't debug library code and instead get stuck 😅 I will let it rest at that because this is all off-topic. Not minifying the way we do is possible, only mangling our properties and not minifying but, it will be a disservice to people using no-build step to create their applications. It's hard to do right for everyone

@lilnasy
Copy link
Contributor

lilnasy commented Dec 14, 2024

It would mean that there are in fact more element nodes than vnodes

You're right, that was the case. To simplify, [ divVNode ] was being compared against [ Text, HTMLDivElement ]. I will share a repro when I can.

@JoviDeCroock
Copy link
Member

JoviDeCroock commented Dec 14, 2024

Hmm, then we probably diff the divs correctly but maybe the unmount of the empty text node causes the state of the HTMLElement to be reset as it moves from position 1 to 0. Maybe it's worth to not unmount empty text nodes for this reason, they should not lead to erroneous child diffing.

@lilnasy
Copy link
Contributor

lilnasy commented Dec 14, 2024

In the meantime maybe this helps:

The message did not make sense to me originally, but I dismissed it thinking engineers are not communicators. Only when I was looking preact/debug's source code did I realise it was supposed to log the diffed element tag names.

Again, because I have the unminified code running, I was able to add this line to node_modules to immediately determine why it wasn't logging the available elements. That's twice in one day!

Here's the repro for the flashing issue: https://stackblitz.com/edit/github-vmfzstte?file=src%2Fpages%2Findex.astro,h1.ts&on=stackblitz

@rschristian
Copy link
Member

rschristian commented Dec 15, 2024

but I dismissed it thinking engineers are not communicators.

Snide comments like this aren't helpful, please try to keep things a bit more positive for people volunteering time.

Again, because I have the unminified code running, I was able to add this line to node_modules to immediately determine why it wasn't logging the available elements. That's twice in one day!

Alternatively, we do offer source maps... you can simply open your debugger and add a breakpoint. I get the value of console.log debugging, it's absolutely my go-to, but our source maps, along with any proper JS debugger, handles this situation (and more) very, very well.

@lilnasy
Copy link
Contributor

lilnasy commented Dec 15, 2024

Sorry. It's hard to communicate tone over text, but I didn't mean to be snide.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
Projects
None yet
Development

No branches or pull requests

8 participants