[ 
https://issues.apache.org/jira/browse/JEXL-476?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Henri Biestro updated JEXL-476:
-------------------------------
    Description: 
  h3. Problem
  JEXL does not separate two different questions consistently:
  * *null*: the object an access applies to has no value;
  * *undefined*: a variable, a property or a method does not exist.

  Four mechanisms can excuse them ({{?.}}, the engine {{safe}} option, {{??}}, 
and {{?:}} / ternary), and each covers a different subset. On a strict engine, 
with a String that has no {{nothing}} member:

  ||Expression||safe(false)||safe(true)||
  |{{'abc'?.nothing}}|null|null|
  |{{'abc'?.nothing()}}|null|null|
  |{{'abc'.nothing?.y}}|error|error|
  |{{'abc'.nothing()?.y}}|null|null|
  |{{'abc'.nothing}}|error|error|
  |{{'abc'.nothing()}}|error|null|
  |{{nothing(1)}}|error|error|
  |{{nothing(1).y}}|error|null|
  |{{'abc'.nothing ?? 1}}|1|1|
  |{{'abc'.nothing() ?? 1}}|error|error|

  A missing method is excused by a following {{?.}} and by the {{safe}} option 
but not by {{??}}; a missing property is the exact opposite. A property may 
hold a function, so the two should not differ.

  h3. Proposal
  Safe navigation is about null only; undefined has its own operators.

  ||Condition||{{?.}} / {{?[}}||{{safe}} option||{{??}}||{{?:}} and 
ternary||strict(false)||
  |null value|null|null|right operand|falsy|null|
  |missing property, method or function of a non-null object|error|error|right 
operand|falsy|null|

  * {{x?.y}}, {{x?.y()}} and {{x?[y]}} evaluate as null when {{x}} is null. 
They no longer excuse a missing property or method on a non-null object, 
whether the safe access is the failing one or the one that 
  follows.
  * The {{safe}} option does the same for every access, without the explicit 
operator.
  * {{??}}, {{?:}} and the ternary tolerate a missing method or function in 
their first operand, as they already do for an undefined variable or a missing 
property. {{??}} tests for null or undefined only; {{?:}} 
  and the ternary coerce to boolean, so {{0 ?: 1}} is 1 where {{0 ?? 1}} is 0.

An undefined variable at the root of a navigation is out of scope: {{u?.y}}, 
and {{u.y}} with the {{safe}} option, keep evaluating as null, exactly as a 
null variable does. A host often cannot tell an absent context key from a null 
one.

  h3. Compatibility
  This changes released behaviour (JEXL-290):
  * {{x?.missing()}} and {{x?.missing}} on a non-null {{x}} become errors on a 
strict engine;
  * {{x.missing()?.y}} becomes an error;
  * with {{safe(true)}}, a missing method becomes an error.

  Scripts that use {{?.}} for duck typing should use {{x.maybe() ?? fallback}} 
instead, which this ticket makes work for methods.

  h3. Related
  JEXL-474 already applies the null half of the rule: only the access applied 
to null decides, so {{m?.get('z').y}} is an error when {{m.get('z')}} is null, 
and the {{safe}} option covers array accesses.


  was:

  h3. Problem
  JEXL does not separate two different questions consistently:
  * *null*: the object an access applies to has no value;
  * *undefined*: a variable, a property or a method does not exist.

  Four mechanisms can excuse them ({{?.}}, the engine {{safe}} option, {{??}}, 
and {{?:}} / ternary), and each covers a different subset. On a strict engine, 
with a String that has no {{nothing}} member:

  ||Expression||safe(false)||safe(true)||
  |{{'abc'?.nothing}}|null|null|
  |{{'abc'?.nothing()}}|null|null|
  |{{'abc'.nothing?.y}}|error|error|
  |{{'abc'.nothing()?.y}}|null|null|
  |{{'abc'.nothing}}|error|error|
  |{{'abc'.nothing()}}|error|null|
  |{{nothing(1)}}|error|error|
  |{{nothing(1).y}}|error|null|
  |{{'abc'.nothing ?? 1}}|1|1|
  |{{'abc'.nothing() ?? 1}}|error|error|

  A missing method is excused by a following {{?.}} and by the {{safe}} option 
but not by {{??}}; a missing property is the exact opposite. A property may 
hold a function, so the two should not differ.

  h3. Proposal
  Safe navigation is about null only; undefined has its own operators.

  ||Condition||{{?.}} / {{?[}}||{{safe}} option||{{??}}||{{?:}} and 
ternary||strict(false)||
  |null value|null|null|right operand|falsy|null|
  |undefined variable, property, method or function|error|error|right 
operand|falsy|null|

  * {{x?.y}}, {{x?.y()}} and {{x?[y]}} evaluate as null when {{x}} is null. 
They no longer excuse a missing property or method on a non-null object, 
whether the safe access is the failing one or the one that 
  follows.
  * The {{safe}} option does the same for every access, without the explicit 
operator.
  * {{??}}, {{?:}} and the ternary tolerate a missing method or function in 
their first operand, as they already do for an undefined variable or a missing 
property. {{??}} tests for null or undefined only; {{?:}} 
  and the ternary coerce to boolean, so {{0 ?: 1}} is 1 where {{0 ?? 1}} is 0.

  h3. Compatibility
  This changes released behaviour (JEXL-290):
  * {{x?.missing()}} and {{x?.missing}} on a non-null {{x}} become errors on a 
strict engine;
  * {{x.missing()?.y}} becomes an error;
  * with {{safe(true)}}, a missing method becomes an error.

  Scripts that use {{?.}} for duck typing should use {{x.maybe() ?? fallback}} 
instead, which this ticket makes work for methods.

  h3. Related
  JEXL-474 already applies the null half of the rule: only the access applied 
to null decides, so {{m?.get('z').y}} is an error when {{m.get('z')}} is null, 
and the {{safe}} option covers array accesses.



> Safe navigation should only shortcut null, not undefined properties or methods
> ------------------------------------------------------------------------------
>
>                 Key: JEXL-476
>                 URL: https://issues.apache.org/jira/browse/JEXL-476
>             Project: Commons JEXL
>          Issue Type: Improvement
>    Affects Versions: 3.7.1
>            Reporter: Henri Biestro
>            Assignee: Henri Biestro
>            Priority: Major
>             Fix For: 3.7.2
>
>
>   h3. Problem
>   JEXL does not separate two different questions consistently:
>   * *null*: the object an access applies to has no value;
>   * *undefined*: a variable, a property or a method does not exist.
>   Four mechanisms can excuse them ({{?.}}, the engine {{safe}} option, 
> {{??}}, and {{?:}} / ternary), and each covers a different subset. On a 
> strict engine, with a String that has no {{nothing}} member:
>   ||Expression||safe(false)||safe(true)||
>   |{{'abc'?.nothing}}|null|null|
>   |{{'abc'?.nothing()}}|null|null|
>   |{{'abc'.nothing?.y}}|error|error|
>   |{{'abc'.nothing()?.y}}|null|null|
>   |{{'abc'.nothing}}|error|error|
>   |{{'abc'.nothing()}}|error|null|
>   |{{nothing(1)}}|error|error|
>   |{{nothing(1).y}}|error|null|
>   |{{'abc'.nothing ?? 1}}|1|1|
>   |{{'abc'.nothing() ?? 1}}|error|error|
>   A missing method is excused by a following {{?.}} and by the {{safe}} 
> option but not by {{??}}; a missing property is the exact opposite. A 
> property may hold a function, so the two should not differ.
>   h3. Proposal
>   Safe navigation is about null only; undefined has its own operators.
>   ||Condition||{{?.}} / {{?[}}||{{safe}} option||{{??}}||{{?:}} and 
> ternary||strict(false)||
>   |null value|null|null|right operand|falsy|null|
>   |missing property, method or function of a non-null 
> object|error|error|right operand|falsy|null|
>   * {{x?.y}}, {{x?.y()}} and {{x?[y]}} evaluate as null when {{x}} is null. 
> They no longer excuse a missing property or method on a non-null object, 
> whether the safe access is the failing one or the one that 
>   follows.
>   * The {{safe}} option does the same for every access, without the explicit 
> operator.
>   * {{??}}, {{?:}} and the ternary tolerate a missing method or function in 
> their first operand, as they already do for an undefined variable or a 
> missing property. {{??}} tests for null or undefined only; {{?:}} 
>   and the ternary coerce to boolean, so {{0 ?: 1}} is 1 where {{0 ?? 1}} is 0.
> An undefined variable at the root of a navigation is out of scope: {{u?.y}}, 
> and {{u.y}} with the {{safe}} option, keep evaluating as null, exactly as a 
> null variable does. A host often cannot tell an absent context key from a 
> null one.
>   h3. Compatibility
>   This changes released behaviour (JEXL-290):
>   * {{x?.missing()}} and {{x?.missing}} on a non-null {{x}} become errors on 
> a strict engine;
>   * {{x.missing()?.y}} becomes an error;
>   * with {{safe(true)}}, a missing method becomes an error.
>   Scripts that use {{?.}} for duck typing should use {{x.maybe() ?? 
> fallback}} instead, which this ticket makes work for methods.
>   h3. Related
>   JEXL-474 already applies the null half of the rule: only the access applied 
> to null decides, so {{m?.get('z').y}} is an error when {{m.get('z')}} is 
> null, and the {{safe}} option covers array accesses.



--
This message was sent by Atlassian Jira
(v10.3.26#10030026)

Reply via email to