Implement #each .. in .. #3560

Merged
merged 16 commits into from Apr 23, 2015

Conversation

Projects
None yet
@Slava
Member

Slava commented Jan 26, 2015

<ul>
{{#each person in people}}
  <li>{{person.name}} from {{people.groupName}}</li>
{{/each}}
</ul>
{{#each people}}
  {{@index}}
{{/each}}
{{#let x=y.z f=w.s}}
  {{x}} {{f}} {{w}}
{{/let}}

@Slava Slava added the Project:Blaze label Jan 26, 2015

packages/blaze/builtins.js
@@ -68,7 +68,7 @@ Blaze.Unless = function (conditionFunc, contentFunc, elseFunc) {
/**
* @summary Constructs a View that renders `contentFunc` for each item in a sequence.
* @locus Client
- * @param {Function} argFunc A function to reactively re-run. The function may return a Cursor, an array, null, or undefined.
+ * @param {Function} argFunc A function to reactively re-run. The function must return an object with two fields: '_variable' and '_sequence'. '_variable' is an optional alias for an item available from the Each body, '_sequence' may be a Cursor, an array, null, or undefined. For backwards-compitablity, can return just sequence (Cursor, array) not wrapped into an object.

This comment has been minimized.

@stubailo

stubailo Jan 27, 2015

Contributor

I think this could be more clear if it said something like "just a cursor or an array."

@stubailo

stubailo Jan 27, 2015

Contributor

I think this could be more clear if it said something like "just a cursor or an array."

@stubailo

This comment has been minimized.

Show comment
Hide comment
@stubailo

stubailo Jan 27, 2015

Contributor

Seems legit, needs some tests though!

Contributor

stubailo commented Jan 27, 2015

Seems legit, needs some tests though!

@Slava

This comment has been minimized.

Show comment
Hide comment
@Slava

Slava Jan 27, 2015

Member

I still need to add changes to public-facing docs but the code and tests are done. Can one of you folks review it? @dgreensp @avital

Member

Slava commented Jan 27, 2015

I still need to add changes to public-facing docs but the code and tests are done. Can one of you folks review it? @dgreensp @avital

@stubailo

This comment has been minimized.

Show comment
Hide comment
@stubailo

stubailo Jan 27, 2015

Contributor

Since talking to Greenspan, I was convinced that we can't afford to clone the data context because people might be putting complex objects in there. We might need to do the thing you were saying about adding some scope to the lookup chain. But then we still have the question of what Template.currentData returns..

Contributor

stubailo commented Jan 27, 2015

Since talking to Greenspan, I was convinced that we can't afford to clone the data context because people might be putting complex objects in there. We might need to do the thing you were saying about adding some scope to the lookup chain. But then we still have the question of what Template.currentData returns..

@Slava

This comment has been minimized.

Show comment
Hide comment
@Slava

Slava Jan 27, 2015

Member

It is a shallow clone. How bad is it?
On Jan 27, 2015 8:51 AM, "Sashko Stubailo" notifications@github.com wrote:

Since talking to Greenspan, I was convinced that we can't afford to clone
the data context because people might be putting complex objects in there.
We might need to do the thing you were saying about adding some scope to
the lookup chain. But then we still have the question of what
Template.currentData returns..


Reply to this email directly or view it on GitHub
#3560 (comment).

Member

Slava commented Jan 27, 2015

It is a shallow clone. How bad is it?
On Jan 27, 2015 8:51 AM, "Sashko Stubailo" notifications@github.com wrote:

Since talking to Greenspan, I was convinced that we can't afford to clone
the data context because people might be putting complex objects in there.
We might need to do the thing you were saying about adding some scope to
the lookup chain. But then we still have the question of what
Template.currentData returns..


Reply to this email directly or view it on GitHub
#3560 (comment).

@stubailo

This comment has been minimized.

Show comment
Hide comment
@stubailo

stubailo Jan 27, 2015

Contributor

The problem is that the binding of the methods on a complex object might be wrong, and maybe people are modifying the "this" inside event handlers or something

Contributor

stubailo commented Jan 27, 2015

The problem is that the binding of the methods on a complex object might be wrong, and maybe people are modifying the "this" inside event handlers or something

@dgreensp

This comment has been minimized.

Show comment
Hide comment
@dgreensp

dgreensp Jan 27, 2015

Contributor

The data context can have a prototype, and a lot of apps (e.g. Respondly) use some kind of Model or Component object as the data context, which can't be safely cloned. Some users also mutate the data context object directly, though this is less of an encouraged thing.

I'd love to help figure out how to expose names properly with lexical scope. It would open the door to #let and be useful for @index as well.

Contributor

dgreensp commented Jan 27, 2015

The data context can have a prototype, and a lot of apps (e.g. Respondly) use some kind of Model or Component object as the data context, which can't be safely cloned. Some users also mutate the data context object directly, though this is less of an encouraged thing.

I'd love to help figure out how to expose names properly with lexical scope. It would open the door to #let and be useful for @index as well.

@Slava

This comment has been minimized.

Show comment
Hide comment
@Slava

Slava Jan 27, 2015

Member

Yeah, initially we had a discussion that a proper lexical scoping is required but later we opted-out to a lazier and easier to implement solution (duplication of data context on multiple levels).

Member

Slava commented Jan 27, 2015

Yeah, initially we had a discussion that a proper lexical scoping is required but later we opted-out to a lazier and easier to implement solution (duplication of data context on multiple levels).

@mquandalle mquandalle referenced this pull request in mquandalle/meteor-jade Jan 27, 2015

Closed

each x in y loop #18

@Slava

This comment has been minimized.

Show comment
Hide comment
@Slava

Slava Feb 13, 2015

Member

I updated the PR. Implemented {{#each foo in foos}} with bindings and scoping (rather than changing the data context). Since I was on the path of creating lexical scopes, I implemented {{#let}} as well.

@dgreensp can you please do another round of review for me?

Member

Slava commented Feb 13, 2015

I updated the PR. Implemented {{#each foo in foos}} with bindings and scoping (rather than changing the data context). Since I was on the path of creating lexical scopes, I implemented {{#let}} as well.

@dgreensp can you please do another round of review for me?

@dgreensp

This comment has been minimized.

Show comment
Hide comment
@dgreensp

dgreensp Feb 13, 2015

Contributor

Wow! I will get to it as soon as I can -- early next week. Thanks for implementing this.

Contributor

dgreensp commented Feb 13, 2015

Wow! I will get to it as soon as I can -- early next week. Thanks for implementing this.

@stubailo stubailo referenced this pull request Feb 14, 2015

Closed

@index in Blaze #each #3716

@mitar

This comment has been minimized.

Show comment
Hide comment
@mitar

mitar Feb 14, 2015

Collaborator

The data context can have a prototype, and a lot of apps (e.g. Respondly) use some kind of Model or Component object as the data context, which can't be safely cloned.

What if those objects would be required to implement EJSON clone method?

Collaborator

mitar commented Feb 14, 2015

The data context can have a prototype, and a lot of apps (e.g. Respondly) use some kind of Model or Component object as the data context, which can't be safely cloned.

What if those objects would be required to implement EJSON clone method?

@stubailo

This comment has been minimized.

Show comment
Hide comment
@stubailo

stubailo Feb 14, 2015

Contributor

I think it would be unfortunate if some common patterns stopped being viable because we wanted to implement @index. I think the real problem is the way the data context works - as soon as we can move to a different pattern where there can be multiple variables instead of just one, a lot of things will become a lot nicer/easier.

Contributor

stubailo commented Feb 14, 2015

I think it would be unfortunate if some common patterns stopped being viable because we wanted to implement @index. I think the real problem is the way the data context works - as soon as we can move to a different pattern where there can be multiple variables instead of just one, a lot of things will become a lot nicer/easier.

@mitar

This comment has been minimized.

Show comment
Hide comment
@mitar

mitar Feb 14, 2015

Collaborator

Hm, so I implemented few useful methods here. From my experience changing data context was only done because there was no easy access to template instance. Now (at least in our code) we put all custom stuff into a template instance, having also nice getters to get them across the ancestry.

as soon as we can move to a different pattern where there can be multiple variables instead of just one, a lot of things will become a lot nicer/easier.

What do you mean by this?

Collaborator

mitar commented Feb 14, 2015

Hm, so I implemented few useful methods here. From my experience changing data context was only done because there was no easy access to template instance. Now (at least in our code) we put all custom stuff into a template instance, having also nice getters to get them across the ancestry.

as soon as we can move to a different pattern where there can be multiple variables instead of just one, a lot of things will become a lot nicer/easier.

What do you mean by this?

@stubailo

This comment has been minimized.

Show comment
Hide comment
@stubailo

stubailo Feb 14, 2015

Contributor

Basically, the data context doesn't give that much flexibility, and requires you to think about the invisible omnipresent variable. Basically, "each in" is the first step towards relying less on replacing the data context all the time, and instead adding additional variables that can be referenced by name.

Example 1:

{{#each people}}
  {{name}}
{{/each}}

Example 2:

{{#each person in people}}
  {{person.name}}
{{/each}}

I think the second one has a lot of benefits - in the first example, you have to remember that there's an invisible variable that you can't see, and remember where it comes from and what keys it has. In the second one, it's more visible that you're accessing the name property on a variable person that's coming from the collection people.

There's a reason people don't use the JavaScript with keyword very often, and in fact MDN has a big red warning about it. It's because it often causes weird problems because of invisible overwriting of the scope.

What do you think? Do you agree?

Contributor

stubailo commented Feb 14, 2015

Basically, the data context doesn't give that much flexibility, and requires you to think about the invisible omnipresent variable. Basically, "each in" is the first step towards relying less on replacing the data context all the time, and instead adding additional variables that can be referenced by name.

Example 1:

{{#each people}}
  {{name}}
{{/each}}

Example 2:

{{#each person in people}}
  {{person.name}}
{{/each}}

I think the second one has a lot of benefits - in the first example, you have to remember that there's an invisible variable that you can't see, and remember where it comes from and what keys it has. In the second one, it's more visible that you're accessing the name property on a variable person that's coming from the collection people.

There's a reason people don't use the JavaScript with keyword very often, and in fact MDN has a big red warning about it. It's because it often causes weird problems because of invisible overwriting of the scope.

What do you think? Do you agree?

@mitar

This comment has been minimized.

Show comment
Hide comment
@mitar

mitar Feb 14, 2015

Collaborator

I think it is just a matter of style. In fact, it is just a matter of style. And I doubt that this is a good thing after 1.0 release. You are moving away from Handlebars. You are introducing more concepts for the same thing. You will now have to support both the new and old style. You will introduce discussions of what is better and many ways to do the same thing. At the end of the day it is really a question where you associate the context with. Do you want to name it, or you make it more like a closure.

Maybe I am coming too much from the functional programming background. But having closures is something pretty natural. And this data context for me feels like a closure. Also "with" is a pretty common pattern in functional programming languages. It is called "let ... in" there. So saying that this is bed practice when this is a basic programming construct of any functional programming language is a bit stretch for me. It is true that before there was no way to navigate parent data context, but now this is possible, so one can really already use "with" as a way to overlay new variables on top of old ones. You just traverse the stack until you get a name. Similar to how I am doing with template instances.

I really do not see the benefit for it. For me this is more a syntactic sugar if anything. Just a way to name stuff temporarily. But at the end of the day, this concept of a simple data context of a template is a very powerful one. We do not want to create whole programming language inside a templating engine, no?

What I would rather see is simply make data context's immutable (declared them as that you should not be changing them yourself), if people want to store stuff they should be storing stuff inside a template instance. And then you can provide some getter which allows you to access a data context field in a way that only registers the dependency on that field. This is what you really want at the end. Not the special case for "each in" loop. Because when you are having complicated for each loops (which take some time to render) you anyway externalize internal content into a template:

{{#each person in people}}
  {{> personTemplate person}}
{{/each}}

Because personTemplate is huge! You do not want to inline it there anyway. So this "each in" works for very simply cases anyway. But you want to know there is that this should rerun only when people changes. Not when whole data context change. Currently, one has to be careful with .find() calls to limit data context only to those fields which are necessary in the data context for the template. This is what I think is the issue with separation of concerns. This should not be needed in the code. It should be automatically computed from what I use in the template of template helpers.

Any data contexts are objects anyway. Knowing the name of the object is one, but at the end you have anyway to remember which properties it has. So I do not see much benefit in providing a way to name a property. If you really want to do that, you simply create a template helper now.

Collaborator

mitar commented Feb 14, 2015

I think it is just a matter of style. In fact, it is just a matter of style. And I doubt that this is a good thing after 1.0 release. You are moving away from Handlebars. You are introducing more concepts for the same thing. You will now have to support both the new and old style. You will introduce discussions of what is better and many ways to do the same thing. At the end of the day it is really a question where you associate the context with. Do you want to name it, or you make it more like a closure.

Maybe I am coming too much from the functional programming background. But having closures is something pretty natural. And this data context for me feels like a closure. Also "with" is a pretty common pattern in functional programming languages. It is called "let ... in" there. So saying that this is bed practice when this is a basic programming construct of any functional programming language is a bit stretch for me. It is true that before there was no way to navigate parent data context, but now this is possible, so one can really already use "with" as a way to overlay new variables on top of old ones. You just traverse the stack until you get a name. Similar to how I am doing with template instances.

I really do not see the benefit for it. For me this is more a syntactic sugar if anything. Just a way to name stuff temporarily. But at the end of the day, this concept of a simple data context of a template is a very powerful one. We do not want to create whole programming language inside a templating engine, no?

What I would rather see is simply make data context's immutable (declared them as that you should not be changing them yourself), if people want to store stuff they should be storing stuff inside a template instance. And then you can provide some getter which allows you to access a data context field in a way that only registers the dependency on that field. This is what you really want at the end. Not the special case for "each in" loop. Because when you are having complicated for each loops (which take some time to render) you anyway externalize internal content into a template:

{{#each person in people}}
  {{> personTemplate person}}
{{/each}}

Because personTemplate is huge! You do not want to inline it there anyway. So this "each in" works for very simply cases anyway. But you want to know there is that this should rerun only when people changes. Not when whole data context change. Currently, one has to be careful with .find() calls to limit data context only to those fields which are necessary in the data context for the template. This is what I think is the issue with separation of concerns. This should not be needed in the code. It should be automatically computed from what I use in the template of template helpers.

Any data contexts are objects anyway. Knowing the name of the object is one, but at the end you have anyway to remember which properties it has. So I do not see much benefit in providing a way to name a property. If you really want to do that, you simply create a template helper now.

@steph643

This comment has been minimized.

Show comment
Hide comment
@steph643

steph643 Feb 14, 2015

@mitar, I have to disagree. Maybe I am coming too much from an object-oriented background, but ever since I began with Meteor, I feel like I am constantly fighting with contexts.

I am not against the concept of context, but it has to have a defined meaning (like 'this' in oo-programming). In Blaze, context looks more like a vaporous global variable, with the additional oddity that it can be changed arbitrarily along the line by some block helpers you add in another file.

So a few weeks ago, after a year with Meteor, I came to the conclusion that Blaze context is evil, and I decided to ban its usage:

{{> projNavbar project=project contributor=contributor profile=profile}}

{{#with profile=getProfile}} instead of {{#with getProfile}}

{{#each profiles}}
    {{> profileWithPicture profile=this}}
{{/each}}
{{#each profiles}}
    {{#with profile=this}}
    ...
    {{/with}}
{{/each}}

What @stubailo said sounds reliving to me. I would even encourage MDG to take more radical actions. For example, I would love to have a 'no context' option, which would prevent usage of 'this' and would reset template context even when no arguments are passed.

@mitar, I have to disagree. Maybe I am coming too much from an object-oriented background, but ever since I began with Meteor, I feel like I am constantly fighting with contexts.

I am not against the concept of context, but it has to have a defined meaning (like 'this' in oo-programming). In Blaze, context looks more like a vaporous global variable, with the additional oddity that it can be changed arbitrarily along the line by some block helpers you add in another file.

So a few weeks ago, after a year with Meteor, I came to the conclusion that Blaze context is evil, and I decided to ban its usage:

{{> projNavbar project=project contributor=contributor profile=profile}}

{{#with profile=getProfile}} instead of {{#with getProfile}}

{{#each profiles}}
    {{> profileWithPicture profile=this}}
{{/each}}
{{#each profiles}}
    {{#with profile=this}}
    ...
    {{/with}}
{{/each}}

What @stubailo said sounds reliving to me. I would even encourage MDG to take more radical actions. For example, I would love to have a 'no context' option, which would prevent usage of 'this' and would reset template context even when no arguments are passed.

@dgreensp

This comment has been minimized.

Show comment
Hide comment
@dgreensp

dgreensp Feb 14, 2015

Contributor

@mitar and @steph643, I appreciate your responses!

For some more data on the matter, {{#each person in people}} has been in Ember for a while. (Ember is the main framework for Handlebars, so whether this is "getting away from Handlebars" is debatable.) New information to me is that they are planning to deprecate the old syntax {{#each people}} because beginners/new-comers find it confusing that both are available, and they cite the context-shifting aspect of {{#each people}} as troublesome.

See the "More Consistent Handlebars Scope" section of emberjs/rfcs#15.

(For present purposes we're going to ignore the further proposals they make about #each, which as far as I can tell amount to changing the syntax again to {{#each people as |person|}} and then eventually deprecating #each altogether in favor of a syntax for repeating an element...)

@mitar, it's interesting that you bring up "let" and closures, because by implementing lexical scope for each-in, it's now actually possible to implement "let"! I don't really see an analogy between the behavior of data contexts and let. When you enter an #each or a #with block, you can no longer access some of the names you could before, and a bunch of new names are introduced, the set of which is determined at runtime and not visible from looking at the code. "let" doesn't unset or hide any names and it only introduces a fixed set of names that is known statically and visible in the code.

Contributor

dgreensp commented Feb 14, 2015

@mitar and @steph643, I appreciate your responses!

For some more data on the matter, {{#each person in people}} has been in Ember for a while. (Ember is the main framework for Handlebars, so whether this is "getting away from Handlebars" is debatable.) New information to me is that they are planning to deprecate the old syntax {{#each people}} because beginners/new-comers find it confusing that both are available, and they cite the context-shifting aspect of {{#each people}} as troublesome.

See the "More Consistent Handlebars Scope" section of emberjs/rfcs#15.

(For present purposes we're going to ignore the further proposals they make about #each, which as far as I can tell amount to changing the syntax again to {{#each people as |person|}} and then eventually deprecating #each altogether in favor of a syntax for repeating an element...)

@mitar, it's interesting that you bring up "let" and closures, because by implementing lexical scope for each-in, it's now actually possible to implement "let"! I don't really see an analogy between the behavior of data contexts and let. When you enter an #each or a #with block, you can no longer access some of the names you could before, and a bunch of new names are introduced, the set of which is determined at runtime and not visible from looking at the code. "let" doesn't unset or hide any names and it only introduces a fixed set of names that is known statically and visible in the code.

@mitar

This comment has been minimized.

Show comment
Hide comment
@mitar

mitar Feb 14, 2015

Collaborator

When you enter an #each or a #with block, you can no longer access some of the names you could before, and a bunch of new names are introduced, the set of which is determined at runtime and not visible from looking at the code.

You can always traverse the data context parents. Now that you allow accessing that.

Anyway, I think this is just a matter of style so I do not mind anyway. I got now more scared with this proposal because it looks like it is focusing on wrong things if we want to make templates extendable. (I wrote my questions at the end.)

Collaborator

mitar commented Feb 14, 2015

When you enter an #each or a #with block, you can no longer access some of the names you could before, and a bunch of new names are introduced, the set of which is determined at runtime and not visible from looking at the code.

You can always traverse the data context parents. Now that you allow accessing that.

Anyway, I think this is just a matter of style so I do not mind anyway. I got now more scared with this proposal because it looks like it is focusing on wrong things if we want to make templates extendable. (I wrote my questions at the end.)

@Slava

This comment has been minimized.

Show comment
Hide comment
@Slava

Slava Feb 14, 2015

Member

@mitar I don't see the comparison between #with and let x=y in block valid at all. My thinking is exact the opposite: the new #each .. in .. is built on top of the new #let construct that I implemented in this PR. And the #let construct actually looks a lot closer to let x=y in block you can find in ML or Scheme.

#with works exactly like with keyword in JS: let's just replace our dynamic variable this - the only variable we have in Handlebars! It is OK in JS, since you have a lot of other ways to create new variables, using the var keyword or by creating a new function.

In my plan, I had an idea to introduce explicit arguments to templates like this:

<template name="myTemplate" args="name age">
  <div>Hello {{name}}! Today you are {{age}} years old!</div>
</template>

...
{{> myTemplate name=person.name age=calculateAge}}

(The exact syntax can be different). Here Templates accept named argument explicitly, this might make it easier to use scoping rules more and rely less on the dynamic data context.

Member

Slava commented Feb 14, 2015

@mitar I don't see the comparison between #with and let x=y in block valid at all. My thinking is exact the opposite: the new #each .. in .. is built on top of the new #let construct that I implemented in this PR. And the #let construct actually looks a lot closer to let x=y in block you can find in ML or Scheme.

#with works exactly like with keyword in JS: let's just replace our dynamic variable this - the only variable we have in Handlebars! It is OK in JS, since you have a lot of other ways to create new variables, using the var keyword or by creating a new function.

In my plan, I had an idea to introduce explicit arguments to templates like this:

<template name="myTemplate" args="name age">
  <div>Hello {{name}}! Today you are {{age}} years old!</div>
</template>

...
{{> myTemplate name=person.name age=calculateAge}}

(The exact syntax can be different). Here Templates accept named argument explicitly, this might make it easier to use scoping rules more and rely less on the dynamic data context.

@mitar

This comment has been minimized.

Show comment
Hide comment
@mitar

mitar Feb 15, 2015

Collaborator

#with works exactly like with keyword in JS: let's just replace our dynamic variable this - the only variable we have in Handlebars! It is OK in JS, since you have a lot of other ways to create new variables, using the var keyword or by creating a new function.

I think this is just a question of style. When I implemented my own Scheme interpreter, I used simply a stack of environments. let ... in just pushed a new environment on top of the stack. And exactly this is what with does, if you look at this as current top-level environment. So instead of looking at it as one and only variable, look at it as (data) context. In all programming language you have only one local scope. And that local scope is unnamed. I do not know what is the fuzz about that here? Data context is template local scope. No programming language provides you a way to name the current local scope. In python you can access it though locals, but there is no name for it. For me this is equal to locals(). All fields in that data context object are variables in the local scope. You can access them directly in the template, the same how you would access local variables in the local scope.

The only current difference is that if a variable is missing at the current level, we do not automatically traverse previous scope levels. But this is really simple to implement as a template helper now that we have Template.parentData() and you can simple traverse all data contexts until you get one with the field.

So I really think that you are solving a no-problem here. If you would like to address something, thing about something really new. How to really make templates reusable and composable. Now you are just moving between very similar paradigms, without really any qualitative jump forward.

Please look at Django. They had the same templating language for 10 years. And they have a great way to make templates modular and extendable. This is how you make a stable community. Do not change templates just because one style is 0.5% better than other in some opinion. Do you really want that all StackOverflow posts and all examples which already exist there would be obsolete?

What I would like to see is way to extend templates. That I could say, OK, this is base template, now make the same one, just add before it this, and after it that. Or that I could parse a template through a filter. Genshi is so amazing templating engine. The most powerful I have ever seen. Even more than Django. That is a qualitative jump forward. It is not the easiest to use (XSLT is really not an easy way to develop things), but something like that would be powerful to have.

Collaborator

mitar commented Feb 15, 2015

#with works exactly like with keyword in JS: let's just replace our dynamic variable this - the only variable we have in Handlebars! It is OK in JS, since you have a lot of other ways to create new variables, using the var keyword or by creating a new function.

I think this is just a question of style. When I implemented my own Scheme interpreter, I used simply a stack of environments. let ... in just pushed a new environment on top of the stack. And exactly this is what with does, if you look at this as current top-level environment. So instead of looking at it as one and only variable, look at it as (data) context. In all programming language you have only one local scope. And that local scope is unnamed. I do not know what is the fuzz about that here? Data context is template local scope. No programming language provides you a way to name the current local scope. In python you can access it though locals, but there is no name for it. For me this is equal to locals(). All fields in that data context object are variables in the local scope. You can access them directly in the template, the same how you would access local variables in the local scope.

The only current difference is that if a variable is missing at the current level, we do not automatically traverse previous scope levels. But this is really simple to implement as a template helper now that we have Template.parentData() and you can simple traverse all data contexts until you get one with the field.

So I really think that you are solving a no-problem here. If you would like to address something, thing about something really new. How to really make templates reusable and composable. Now you are just moving between very similar paradigms, without really any qualitative jump forward.

Please look at Django. They had the same templating language for 10 years. And they have a great way to make templates modular and extendable. This is how you make a stable community. Do not change templates just because one style is 0.5% better than other in some opinion. Do you really want that all StackOverflow posts and all examples which already exist there would be obsolete?

What I would like to see is way to extend templates. That I could say, OK, this is base template, now make the same one, just add before it this, and after it that. Or that I could parse a template through a filter. Genshi is so amazing templating engine. The most powerful I have ever seen. Even more than Django. That is a qualitative jump forward. It is not the easiest to use (XSLT is really not an easy way to develop things), but something like that would be powerful to have.

@Slava

This comment has been minimized.

Show comment
Hide comment
@Slava

Slava Feb 15, 2015

Member

@mitar right now we are not talking about Sashko's brainstorming hackpad. We are just talking about this new feature that still allows you to use the data context (but creates a more flexible way to achieve everything data context could with semantics closer to mainstream PL and JS itself). So let's not discuss Django keeping the same system for 10 years in this thread. Extending templates is also not covered in this PR and this is a feature that has nothing to do with this PR.

I think this is just a question of style. When I implemented my own Scheme interpreter, I used simply a stack of environments. let ... in just pushed a new environment on top of the stack. And exactly this is what with does, if you look at this as current top-level environment.

Exactly. Isn't it what is called "dynamic scoping", something that is not used as a main way to bind variables in any modern PL except for ELISP?

There is also a big difference in the data context: to access a variable from a part of the "stack" that is not on the top, you need to specify a path relative to your current position on stack. So if you want to access a variable from a parent you would access it as ../foo, one stack up is ../../foo and so on. Same in the helpers, you would need to know the exact number of jumps you need to make to access something with parentData. Later, if you move your code around, you would need to go and fix it all up? If you want to use this template in a similar but a differently structured call-stack, your template just wouldn't work? In PLs with dynamic variables, at least you can just access it by name and the interpreter will pull out the correct binding.

Member

Slava commented Feb 15, 2015

@mitar right now we are not talking about Sashko's brainstorming hackpad. We are just talking about this new feature that still allows you to use the data context (but creates a more flexible way to achieve everything data context could with semantics closer to mainstream PL and JS itself). So let's not discuss Django keeping the same system for 10 years in this thread. Extending templates is also not covered in this PR and this is a feature that has nothing to do with this PR.

I think this is just a question of style. When I implemented my own Scheme interpreter, I used simply a stack of environments. let ... in just pushed a new environment on top of the stack. And exactly this is what with does, if you look at this as current top-level environment.

Exactly. Isn't it what is called "dynamic scoping", something that is not used as a main way to bind variables in any modern PL except for ELISP?

There is also a big difference in the data context: to access a variable from a part of the "stack" that is not on the top, you need to specify a path relative to your current position on stack. So if you want to access a variable from a parent you would access it as ../foo, one stack up is ../../foo and so on. Same in the helpers, you would need to know the exact number of jumps you need to make to access something with parentData. Later, if you move your code around, you would need to go and fix it all up? If you want to use this template in a similar but a differently structured call-stack, your template just wouldn't work? In PLs with dynamic variables, at least you can just access it by name and the interpreter will pull out the correct binding.

@mitar

This comment has been minimized.

Show comment
Hide comment
@mitar

mitar Feb 15, 2015

Collaborator

So let's not discuss Django keeping the same system for 10 years in this thread.

I mentioned that as an example where changing things would really add some benefit. Here, for me it looks just a question of style which libraries could address them as well.

to access a variable from a part of the "stack" that is not on the top, you need to specify a path relative to your current position on stack.

No you don't. You can use a simple helper to do that. Templates already provide things needed to implement it. Of course you could define a default helper or default traversal to parent data contexts, how I described above, but that traversal is orthogonal to changes in this pull request. In this pull request you are trying to force naming of scopes. But no language I am aware of has naming of scopes.

Yes, currently with vanilla Meteor you have to define static number of levels to find both template instance or correct data context. But we addressed this with simple API additions in meteor-template-extension. If you have a template instance, you can do templateInstance.get('foo') and it will return you the first field in template instance chain which contains that field. Similarly, for data contexts, you can do Template.parentData(hasField('foo')).foo to achieve the same (we could even wrap this into a simple get helper if it shows as a common pattern, personally I prefer using template instances for custom contexts, and data context really just for data).

Later, if you move your code around, you would need to go and fix it all up?

Not in our code using those template extensions which did not require any changes to templates themselves. Question of changing template syntax is in my opinion just a question of changing style and I argue that changing the style is less important than having stable templating languages people can rely upon. This is why I referenced Django and its stability.

If you want to use this template in a similar but a differently structured call-stack, your template just wouldn't work?

Not in our code using those template extensions.

In PLs with dynamic variables, at least you can just access it by name and the interpreter will pull out the correct binding.

This can be done with a simple helper.

What cannot be done is extending templates. We have to do ugly internal stuff to make other methods in the template extensions package work.

I think what I am saying is that in my opinion you are spending time on a wrong thing. In meantime community has found solutions and are proposing those solutions. Maybe an user study would be useful before we just theorize which one of those approaches is better for users. Yours:

  • modifying the syntax of #each
  • mine, leaving syntax as it is, just automatically traverse to parent data context when a field is not found in the current one, with ../../foo and ./foo syntax to no traversal

Which problem are you solving? Is it a real problem or just perceived problem? Will it really allow users to do more? Or would some other solutions improve it as well, or even better?

In my opinion, yea, add #each ... in, this is really just a style. Do not deprecate old style. But to really improve things which are missing, not just a question of style, focus on getting things which community seeks in template extensions package into Meteor proper in some form or another. Those are the real problems which has to be addressed. Not style ones. How we copy things around. How we override templates. How we provide easy declarative ways to access parent data contexts and template instances without having to hard-code number of steps. In template extensions package we are proposing one approach. I would love to see more discussion about how exactly to address that. This #each ... in feels more like a sugar thing. And I think it is wrong to confuse it with real changes needed.

Collaborator

mitar commented Feb 15, 2015

So let's not discuss Django keeping the same system for 10 years in this thread.

I mentioned that as an example where changing things would really add some benefit. Here, for me it looks just a question of style which libraries could address them as well.

to access a variable from a part of the "stack" that is not on the top, you need to specify a path relative to your current position on stack.

No you don't. You can use a simple helper to do that. Templates already provide things needed to implement it. Of course you could define a default helper or default traversal to parent data contexts, how I described above, but that traversal is orthogonal to changes in this pull request. In this pull request you are trying to force naming of scopes. But no language I am aware of has naming of scopes.

Yes, currently with vanilla Meteor you have to define static number of levels to find both template instance or correct data context. But we addressed this with simple API additions in meteor-template-extension. If you have a template instance, you can do templateInstance.get('foo') and it will return you the first field in template instance chain which contains that field. Similarly, for data contexts, you can do Template.parentData(hasField('foo')).foo to achieve the same (we could even wrap this into a simple get helper if it shows as a common pattern, personally I prefer using template instances for custom contexts, and data context really just for data).

Later, if you move your code around, you would need to go and fix it all up?

Not in our code using those template extensions which did not require any changes to templates themselves. Question of changing template syntax is in my opinion just a question of changing style and I argue that changing the style is less important than having stable templating languages people can rely upon. This is why I referenced Django and its stability.

If you want to use this template in a similar but a differently structured call-stack, your template just wouldn't work?

Not in our code using those template extensions.

In PLs with dynamic variables, at least you can just access it by name and the interpreter will pull out the correct binding.

This can be done with a simple helper.

What cannot be done is extending templates. We have to do ugly internal stuff to make other methods in the template extensions package work.

I think what I am saying is that in my opinion you are spending time on a wrong thing. In meantime community has found solutions and are proposing those solutions. Maybe an user study would be useful before we just theorize which one of those approaches is better for users. Yours:

  • modifying the syntax of #each
  • mine, leaving syntax as it is, just automatically traverse to parent data context when a field is not found in the current one, with ../../foo and ./foo syntax to no traversal

Which problem are you solving? Is it a real problem or just perceived problem? Will it really allow users to do more? Or would some other solutions improve it as well, or even better?

In my opinion, yea, add #each ... in, this is really just a style. Do not deprecate old style. But to really improve things which are missing, not just a question of style, focus on getting things which community seeks in template extensions package into Meteor proper in some form or another. Those are the real problems which has to be addressed. Not style ones. How we copy things around. How we override templates. How we provide easy declarative ways to access parent data contexts and template instances without having to hard-code number of steps. In template extensions package we are proposing one approach. I would love to see more discussion about how exactly to address that. This #each ... in feels more like a sugar thing. And I think it is wrong to confuse it with real changes needed.

@dgreensp

This comment has been minimized.

Show comment
Hide comment
@dgreensp

dgreensp Feb 15, 2015

Contributor

Minor clarification, Slava: Scheme's environment chain is the chain of lexical scopes, not dynamic.

This PR includes each-in and #let. I don't think each-in is very controversial. The point of #let is to behave precisely like a "let" in Scheme (or JS). No matter how much I try, I can't twist it around into the opposite of that, so that arguing in favor of Scheme's lexical scoping becomes an argument against {{#let}}.

The hallmark of lexical scoping is that when you write var x = 1 in JavaScript, say, it exposes x to the current function and functions that are inside it, where "inside" means they are nested in your text editor, i.e. lexically. Functions you call do not see x, and nested functions (like callbacks) do not have to dig up through the scopes of the code that called them to get to x. Similarly, the environment chain in Scheme only includes nested scopes that are nested lexically, in the code.

There is definitely a valid point about how if you think of the data context as a scope, you might come to different conclusions than if you think about it as an object whose properties are dumped out into the scope. However, the fact that the data context is often set to an object that comes out of Mongo, for example by {{#each someCursor}}, or other times a model object with arbitrary fields and methods, makes it hard to treat it as a scope. In order for {{#with x=1}}...{{#with y=2}}... to let you use both x and y inside, we would have to make {{#each cursor1}}...{{#each cursor2}}... expose all the fields of both objects inside! If a document in cursor2 were missing an optional property, we would read through to the document from cursor1. Also, this would happen regardless of whether {{#each cursor2}} were nested lexically in the same template or occurred in a different template. There's no such problem with {{#let x=1}}...{{#let y=2}}... and {{#each a in cursor1}}...{{#each b in cursor2}}... where x, y, a, and b are lexically scoped.

There do exist Handlebars variants where lookups search all the enclosing data contexts. I think they use one at AirBnB in Rendr, if I remember correctly. It works if you are keeping in mind all the fields of all the model objects on the data context stack, and you are mindful of collisions. It hurts composability quite a lot, though.

@mitar, we have template reusability and composability in mind at all times. We are working towards a world where templates take arguments, and data doesn't leak into things where it doesn't belong while at other times being strangely inaccessible. You are asking for a focus on increased composability, and almost in the same breath asking that names always filter down from parent components. You say our focus is in the wrong place if we are thinking about improving scoping, but then admit you find Meteor's built-in scoping unusable and have overridden it?

What does "forcing naming of scopes" mean to you, and what are you worried about losing (not from this PR but presumably in the future)?

Contributor

dgreensp commented Feb 15, 2015

Minor clarification, Slava: Scheme's environment chain is the chain of lexical scopes, not dynamic.

This PR includes each-in and #let. I don't think each-in is very controversial. The point of #let is to behave precisely like a "let" in Scheme (or JS). No matter how much I try, I can't twist it around into the opposite of that, so that arguing in favor of Scheme's lexical scoping becomes an argument against {{#let}}.

The hallmark of lexical scoping is that when you write var x = 1 in JavaScript, say, it exposes x to the current function and functions that are inside it, where "inside" means they are nested in your text editor, i.e. lexically. Functions you call do not see x, and nested functions (like callbacks) do not have to dig up through the scopes of the code that called them to get to x. Similarly, the environment chain in Scheme only includes nested scopes that are nested lexically, in the code.

There is definitely a valid point about how if you think of the data context as a scope, you might come to different conclusions than if you think about it as an object whose properties are dumped out into the scope. However, the fact that the data context is often set to an object that comes out of Mongo, for example by {{#each someCursor}}, or other times a model object with arbitrary fields and methods, makes it hard to treat it as a scope. In order for {{#with x=1}}...{{#with y=2}}... to let you use both x and y inside, we would have to make {{#each cursor1}}...{{#each cursor2}}... expose all the fields of both objects inside! If a document in cursor2 were missing an optional property, we would read through to the document from cursor1. Also, this would happen regardless of whether {{#each cursor2}} were nested lexically in the same template or occurred in a different template. There's no such problem with {{#let x=1}}...{{#let y=2}}... and {{#each a in cursor1}}...{{#each b in cursor2}}... where x, y, a, and b are lexically scoped.

There do exist Handlebars variants where lookups search all the enclosing data contexts. I think they use one at AirBnB in Rendr, if I remember correctly. It works if you are keeping in mind all the fields of all the model objects on the data context stack, and you are mindful of collisions. It hurts composability quite a lot, though.

@mitar, we have template reusability and composability in mind at all times. We are working towards a world where templates take arguments, and data doesn't leak into things where it doesn't belong while at other times being strangely inaccessible. You are asking for a focus on increased composability, and almost in the same breath asking that names always filter down from parent components. You say our focus is in the wrong place if we are thinking about improving scoping, but then admit you find Meteor's built-in scoping unusable and have overridden it?

What does "forcing naming of scopes" mean to you, and what are you worried about losing (not from this PR but presumably in the future)?

@mitar

This comment has been minimized.

Show comment
Hide comment
@mitar

mitar Feb 15, 2015

Collaborator

However, the fact that the data context is often set to an object that comes out of Mongo, for example by {{#each someCursor}}, or other times a model object with arbitrary fields and methods, makes it hard to treat it as a scope.

It is normal that scopes have functions defined in them. So object methods become such functions. Object fields become variables in the scope. someCursor is then more like iterable in Python. But yes, that is a bit of a special case. :-)

There's no such problem with {{#let x=1}}...{{#let y=2}}... and {{#each a in cursor1}}...{{#each b in cursor2}}... where x, y, a, and b are lexically scoped.

You have a valid point here. I see the benefits of #each ... in for cursors, but I still do not understand why introducing #let when we could just use #with we already have?

It hurts composability quite a lot, though.

Yes, that why in template extensions we are not providing a way to traverse data contexts to find a field, but instead search for a data context which is an instance of some JavaScript class. We use collection's transform to transform each document from a collection into a JavaScript object, and then we use Template.parentData(isInstanceOf(DocumentFooBar)) to get that data context, no matter where in the data context stack it is.

What does "forcing naming of scopes" mean to you

I understood @Slava above, that with #each person in persons will now name the previously unnamed scope as person.

and what are you worried about losing (not from this PR but presumably in the future)?

I worry that by offering to many ways to do the same thing we will introduce confusion. That instead of fixing core issues we are doing band-aids. And that in the future this will lead to deprecating of features like normal #each.

As I said, I like things proposed in this pull request in a way. I like sweet/sugary things. :-) But from things beside the style/syntax I got a feeling that we are also going into one other direction. Where using whole data contexts will be discouraged (or even deprecated). And my worry was that this is maybe just because we do not yet have nice tools to work with data contexts in place and that why they feel cumbersome. So I wanted to present some other alternatives which can make work with data contexts quite nice, declarative, decomposable. That it is maybe just a question on how you look at what a data context is.

Anyway, I expressed my concerns. I think you took them into consideration and this is all what I hoped for. Maybe I do not agree with all details here, but I do not want to take more of your time anyway.

Collaborator

mitar commented Feb 15, 2015

However, the fact that the data context is often set to an object that comes out of Mongo, for example by {{#each someCursor}}, or other times a model object with arbitrary fields and methods, makes it hard to treat it as a scope.

It is normal that scopes have functions defined in them. So object methods become such functions. Object fields become variables in the scope. someCursor is then more like iterable in Python. But yes, that is a bit of a special case. :-)

There's no such problem with {{#let x=1}}...{{#let y=2}}... and {{#each a in cursor1}}...{{#each b in cursor2}}... where x, y, a, and b are lexically scoped.

You have a valid point here. I see the benefits of #each ... in for cursors, but I still do not understand why introducing #let when we could just use #with we already have?

It hurts composability quite a lot, though.

Yes, that why in template extensions we are not providing a way to traverse data contexts to find a field, but instead search for a data context which is an instance of some JavaScript class. We use collection's transform to transform each document from a collection into a JavaScript object, and then we use Template.parentData(isInstanceOf(DocumentFooBar)) to get that data context, no matter where in the data context stack it is.

What does "forcing naming of scopes" mean to you

I understood @Slava above, that with #each person in persons will now name the previously unnamed scope as person.

and what are you worried about losing (not from this PR but presumably in the future)?

I worry that by offering to many ways to do the same thing we will introduce confusion. That instead of fixing core issues we are doing band-aids. And that in the future this will lead to deprecating of features like normal #each.

As I said, I like things proposed in this pull request in a way. I like sweet/sugary things. :-) But from things beside the style/syntax I got a feeling that we are also going into one other direction. Where using whole data contexts will be discouraged (or even deprecated). And my worry was that this is maybe just because we do not yet have nice tools to work with data contexts in place and that why they feel cumbersome. So I wanted to present some other alternatives which can make work with data contexts quite nice, declarative, decomposable. That it is maybe just a question on how you look at what a data context is.

Anyway, I expressed my concerns. I think you took them into consideration and this is all what I hoped for. Maybe I do not agree with all details here, but I do not want to take more of your time anyway.

@dgreensp

This comment has been minimized.

Show comment
Hide comment
@dgreensp

dgreensp Feb 15, 2015

Contributor

We are very interested in user feedback. Thank you for engaging us! I think I understand now that you are mainly talking about a proposal or style where the data context is inherited, and wanting to know if that's the world we are designing for or not?

We've discussed ideas before where you might be able to write something like {{Form.errors}} and it will look up the stack for an instance of Form and get the errors. I think of this as a way to dynamically bind one component to another. What we're discussing with each-in is how to plumb data within a single template. For example (somewhat artificial for brevity's sake):

{{#each p in people}}
  {{#each c in chairs}}
    {{c.size}} {{p.name}} {{c.color}} {{p.height}} {{c.type}}
  {{/each}}
{{/each}}

Or:

{{#let p=myPerson}}
  {{#let c=myChair}}
    {{c.size}} {{p.name}} {{c.color}} {{p.height}} {{c.type}}
  {{/let}}
{{/let}}

Where as the proposal that searches all the data contexts looks like this, and you can't tell where each field comes from:

{{#each people}}
  {{#each chairs}}
    {{size}} {{name}} {{color}} {{height}} {{type}}
  {{/each}}
{{/each}}

Maybe that's not what you meant at all -- you clarified that your helpers look at the types of things -- but it sounds like you are having to define a lot of helpers right now to achieve what you want.

My point about how objects that come out of the database, or model objects with methods, are sort of different from scopes, was mainly about how names get into scope. Normally names get into a scope by appearing in the source code. A model object with methods gains a name when someone adds a new instance method -- which is usually not a backwards-compatibility concern, but would be in this case! A database object gains a field when someone performs a database operation. Again, not usually how a name gets into a scope.

However, there is one very important point here: Currently, since there is no real guidance on how to use the data context and how to use the template instance, we probably have a situation where people are using them in different ways. For example, some users probably use the data context as the Model, and others the Component, and others the scope. Likewise with template instances. We should definitely keep that in mind as we figure out how best to serve our users. We will almost certainly have to standardize on something, though. As you said, too many ways to do things leads to confusion.

If you've picked up that we're kind of down on the data context, that's true -- a lot of what it does would be done better by real local variables (statically declared), template arguments, and better access to the template instance. However, we will keep in mind what you've said.

Contributor

dgreensp commented Feb 15, 2015

We are very interested in user feedback. Thank you for engaging us! I think I understand now that you are mainly talking about a proposal or style where the data context is inherited, and wanting to know if that's the world we are designing for or not?

We've discussed ideas before where you might be able to write something like {{Form.errors}} and it will look up the stack for an instance of Form and get the errors. I think of this as a way to dynamically bind one component to another. What we're discussing with each-in is how to plumb data within a single template. For example (somewhat artificial for brevity's sake):

{{#each p in people}}
  {{#each c in chairs}}
    {{c.size}} {{p.name}} {{c.color}} {{p.height}} {{c.type}}
  {{/each}}
{{/each}}

Or:

{{#let p=myPerson}}
  {{#let c=myChair}}
    {{c.size}} {{p.name}} {{c.color}} {{p.height}} {{c.type}}
  {{/let}}
{{/let}}

Where as the proposal that searches all the data contexts looks like this, and you can't tell where each field comes from:

{{#each people}}
  {{#each chairs}}
    {{size}} {{name}} {{color}} {{height}} {{type}}
  {{/each}}
{{/each}}

Maybe that's not what you meant at all -- you clarified that your helpers look at the types of things -- but it sounds like you are having to define a lot of helpers right now to achieve what you want.

My point about how objects that come out of the database, or model objects with methods, are sort of different from scopes, was mainly about how names get into scope. Normally names get into a scope by appearing in the source code. A model object with methods gains a name when someone adds a new instance method -- which is usually not a backwards-compatibility concern, but would be in this case! A database object gains a field when someone performs a database operation. Again, not usually how a name gets into a scope.

However, there is one very important point here: Currently, since there is no real guidance on how to use the data context and how to use the template instance, we probably have a situation where people are using them in different ways. For example, some users probably use the data context as the Model, and others the Component, and others the scope. Likewise with template instances. We should definitely keep that in mind as we figure out how best to serve our users. We will almost certainly have to standardize on something, though. As you said, too many ways to do things leads to confusion.

If you've picked up that we're kind of down on the data context, that's true -- a lot of what it does would be done better by real local variables (statically declared), template arguments, and better access to the template instance. However, we will keep in mind what you've said.

@mitar

This comment has been minimized.

Show comment
Hide comment
@mitar

mitar Feb 15, 2015

Collaborator

I agree with your examples above. The double each is really a mess. Not sure though why you have to do let instead of with:

{{#with p=myPerson}}
  {{#with c=myChair}}
    {{c.size}} {{p.name}} {{c.color}} {{p.height}} {{c.type}}
  {{/with}}
{{/with}}

Because with replaces whole data context inside? But with the traversal going on to parent data context then if p would not be found in inner data context, it would be found in the parent one? OK, this is not completely different idea, but maybe worth considering if we do not want to change things too much. Maybe for some cases this is OK:

{{#with defaults}}
  {{#each localChanges}}
    {{> documentTemplate}}
  {{/each}}
{{/with}}

A model object with methods gains a name when someone adds a new instance method -- which is usually not a backwards-compatibility concern, but would be in this case! A database object gains a field when someone performs a database operation. Again, not usually how a name gets into a scope.

I found this as a good thing. :-) I like the idea of helpers and object methods being merged as one, and then inheritance in searching for the first defined.

We should definitely keep that in mind as we figure out how best to serve our users. We will almost certainly have to standardize on something, though. As you said, too many ways to do things leads to confusion.

I completely agree. I think the initial confusion was because there was no way to access template instance inside template helpers. So one had to misuse data contexts. Now, the issue is that there is no way to access parent template instances by default. But once I introduced parent method on template instances, things really became very intuitive for us. You use template instances as scopes. You add stuff you want, you use get to traverse scopes if you do not want to hard-code the number of levels (so you can refactor your templates and template inclusions without issues). You use the same code both in template helpers, created, rendered, destroyed, event handlers. And you do not modify data contexts at all. They are coming from the database or you generate custom ones as needed in helpers, but reactively based on data. In this way you do have to worry about reactivity overriding your changes to the data context. I would even propose that conceptually data contexts should be declared immutable. In fact it is a immutable part of template instance, template.data (sadly not a reactive source). Having data contexts immutable might be also a great optimization, like used in Om.

If you've picked up that we're kind of down on the data context, that's true -- a lot of what it does would be done better by real local variables (statically declared), template arguments, and better access to the template instance. However, we will keep in mind what you've said.

My main issue with data context is that it declares Tracker dependency on the whole document. So if II am passing DB document to the template, and I am using only one field in the template, I have to limit the DB query to only that one field if I want to optimize stuff. And if I add another field in the template, I have to update the code. I think this is the main concern separation issue I have with data contexts (and lack of getters).

Collaborator

mitar commented Feb 15, 2015

I agree with your examples above. The double each is really a mess. Not sure though why you have to do let instead of with:

{{#with p=myPerson}}
  {{#with c=myChair}}
    {{c.size}} {{p.name}} {{c.color}} {{p.height}} {{c.type}}
  {{/with}}
{{/with}}

Because with replaces whole data context inside? But with the traversal going on to parent data context then if p would not be found in inner data context, it would be found in the parent one? OK, this is not completely different idea, but maybe worth considering if we do not want to change things too much. Maybe for some cases this is OK:

{{#with defaults}}
  {{#each localChanges}}
    {{> documentTemplate}}
  {{/each}}
{{/with}}

A model object with methods gains a name when someone adds a new instance method -- which is usually not a backwards-compatibility concern, but would be in this case! A database object gains a field when someone performs a database operation. Again, not usually how a name gets into a scope.

I found this as a good thing. :-) I like the idea of helpers and object methods being merged as one, and then inheritance in searching for the first defined.

We should definitely keep that in mind as we figure out how best to serve our users. We will almost certainly have to standardize on something, though. As you said, too many ways to do things leads to confusion.

I completely agree. I think the initial confusion was because there was no way to access template instance inside template helpers. So one had to misuse data contexts. Now, the issue is that there is no way to access parent template instances by default. But once I introduced parent method on template instances, things really became very intuitive for us. You use template instances as scopes. You add stuff you want, you use get to traverse scopes if you do not want to hard-code the number of levels (so you can refactor your templates and template inclusions without issues). You use the same code both in template helpers, created, rendered, destroyed, event handlers. And you do not modify data contexts at all. They are coming from the database or you generate custom ones as needed in helpers, but reactively based on data. In this way you do have to worry about reactivity overriding your changes to the data context. I would even propose that conceptually data contexts should be declared immutable. In fact it is a immutable part of template instance, template.data (sadly not a reactive source). Having data contexts immutable might be also a great optimization, like used in Om.

If you've picked up that we're kind of down on the data context, that's true -- a lot of what it does would be done better by real local variables (statically declared), template arguments, and better access to the template instance. However, we will keep in mind what you've said.

My main issue with data context is that it declares Tracker dependency on the whole document. So if II am passing DB document to the template, and I am using only one field in the template, I have to limit the DB query to only that one field if I want to optimize stuff. And if I add another field in the template, I have to update the code. I think this is the main concern separation issue I have with data contexts (and lack of getters).

@mitar mitar referenced this pull request Feb 17, 2015

Closed

Handlebars version update #489

@Slava

This comment has been minimized.

Show comment
Hide comment
@Slava

Slava Mar 3, 2015

Member

As was pointed out in a thread on Meteor forums, accessing variables defined with #let or #each in is hard in event handlers.

Member

Slava commented Mar 3, 2015

As was pointed out in a thread on Meteor forums, accessing variables defined with #let or #each in is hard in event handlers.

@mitar

This comment has been minimized.

Show comment
Hide comment
@mitar

mitar Mar 25, 2015

Collaborator

After more thought, I like it. :-)

Collaborator

mitar commented Mar 25, 2015

After more thought, I like it. :-)

Slava added some commits Jan 26, 2015

#each .. in extends the data context for children
Regular #each ({{#each items}}) sets a new data context for each child view to
the item of the sequence.

The new #each ({{#each item in items}}) preserves the data context and extends
it to have the variable like 'item'
Reimplement the `each foo in foos` with the scope lookup
Remove the old implementation with modifying the data context,
implement it in the new way to keep the same data context but lookup
iterator in the scope.
@Slava

This comment has been minimized.

Show comment
Hide comment
@Slava

Slava Apr 1, 2015

Member

I rebased everything on top of the latest devel. Should be easier to merge later ;)

Member

Slava commented Apr 1, 2015

I rebased everything on top of the latest devel. Should be easier to merge later ;)

@Sewdn

This comment has been minimized.

Show comment
Hide comment
@Sewdn

Sewdn Apr 2, 2015

nice! Any idea in what release this will be released?

Sewdn commented Apr 2, 2015

nice! Any idea in what release this will be released?

@Slava Slava merged commit 4a11850 into devel Apr 23, 2015

1 check passed

default The author has signed the Meteor Contributor Agreement.
Details

@Slava Slava deleted the each-in branch Apr 23, 2015

Slava added a commit that referenced this pull request Apr 23, 2015

}
- return null;
+
+ // 5. throw an error when called: nothing is found

This comment has been minimized.

@dgreensp

dgreensp May 8, 2015

Contributor

This isn't all the "not found" case, it's the "data context" case.

@dgreensp

dgreensp May 8, 2015

Contributor

This isn't all the "not found" case, it's the "data context" case.

This comment has been minimized.

@mitar

mitar May 8, 2015

Collaborator

I can fix this as part of #4036 ;-)

@mitar

mitar May 8, 2015

Collaborator

I can fix this as part of #4036 ;-)

This comment has been minimized.

@dgreensp

dgreensp May 8, 2015

Contributor

Sure! I know I said I would merge that one.

@dgreensp

dgreensp May 8, 2015

Contributor

Sure! I know I said I would merge that one.

This comment has been minimized.

@mitar

mitar May 8, 2015

Collaborator

Fixed. :-)

@mitar

mitar May 8, 2015

Collaborator

Fixed. :-)

+ */
+Blaze.Let = function (bindingsFunc, contentFunc) {
+ var view = Blaze.View('let', contentFunc);
+ Blaze._attachBindingsToView(bindingsFunc(), view);

This comment has been minimized.

@dgreensp

dgreensp May 8, 2015

Contributor

I have to look into what's going on here more, but we shouldn't be calling a function like bindingsFunc() from this function (a View factory) if it's a function that is to be reactively re-rendered.

@dgreensp

dgreensp May 8, 2015

Contributor

I have to look into what's going on here more, but we shouldn't be calling a function like bindingsFunc() from this function (a View factory) if it's a function that is to be reactively re-rendered.

@dgreensp

This comment has been minimized.

Show comment
Hide comment
@dgreensp

dgreensp May 12, 2015

Contributor

This will break apps that do {{#each helper arg}}. Is that something we thought about?

Contributor

dgreensp commented May 12, 2015

This will break apps that do {{#each helper arg}}. Is that something we thought about?

@dgreensp

This comment has been minimized.

Show comment
Hide comment
@dgreensp

dgreensp May 12, 2015

Contributor

Ok, I fixed bugs on devel. Consider this reviewed.

Contributor

dgreensp commented May 12, 2015

Ok, I fixed bugs on devel. Consider this reviewed.

@Slava

This comment has been minimized.

Show comment
Hide comment
@Slava

Slava May 12, 2015

Member

Thank you!

On Tue, May 12, 2015 at 3:46 PM, David Greenspan notifications@github.com
wrote:

Ok, I fixed bugs on devel. Consider this reviewed.


Reply to this email directly or view it on GitHub
#3560 (comment).

Member

Slava commented May 12, 2015

Thank you!

On Tue, May 12, 2015 at 3:46 PM, David Greenspan notifications@github.com
wrote:

Ok, I fixed bugs on devel. Consider this reviewed.


Reply to this email directly or view it on GitHub
#3560 (comment).

@avital

This comment has been minimized.

Show comment
Hide comment
@avital

avital May 12, 2015

Contributor

@dgreensp @Slava Looks like the frontpage app doesn't work off devel. We use {{#each jobs this}}... I guess I'll just fix it by getting rid of the second argument and instead using a new helper?

But yes, this will definitely break people's apps... We might have to think about this some more.

Contributor

avital commented May 12, 2015

@dgreensp @Slava Looks like the frontpage app doesn't work off devel. We use {{#each jobs this}}... I guess I'll just fix it by getting rid of the second argument and instead using a new helper?

But yes, this will definitely break people's apps... We might have to think about this some more.

@avital

This comment has been minimized.

Show comment
Hide comment
@avital

avital May 12, 2015

Contributor

We also had {{#each most_recent_posts 10}} that I'm changing to {{#each ten_most_recent_posts}}

Contributor

avital commented May 12, 2015

We also had {{#each most_recent_posts 10}} that I'm changing to {{#each ten_most_recent_posts}}

@dgreensp

This comment has been minimized.

Show comment
Hide comment
@dgreensp

dgreensp May 12, 2015

Contributor

You can add parentheses around (jobs this) to fix, right? Go
subexpressions!

But yes, this may be a problem. Proposal: if we don't see "in", do the old
thing.

On Tuesday, May 12, 2015, Avital Oliver notifications@github.com wrote:

@dgreensp https://github.com/dgreensp @Slava https://github.com/Slava
Looks like the frontpage app doesn't work off devel. We use {{#each jobs
this}}... I guess I'll just fix it by getting rid of the second argument
and instead using a new helper?

But yes, this will definitely break people's apps... We might have to
think about this some more.


Reply to this email directly or view it on GitHub
#3560 (comment).

Contributor

dgreensp commented May 12, 2015

You can add parentheses around (jobs this) to fix, right? Go
subexpressions!

But yes, this may be a problem. Proposal: if we don't see "in", do the old
thing.

On Tuesday, May 12, 2015, Avital Oliver notifications@github.com wrote:

@dgreensp https://github.com/dgreensp @Slava https://github.com/Slava
Looks like the frontpage app doesn't work off devel. We use {{#each jobs
this}}... I guess I'll just fix it by getting rid of the second argument
and instead using a new helper?

But yes, this will definitely break people's apps... We might have to
think about this some more.


Reply to this email directly or view it on GitHub
#3560 (comment).

@avital

This comment has been minimized.

Show comment
Hide comment
@avital

avital May 12, 2015

Contributor

Ah, right! {{#each (most_recent_posts 10)}} works. I think it's not too bad to ask people to change their apps in this way -- it's probably clearer than our original syntax.

Contributor

avital commented May 12, 2015

Ah, right! {{#each (most_recent_posts 10)}} works. I think it's not too bad to ask people to change their apps in this way -- it's probably clearer than our original syntax.

@dgreensp

This comment has been minimized.

Show comment
Hide comment
@dgreensp

dgreensp May 13, 2015

Contributor

I guess we should have an error message that suggests this. Or just
implement back-compat (because it is easy), and consider throwing a
deprecation warning?

On Tuesday, May 12, 2015, Avital Oliver notifications@github.com wrote:

Ah, right! {{#each (most_recent_posts 10)}} works. I think it's not too
bad to ask people to change their apps in this way -- it's probably clearer
than our original syntax.


Reply to this email directly or view it on GitHub
#3560 (comment).

Contributor

dgreensp commented May 13, 2015

I guess we should have an error message that suggests this. Or just
implement back-compat (because it is easy), and consider throwing a
deprecation warning?

On Tuesday, May 12, 2015, Avital Oliver notifications@github.com wrote:

Ah, right! {{#each (most_recent_posts 10)}} works. I think it's not too
bad to ask people to change their apps in this way -- it's probably clearer
than our original syntax.


Reply to this email directly or view it on GitHub
#3560 (comment).

@stubailo

This comment has been minimized.

Show comment
Hide comment
@stubailo

stubailo May 13, 2015

Contributor

Yeah, I think this is an unnecessary breaking change, and is more of an oversight than something intentional.

Contributor

stubailo commented May 13, 2015

Yeah, I think this is an unnecessary breaking change, and is more of an oversight than something intentional.

@dgreensp

This comment has been minimized.

Show comment
Hide comment
@dgreensp

dgreensp May 13, 2015

Contributor

Ok, fixed in e823459

Contributor

dgreensp commented May 13, 2015

Ok, fixed in e823459

@stubailo

This comment has been minimized.

Show comment
Hide comment
@stubailo

stubailo May 13, 2015

Contributor

Thanks @dgreensp!

Contributor

stubailo commented May 13, 2015

Thanks @dgreensp!

@JulianKingman

This comment has been minimized.

Show comment
Hide comment
@JulianKingman

JulianKingman May 16, 2015

Thanks guys, the {{@index}} is going to be really useful, glad to see it's close to being implemented.

Thanks guys, the {{@index}} is going to be really useful, glad to see it's close to being implemented.

@mquandalle mquandalle referenced this pull request in mquandalle/meteor-jade May 19, 2015

Open

webstorm compatibility #140

@rclai

This comment has been minimized.

Show comment
Hide comment
@rclai

rclai May 22, 2015

Hey guys, were you guys aware that Handlebars had implemented something like this here?

I told them to document it.

rclai commented May 22, 2015

Hey guys, were you guys aware that Handlebars had implemented something like this here?

I told them to document it.

@Slava

This comment has been minimized.

Show comment
Hide comment
@Slava

Slava May 22, 2015

Member

@rclai for some reason I cannot find the doc right now, but we looked at every syntax used in different templating languages and we noticed that Ember pushed Handlebars into implementing 2 different syntaxes. One, just the same as described in this thread, and the newer one, like described on your link.

After the internal conversation, nobody liked the direction of the pipes syntax, it felt like it was trying too hard to be like ruby.

Member

Slava commented May 22, 2015

@rclai for some reason I cannot find the doc right now, but we looked at every syntax used in different templating languages and we noticed that Ember pushed Handlebars into implementing 2 different syntaxes. One, just the same as described in this thread, and the newer one, like described on your link.

After the internal conversation, nobody liked the direction of the pipes syntax, it felt like it was trying too hard to be like ruby.

@sujith3g

This comment has been minimized.

Show comment
Hide comment
@sujith3g

sujith3g Aug 27, 2015

@Slava : How to use this @index in a template ? I have tried {{@index}} inside an #each block, and it gave me "Expected IDENTIFIER" error.

@Slava : How to use this @index in a template ? I have tried {{@index}} inside an #each block, and it gave me "Expected IDENTIFIER" error.

This comment has been minimized.

Show comment
Hide comment
@Konafets

Konafets Sep 8, 2015

I ran in the same error. Any hints?

I ran in the same error. Any hints?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment