Standard modules - concept or concrete?
On Sat, Jun 8, 2013 at 5:08 AM, Brian Di Palma <offler at gmail.com> wrote:
The standard modules wiki page ( harmony:modules_standard ) is not clear as to whether what it describes is a concrete proposal and that ES6 will include it or it's purely a concept.
ES6 will definitely provide some set of standard modules. The primary open questions are (a) what will the modules be named and (b) how fine-grained will the module split be.
Is this a prerequisite for static checks in modules ( forbid all globals in modules unless explicitly imported )? I'm sure the checks aren't as harsh, but I'd love such strictness as it can make tooling more powerful and code simpler to understand and follow. The idea that you can grab anything from the global object without first importing it seems wrong.
The global object will still be accessible in modules. Of course, you can create new module loaders with an empty global.
Good, I like the standard modules idea.
On Sat, Jun 8, 2013 at 2:33 PM, Sam Tobin-Hochstadt <samth at ccs.neu.edu> wrote:
The global object will still be accessible in modules. Of course, you can create new module loaders with an empty global.
Umm. It makes porting old code easier.
If we could guarantee that any reference inside a module had to have an import definition I imagine IDEs and development concatenation tools would provide fast feedback when those rules are broken. Why would people use the standard modules if they can just access the global?
Is it expected that developers import things like Date because it would be good practice?
import { Date } from "@std";
or simply
new Date();
I can imagine many people just taking option 2. Seem to make standard modules somewhat redundant, or at least it undermines them. I suppose then the static checks are only to check that a module imports the identifiers that another module exports, is that it? If we can grab anything from the global that means un-imported references can be used all over module code and the environment will just have to shrug its shoulders an accept it. No compile time error. Somewhat disappointed with that.
So modules will be allowed to be polluted by the global state, not just build in globals but any possible user defined global state.
The possibility of strengthening module consistency is off the table?
On Sat, Jun 8, 2013 at 2:00 PM, Brian Di Palma <offler at gmail.com> wrote:
Good, I like the standard modules idea.
On Sat, Jun 8, 2013 at 2:33 PM, Sam Tobin-Hochstadt <samth at ccs.neu.edu> wrote:
The global object will still be accessible in modules. Of course, you can create new module loaders with an empty global.
Umm. It makes porting old code easier.
If we could guarantee that any reference inside a module had to have an import definition I imagine IDEs and development concatenation tools would provide fast feedback when those rules are broken. Why would people use the standard modules if they can just access the global?
Is it expected that developers import things like Date because it would be good practice?
import { Date } from "@std";
or simply
new Date();
I can imagine many people just taking option 2. Seem to make standard modules somewhat redundant, or at least it undermines them. I suppose then the static checks are only to check that a module imports the identifiers that another module exports, is that it? If we can grab anything from the global that means un-imported references can be used all over module code and the environment will just have to shrug its shoulders an accept it. No compile time error. Somewhat disappointed with that.
I think you misunderstand. The requirement that modules not have free variables at compile time includes global references. I expect that development environments won't have a problem handling this or enforcing whatever properties you're looking for.
On Sat, Jun 8, 2013 at 7:07 PM, Sam Tobin-Hochstadt <samth at ccs.neu.edu> wrote:
I think you misunderstand. The requirement that modules not have free variables at compile time includes global references. I expect that development environments won't have a problem handling this or enforcing whatever properties you're looking for.
I think I see what you're saying. Let me just see if I'm correct.
At compile time any references in a module which are not explicitly imported but are language globals will not cause compile errors. Any references which aren't explicitly imported and aren't language globals will cause a compile error?
So
module "test" { new Date(); }
is fine.
While
module "test2" { $ }
will throw an error unless you add the line
import $ from "jquery";
even if jQuery was available in the global scope and had been loaded in by a normal script tag?
On Sat, Jun 8, 2013 at 2:30 PM, Brian Di Palma <offler at gmail.com> wrote:
On Sat, Jun 8, 2013 at 7:07 PM, Sam Tobin-Hochstadt <samth at ccs.neu.edu> wrote:
I think you misunderstand. The requirement that modules not have free variables at compile time includes global references. I expect that development environments won't have a problem handling this or enforcing whatever properties you're looking for.
I think I see what you're saying. Let me just see if I'm correct.
At compile time any references in a module which are not explicitly imported but are language globals will not cause compile errors. Any references which aren't explicitly imported and aren't language globals will cause a compile error?
This is close, but not quite right. The rule is that any unbound
variables in modules are errors. The variables may be bound by import
declarations, or by lexical bindings such as var or let, or by
bindings on the global object, or by top-level let bindings (which
are not on the global object, IIRC).
From: Sam Tobin-Hochstadt
This is close, but not quite right. The rule is that any unbound variables in modules are errors. The variables may be bound by import declarations, or by lexical bindings such as
varorlet, or by bindings on the global object, or by top-levelletbindings (which are not on the global object, IIRC).
Is this correct then?
Date.now();
is checked at compile time and found to be OK, because it is referencing a binding that is a property of the global object that exists at the time of static-checking. But
setTimeout(() =>
asdf(); // (A)
}, 5000);
setTimeout(() =>
window.asdf = () =>; // (B)
}, 1000);
is checked at compile time and found to error, because (A) is referencing a binding that is a not a property of the global object at the time of static checking? (Assuming an asdf binding is not introduced through any of the other mechanisms you mention.) And this is true even though (B) adds such a property to the global object before (A) ever runs?
On Tue, Jun 18, 2013 at 11:29 PM, Domenic Denicola <domenic at domenicdenicola.com> wrote:
From: Sam Tobin-Hochstadt
This is close, but not quite right. The rule is that any unbound variables in modules are errors. The variables may be bound by import declarations, or by lexical bindings such as
varorlet, or by bindings on the global object, or by top-levelletbindings (which are not on the global object, IIRC).Is this correct then?
Yes, the below is all correct.
But a compile-time error as Domenic is wondering? That doesn't seem quite right to me. I would think that those would be runtime errors (as is the case in strict-mode). Otherwise this is a big semantic change that I haven't previously considered.
First, I meant what I said in my previous email -- that program is a compile time error inside a module.
Second, the meaning of that program doesn't change in strict mode. If the reference is evaluated before the assignment, it's a ReferenceError even in non-strict mode. If the assignment is evaluated first, it will work even in strict mode.
OK - I see it on the wiki here:
"Compilation resolves and validates all variable definitions and references"
It still seems odd to me that we're going to check a dynamic object (the global object) at link-time for references. What if the global object is changed after the linking pass, but before the module executes? Does the variable reference still point to the old thing?
// Before linking, window.bar = "before"
// foo.js
bar = "after";
// main.js
import "foo";
console.log(bar); // "before"?
Thanks,
On Wed, Jun 19, 2013 at 11:37 AM, Kevin Smith <zenparsing at gmail.com> wrote:
OK - I see it on the wiki here:
"Compilation resolves and validates all variable definitions and references"
It still seems odd to me that we're going to check a dynamic object (the global object) at link-time for references. What if the global object is changed after the linking pass, but before the module executes? Does the variable reference still point to the old thing?
// Before linking, window.bar = "before" // foo.js bar = "after"; // main.js import "foo"; console.log(bar); // "before"?
This produces "after" -- mutable variables are still mutable. Saving an old version would be pretty strange.
Note that there are situations where you can still get a
ReferenceError inside a module, by deleting properties off of
window. Ruling this out would require:
- not giving modules access to the global scope, or
- changing the global object when compiling modules
Neither of these seem like good ideas.
Note that there are situations where you can still get a ReferenceError inside a module, by deleting properties off of
window. Ruling this out would require:
- not giving modules access to the global scope, or
- changing the global object when compiling modules
Neither of these seem like good ideas.
I agree. I wonder, though, why my previous example should work, but this should fail:
// Before linking, window.bar is not defined
// foo.js
window.bar = "bar";
// main.js
import "foo";
console.log(bar); // Link time error
But if we ran "foo" in a previous linking/execution pass, then it would work.
Since this is a departure from the current non-module handling of global object variable references, what is the motivation?
I would think the advantage of running compile-time checks against the global object is that it can catch errors that we currently use linters for:
// OOPS - forgot this line!
// import { x } from "foo";
function someRareCase() {
x(); // Reference error?
}
That's useful, but it comes at the price of treating the global object as if it were a static thing, and not dynamic. From my point of view, though, a dynamic global object is just how it goes with Javascript. I think this kind of static checking should be left to linters, unless we are adopting a policy of actively discouraging dynamism for the global object.
On Thu, Jun 20, 2013 at 9:55 AM, Kevin Smith <zenparsing at gmail.com> wrote:
I would think the advantage of running compile-time checks against the global object is that it can catch errors that we currently use linters for:
// OOPS - forgot this line! // import { x } from "foo"; function someRareCase() { x(); // Reference error? }That's useful, but it comes at the price of treating the global object as if it were a static thing, and not dynamic. From my point of view, though, a dynamic global object is just how it goes with Javascript. I think this kind of static checking should be left to linters, unless we are adopting a policy of actively discouraging dynamism for the global object.
We could:
- Give up on static checking of unbound variables in modules.
- Take the global object off the scope chain in modules.
- Adopt a compromise.
We think static checking for unbound variables is valuable, and
letting people write console.log without having to import anything
is valuable. Thus, option 3.
On 20 June 2013 15:55, Kevin Smith <zenparsing at gmail.com> wrote:
I would think the advantage of running compile-time checks against the global object is that it can catch errors that we currently use linters for:
// OOPS - forgot this line! // import { x } from "foo"; function someRareCase() { x(); // Reference error? }That's useful, but it comes at the price of treating the global object as if it were a static thing, and not dynamic. From my point of view, though, a dynamic global object is just how it goes with Javascript. I think this kind of static checking should be left to linters, unless we are adopting a policy of actively discouraging dynamism for the global object.
This actually is the sort of thing that can be difficult to check for off-line linters, because the use of global variables may depend on some staged dynamic configuration that a linter cannot easily see, verify, or assume.
This actually is the sort of thing that can be difficult to check for off-line linters, because the use of global variables may depend on some staged dynamic configuration that a linter cannot easily see, verify, or assume.
That's pretty much true for everything about javascript : )
In my usage of linters for this task (using my own linters of course), I would have a predefined set of global variables that are "allowed" - essentially treating the global object as if it were static for the purposes of linting.
We think static checking for unbound variables is valuable, and letting people write
console.logwithout having to import anything is valuable. Thus, option 3.
I think that's fine, if we're willing to discourage dynamic usage of the global object for unbound variables. Static checking of the global object might create some iffy edge cases for users who want to treat the global object as a more dynamic thing.
I wonder, though, if this might create issues for polyfilling:
// polyfillz.js
if (this.Promise === void 0)
this.Promise = function() { ... }
// main.js
import "polyfillz.js";
new Promise();
This would refuse to compile, right? We'd have to introduce all of our polyfills in a separate (previous) compilation/execution cycle.
On Thu, Jun 20, 2013 at 10:26 AM, Kevin Smith <zenparsing at gmail.com> wrote:
I wonder, though, if this might create issues for polyfilling:
// polyfillz.js if (this.Promise === void 0) this.Promise = function() { ... } // main.js import "polyfillz.js"; new Promise();This would refuse to compile, right? We'd have to introduce all of our polyfills in a separate (previous) compilation/execution cycle.
Yes, like so:
<script src="polyfillz.js"/>
Note that this is already the way people suggest using polyfills; see [1] for an example.
Sam
Yes, like so:
<script src="polyfillz.js"/>
Sure. In a server environment, you'd have to do your monkey-patching and then load your "main module" dynamically through the loader api.
On Thu, Jun 20, 2013 at 9:08 AM, Sam Tobin-Hochstadt <samth at ccs.neu.edu> wrote:
On Thu, Jun 20, 2013 at 10:26 AM, Kevin Smith <zenparsing at gmail.com> wrote:
I wonder, though, if this might create issues for polyfilling:
// polyfillz.js if (this.Promise === void 0) this.Promise = function() { ... } // main.js import "polyfillz.js"; new Promise();This would refuse to compile, right? We'd have to introduce all of our polyfills in a separate (previous) compilation/execution cycle.
Yes, like so:
<script src="polyfillz.js"/>Note that this is already the way people suggest using polyfills; see [1] for an example.
I have found that once I have module loading, I want the dependencies to be specified by the modules that use them, either via the declarative dependency syntax or via module loader APIs, and at the very least, avoid script tags as the optimization tools can work solely by tracing module/JS loading APIs. In this case, only the "model" set of modules would care about setting up indexeddb access, not the top level of the app.
Example, this AMD module:
jrburke/carmino/blob/master/www/lib/IDB.js
Asks for "indexedDB!", which is an AMD loader plugin:
jrburke/carmino/blob/master/www/lib/indexedDB.js
which feature detects and uses a module loader API to load a shim if it is needed. So the "IDB" module will not execute until that optional shim work is done.
I believe this will also work via the ES Module Loader API, but calling it out just in case I missed something. I want to be sure there are options that do not require using <script src> tags, except
maybe one to bootstrap a set of Module Loader hooks.
James
Not sure how to answer your question exactly, James, but the takeaway is that under the current design, it is not sufficient to "import" global-object polyfills from the module that uses the polyfills. Global object polyfills must be loaded in a prior compilation/execution cycle.
Bascially, you'll have to somehow (a) setup your global object with polyfills, and then (b) load your main module, with (a) and (b) happening in separate stages.
On Jun 20, 2013 7:53 PM, "James Burke" <jrburke at gmail.com> wrote:
On Thu, Jun 20, 2013 at 9:08 AM, Sam Tobin-Hochstadt <samth at ccs.neu.edu>
wrote:
On Thu, Jun 20, 2013 at 10:26 AM, Kevin Smith <zenparsing at gmail.com>
wrote:
I wonder, though, if this might create issues for polyfilling:
// polyfillz.js if (this.Promise === void 0) this.Promise = function() { ... } // main.js import "polyfillz.js"; new Promise();This would refuse to compile, right? We'd have to introduce all of our polyfills in a separate (previous) compilation/execution cycle.
Yes, like so:
<script src="polyfillz.js"/>Note that this is already the way people suggest using polyfills; see [1] for an example.
I have found that once I have module loading, I want the dependencies to be specified by the modules that use them, either via the declarative dependency syntax or via module loader APIs, and at the very least, avoid script tags as the optimization tools can work solely by tracing module/JS loading APIs. In this case, only the "model" set of modules would care about setting up indexeddb access, not the top level of the app.
Example, this AMD module:
jrburke/carmino/blob/master/www/lib/IDB.js
Asks for "indexedDB!", which is an AMD loader plugin:
jrburke/carmino/blob/master/www/lib/indexedDB.js
which feature detects and uses a module loader API to load a shim if it is needed. So the "IDB" module will not execute until that optional shim work is done.
I believe this will also work via the ES Module Loader API, but calling it out just in case I missed something. I want to be sure there are options that do not require using <script src> tags, except maybe one to bootstrap a set of Module Loader hooks.
Yes, this will work fine. Loader hooks can explicitly add modules using the loader API, allowing them to polyfill in exactly this way.
We think static checking for unbound variables is valuable, and letting people write
console.logwithout having to import anything is valuable. Thus, option 3.
Another option would be to check unbound variables not in the linking phase, but immediately before executing the module body. That would give us the advantage of variable checks, but also allow more flexibility when polyfilling or otherwise tweaking the global object.
The standard modules wiki page ( harmony:modules_standard ) is not clear as to whether what it describes is a concrete proposal and that ES6 will include it or it's purely a concept.
The page does not seem to be linked/mentioned from any of the other module pages ( at least based on a cursory read of the main module pages ).
Is this a prerequisite for static checks in modules ( forbid all globals in modules unless explicitly imported )? I'm sure the checks aren't as harsh, but I'd love such strictness as it can make tooling more powerful and code simpler to understand and follow. The idea that you can grab anything from the global object without first importing it seems wrong.
Are standard modules the future of built-ins for ES?