As I advocated on the call, I think that by selecting "strict mode" a developer is really choosing to restrict themselves to using a subset of the complete language. The future-proofing issues of this relate to the possibility that there might be multiple such subsets that a developer might need to choose among. Should there be multiple named "strict modes" to choose among, how should they be named, can "strictness" of a mode increase in future versions, etc?
I think some of the controversy could be eliminated if we simply eliminate the term "Strict Mode". Instead I propose we have a "Use Subset" directive and that we name specific subsets in a meaningful and generally positive manner. For example, since the motivation of most of the proposed restrictions in ES3.1 strict mode is to provide a safer subset language I suggest that we call this subset "safety" (or perhaps "safety1" or "safetyA" or "safety2008" implying that in the future different safe subsets might be defined and we don't want to give undo importance to this initial one).
So, the first line of a "strict mode" compilation unit would now look like"
"use subset safety"
I would suggest that we actually define "use subset" such that a comma separated list of subsets is allowed. So, if somebody decided to define a subset that only contained the features of ES3 you might see something like this:
"use subset safety,ES3"
Since subsets are sets of restrictions, listing multiple subsets means to take the union of the restrictions imposed by all of the listed subsets. So "use subset safty,ES3" means that this compilation unit may only use those features defined by ECMA 262-3 and which are not excluded by the "safety" subset. So, assuming that "safety" excludes use of the with statement, such a compilation unit could not include use of the with statement nor could it include any use of a feature that is new in ES3.1 or ES4.
Future versions of ECMAScript may add exclusions to a subset defined by an earlier version as long as the added exclusions only restrict capabilities that didn't exist in the earlier version. For example, ES 4 in supporting the ES3.1 "safety" subset but add to it any features that are added in ES 4 but which are considered to be unsafe.
A future version may not add exclusion to an pre-existing subset that restricts features that existed when the original subset was defined. For example if ES3.14 or ES5 decided that the for-in statement was unsafe, it could not add that restriction to the "safety" subset. However, it could define a new subset named perhaps "safety2010" that includes all the restrictions of the "safety" subset and in addition disallows use of the "for" statement.
If a compilation unit specifies a subset that is not known to the implementation that is processing it, that subset restriction is simply ignored. The code in the unit is still either valid or invalid on its own merit just like is the case when no subset had been specified.
Following up on the "Strict Mode" discussion...
As I advocated on the call, I think that by selecting "strict mode" a developer is really choosing to restrict themselves to using a subset of the complete language. The future-proofing issues of this relate to the possibility that there might be multiple such subsets that a developer might need to choose among. Should there be multiple named "strict modes" to choose among, how should they be named, can "strictness" of a mode increase in future versions, etc?
I think some of the controversy could be eliminated if we simply eliminate the term "Strict Mode". Instead I propose we have a "Use Subset" directive and that we name specific subsets in a meaningful and generally positive manner. For example, since the motivation of most of the proposed restrictions in ES3.1 strict mode is to provide a safer subset language I suggest that we call this subset "safety" (or perhaps "safety1" or "safetyA" or "safety2008" implying that in the future different safe subsets might be defined and we don't want to give undo importance to this initial one).
So, the first line of a "strict mode" compilation unit would now look like"
"use subset safety"
I would suggest that we actually define "use subset" such that a comma separated list of subsets is allowed. So, if somebody decided to define a subset that only contained the features of ES3 you might see something like this:
"use subset safety,ES3"
Since subsets are sets of restrictions, listing multiple subsets means to take the union of the restrictions imposed by all of the listed subsets. So "use subset safty,ES3" means that this compilation unit may only use those features defined by ECMA 262-3 and which are not excluded by the "safety" subset. So, assuming that "safety" excludes use of the with statement, such a compilation unit could not include use of the with statement nor could it include any use of a feature that is new in ES3.1 or ES4.
Future versions of ECMAScript may add exclusions to a subset defined by an earlier version as long as the added exclusions only restrict capabilities that didn't exist in the earlier version. For example, ES 4 in supporting the ES3.1 "safety" subset but add to it any features that are added in ES 4 but which are considered to be unsafe.
A future version may not add exclusion to an pre-existing subset that restricts features that existed when the original subset was defined. For example if ES3.14 or ES5 decided that the for-in statement was unsafe, it could not add that restriction to the "safety" subset. However, it could define a new subset named perhaps "safety2010" that includes all the restrictions of the "safety" subset and in addition disallows use of the "for" statement.
If a compilation unit specifies a subset that is not known to the implementation that is processing it, that subset restriction is simply ignored. The code in the unit is still either valid or invalid on its own merit just like is the case when no subset had been specified.
_____________________________________________
From: Pratap Lakshman (VJ#SDK)
Sent: Tuesday, June 24, 2008 11:43 AM
To: Adam Peller; Sam Ruby; Mark S. Miller; Allen Wirfs-Brock; Pratap Lakshman (VJ#SDK)
Subject: Brief Minutes [RE: ES3.1 WG phone conference 24 June 08:00 PT]
Here are brief minutes from our call.
Please take a look, and let me know if you want any changes by your EOD.
I'll upload it to the wiki and send a copy to Patrick Charollais (ECMA) for posting on the ECMA site tomorrow night (Redmond time).
Attendees
Adam Peller (IBM)
Sam Ruby (IBM)
Mark Miller (Google)
Allen Wirfs-Brock (Microsoft)
Pratap Lakshman (Microsoft)
Agenda
On posting the latest draft to the wiki
Getters/Setters
Decimal
Setting up a review based on Lars' feedback on the 11 June draft
Minutes
Would like to add a couple more items to agenda that we can get to if we have the time (1) inconsistence language like "as if by the expression ..." pervasive in ES3 (e.g. section 11.1.4 Array Initializer: "create a new array as if by the expression new Array()"; needs to be fixed in ES3.1 (2) ES3/ES4 based review.
On posting the latest draft to the wiki
The latest draft has been uploaded to the wiki. This the draft as of 24 June 2008 - it has all the edits related to the statics on Object, the introduction of the [[Extensible]] property, the revised notation for JSON, and placeholders for Decimal - by next week we should try to have a technically complete draft - the draft may still have some place holders but it should still be enough to get circulated within TC39 as an artifact that can be reviewed - by the time we meet in Oslo, each place holder must have a supplementary doc that can be discussed F2F.
'const' needs to be introduced into the grammar - what does it mean to be a reserved word? - you cannot use it as a variable name - note that we have introduced the ability to use is as the name of a property though - const has letrec like scoping - not subject to hoisting.
reformed scoping needs to be introduced - but for doing that we may want to introduce the notion of a 'block activation' as an expository device
'abc'[i] is using the ES3 like algorithm convention - can we use the new convention that eliminates the need for writing gotos - pratapL to investigate - also, for all new functionality we need to worry about argument conditioning - for e.g. if an algorithm expects to be handed an object, make sure we call toObject on the argument that is passed in.
Getters/Setters
Surface syntax spec from Kris looked Ok - need to ensure it is using the Meta APIs - Allen to check surface syntax integration.
Decimal
Decimal changes are isolated and can be done without impacting content in the rest of the doc - Sam can use the latest draft from the wiki as a starting point.
Strict mode
2 controversies coupled to each other - proposed ES4 wants to keep 'with' in their strict mode; ES3.1 wants to remove 'with' in its strict mode - need to be clear about the purpose of strict mode - the other is "what do we mean by subset?" - there was one formal definition from Lars, and Doug had called out a less formal notion - we may end up using Doug's less formal notion (especially if we cannot resolve the 'with' issue).
Binary strict mode is naively limiting - each version may want to allow/limit specific features - strict mode allows users to say 'I am subscribing to this particular subset, and I am aware of all its limitations' - but if the subset is a moving target how do they do that? - sure, we can introduce a more elaborate mechanism - it would make it easier for us to not have to resolve arguments - but it would open up a larger combinatorial space of possibilities - the SunOS vs. MacOS problem; the former was highly customizable, however you invariably got it wrong; the latter was not customizable, but what you got was good enough to get the job done - more knobs need not mean better.
An elaborate mechanism enables chaos at the composability boundary - but how does it matter if you are opaquely including a module? - Caja will use ES3.1 strict mode for all uncajoled code - OTOH an elaborate mechanism can enforce a subsetting profile - languages can now more directly enforce Caja like semantics.
'use strict 3.1' could be a potential syntax to turn on 3.1 level strict mode - with the possibility that the 3.1 may be replaced with a list - instead of tying it to the language version number, we can also just say something like 'use strict a' or 'use strict 1' - can also use Perl-style 'use strict no with' (where we mention the specific restriction we want to enable) - that seems a good idea too - in any case, named restriction sets for ES3.1 and proposed ES4 will be useful to discuss this proposal further.
Meeting adjourned.
pratap
_____________________________________________
From: Pratap Lakshman (VJ#SDK)
Sent: Tuesday, June 24, 2008 6:46 PM
To: Pratap Lakshman (VJ#SDK); crock at yahoo-inc.com; Mark S. Miller; Kris Zyp; Mike Cowlishaw; Adam Peller; Sam Ruby; Lars Hansen; ggaren at apple.com; Allen Wirfs-Brock
Subject: ES3.1 WG phone conference 24 June 08:00 PT
Agenda:
(1) On posting the latest draft to the wiki
(2) Getters/Setters
(3) Decimal
(4) setting up a review based on Lars' feedback on the 11 June draft
Let me know if you want to add any items to the agenda
Here is the dial-in information for our phone conference at 08:00 AM (PT):
Tel: 866 500 6738 (US); 203 480 8000 (intl)
Participant code: 885535
pratap
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.mozilla.org/pipermail/es-discuss/attachments/20080625/4532bd09/attachment-0002.html
Following up on the "Strict Mode" discussion...
As I advocated on the call, I think that by selecting "strict mode" a developer is really choosing to restrict themselves to using a subset of the complete language. The future-proofing issues of this relate to the possibility that there might be multiple such subsets that a developer might need to choose among. Should there be multiple named "strict modes" to choose among, how should they be named, can "strictness" of a mode increase in future versions, etc?
I think some of the controversy could be eliminated if we simply eliminate the term "Strict Mode". Instead I propose we have a "Use Subset" directive and that we name specific subsets in a meaningful and generally positive manner. For example, since the motivation of most of the proposed restrictions in ES3.1 strict mode is to provide a safer subset language I suggest that we call this subset "safety" (or perhaps "safety1" or "safetyA" or "safety2008" implying that in the future different safe subsets might be defined and we don't want to give undo importance to this initial one).
So, the first line of a "strict mode" compilation unit would now look like" "use subset safety"
I would suggest that we actually define "use subset" such that a comma separated list of subsets is allowed. So, if somebody decided to define a subset that only contained the features of ES3 you might see something like this: "use subset safety,ES3"
Since subsets are sets of restrictions, listing multiple subsets means to take the union of the restrictions imposed by all of the listed subsets. So "use subset safty,ES3" means that this compilation unit may only use those features defined by ECMA 262-3 and which are not excluded by the "safety" subset. So, assuming that "safety" excludes use of the with statement, such a compilation unit could not include use of the with statement nor could it include any use of a feature that is new in ES3.1 or ES4.
Future versions of ECMAScript may add exclusions to a subset defined by an earlier version as long as the added exclusions only restrict capabilities that didn't exist in the earlier version. For example, ES 4 in supporting the ES3.1 "safety" subset but add to it any features that are added in ES 4 but which are considered to be unsafe.
A future version may not add exclusion to an pre-existing subset that restricts features that existed when the original subset was defined. For example if ES3.14 or ES5 decided that the for-in statement was unsafe, it could not add that restriction to the "safety" subset. However, it could define a new subset named perhaps "safety2010" that includes all the restrictions of the "safety" subset and in addition disallows use of the "for" statement.
If a compilation unit specifies a subset that is not known to the implementation that is processing it, that subset restriction is simply ignored. The code in the unit is still either valid or invalid on its own merit just like is the case when no subset had been specified.
Following up on the "Strict Mode" discussion... As I advocated on the call, I think that by selecting "strict mode" a developer is really choosing to restrict themselves to using a subset of the complete language. The future-proofing issues of this relate to the possibility that there might be multiple such subsets that a developer might need to choose among. Should there be multiple named "strict modes" to choose among, how should they be named, can "strictness" of a mode increase in future versions, etc? I think some of the controversy could be eliminated if we simply eliminate the term "Strict Mode". Instead I propose we have a "Use Subset" directive and that we name specific subsets in a meaningful and generally positive manner. For example, since the motivation of most of the proposed restrictions in ES3.1 strict mode is to provide a safer subset language I suggest that we call this subset "safety" (or perhaps "safety1" or "safetyA" or "safety2008" implying that in the future different safe subsets might be defined and we don't want to give undo importance to this initial one). So, the first line of a "strict mode" compilation unit would now look like" "use subset safety" I would suggest that we actually define "use subset" such that a comma separated list of subsets is allowed. So, if somebody decided to define a subset that only contained the features of ES3 you might see something like this: "use subset safety,ES3" Since subsets are sets of restrictions, listing multiple subsets means to take the union of the restrictions imposed by all of the listed subsets. So "use subset safty,ES3" means that this compilation unit may only use those features defined by ECMA 262-3 and which are not excluded by the "safety" subset. So, assuming that "safety" excludes use of the with statement, such a compilation unit could not include use of the with statement nor could it include any use of a feature that is new in ES3.1 or ES4. Future versions of ECMAScript may add exclusions to a subset defined by an earlier version as long as the added exclusions only restrict capabilities that didn't exist in the earlier version. For example, ES 4 in supporting the ES3.1 "safety" subset but add to it any features that are added in ES 4 but which are considered to be unsafe. A future version may not add exclusion to an pre-existing subset that restricts features that existed when the original subset was defined. For example if ES3.14 or ES5 decided that the for-in statement was unsafe, it could not add that restriction to the "safety" subset. However, it could define a new subset named perhaps "safety2010" that includes all the restrictions of the "safety" subset and in addition disallows use of the "for" statement. If a compilation unit specifies a subset that is not known to the implementation that is processing it, that subset restriction is simply ignored. The code in the unit is still either valid or invalid on its own merit just like is the case when no subset had been specified. _____________________________________________ From: Pratap Lakshman (VJ#SDK) Sent: Tuesday, June 24, 2008 11:43 AM To: Adam Peller; Sam Ruby; Mark S. Miller; Allen Wirfs-Brock; Pratap Lakshman (VJ#SDK) Subject: Brief Minutes [RE: ES3.1 WG phone conference 24 June 08:00 PT] Here are brief minutes from our call. Please take a look, and let me know if you want any changes by your EOD. I'll upload it to the wiki and send a copy to Patrick Charollais (ECMA) for posting on the ECMA site tomorrow night (Redmond time). Attendees Adam Peller (IBM) Sam Ruby (IBM) Mark Miller (Google) Allen Wirfs-Brock (Microsoft) Pratap Lakshman (Microsoft) Agenda On posting the latest draft to the wiki Getters/Setters Decimal Setting up a review based on Lars' feedback on the 11 June draft Minutes Would like to add a couple more items to agenda that we can get to if we have the time (1) inconsistence language like "as if by the expression ..." pervasive in ES3 (e.g. section 11.1.4 Array Initializer: "create a new array as if by the expression new Array()"; needs to be fixed in ES3.1 (2) ES3/ES4 based review. On posting the latest draft to the wiki The latest draft has been uploaded to the wiki. This the draft as of 24 June 2008 - it has all the edits related to the statics on Object, the introduction of the [[Extensible]] property, the revised notation for JSON, and placeholders for Decimal - by next week we should try to have a technically complete draft - the draft may still have some place holders but it should still be enough to get circulated within TC39 as an artifact that can be reviewed - by the time we meet in Oslo, each place holder must have a supplementary doc that can be discussed F2F. 'const' needs to be introduced into the grammar - what does it mean to be a reserved word? - you cannot use it as a variable name - note that we have introduced the ability to use is as the name of a property though - const has letrec like scoping - not subject to hoisting. reformed scoping needs to be introduced - but for doing that we may want to introduce the notion of a 'block activation' as an expository device 'abc'[i] is using the ES3 like algorithm convention - can we use the new convention that eliminates the need for writing gotos - pratapL to investigate - also, for all new functionality we need to worry about argument conditioning - for e.g. if an algorithm expects to be handed an object, make sure we call toObject on the argument that is passed in. Getters/Setters Surface syntax spec from Kris looked Ok - need to ensure it is using the Meta APIs - Allen to check surface syntax integration. Decimal Decimal changes are isolated and can be done without impacting content in the rest of the doc - Sam can use the latest draft from the wiki as a starting point. Strict mode 2 controversies coupled to each other - proposed ES4 wants to keep 'with' in their strict mode; ES3.1 wants to remove 'with' in its strict mode - need to be clear about the purpose of strict mode - the other is "what do we mean by subset?" - there was one formal definition from Lars, and Doug had called out a less formal notion - we may end up using Doug's less formal notion (especially if we cannot resolve the 'with' issue). Binary strict mode is naively limiting - each version may want to allow/limit specific features - strict mode allows users to say 'I am subscribing to this particular subset, and I am aware of all its limitations' - but if the subset is a moving target how do they do that? - sure, we can introduce a more elaborate mechanism - it would make it easier for us to not have to resolve arguments - but it would open up a larger combinatorial space of possibilities - the SunOS vs. MacOS problem; the former was highly customizable, however you invariably got it wrong; the latter was not customizable, but what you got was good enough to get the job done - more knobs need not mean better. An elaborate mechanism enables chaos at the composability boundary - but how does it matter if you are opaquely including a module? - Caja will use ES3.1 strict mode for all uncajoled code - OTOH an elaborate mechanism can enforce a subsetting profile - languages can now more directly enforce Caja like semantics. 'use strict 3.1' could be a potential syntax to turn on 3.1 level strict mode - with the possibility that the 3.1 may be replaced with a list - instead of tying it to the language version number, we can also just say something like 'use strict a' or 'use strict 1' - can also use Perl-style 'use strict no with' (where we mention the specific restriction we want to enable) - that seems a good idea too - in any case, named restriction sets for ES3.1 and proposed ES4 will be useful to discuss this proposal further. Meeting adjourned. pratap _____________________________________________ From: Pratap Lakshman (VJ#SDK) Sent: Tuesday, June 24, 2008 6:46 PM To: Pratap Lakshman (VJ#SDK); crock at yahoo-inc.com; Mark S. Miller; Kris Zyp; Mike Cowlishaw; Adam Peller; Sam Ruby; Lars Hansen; ggaren at apple.com; Allen Wirfs-Brock Subject: ES3.1 WG phone conference 24 June 08:00 PT Agenda: (1) On posting the latest draft to the wiki (2) Getters/Setters (3) Decimal (4) setting up a review based on Lars' feedback on the 11 June draft Let me know if you want to add any items to the agenda Here is the dial-in information for our phone conference at 08:00 AM (PT): Tel: 866 500 6738 (US); 203 480 8000 (intl) Participant code: 885535 pratap -------------- next part -------------- An HTML attachment was scrubbed... URL: http://mail.mozilla.org/pipermail/es-discuss/attachments/20080625/4532bd09/attachment-0002.html