[Epic] Holes in our evaluation model

[rebolbot]

Submitted by: BrianH

For Rebol 3 we have done a lot to clean up our semantics, and build a much cleaner, more uniform, more intentionally designed evaluation model. Between the rebalancing of error triggering, or of none propagation, or of type conversions, we're well on our way to making a much more solid base for us to build on going forward.

However, we still have some holes in the new model that need fixing or filling in. This is a list of the minor changes, fixes or additions that we should do before 3.0, at least ones that aren't referenced in the other epics (though a few of those too). Limited to semantic model or API changes - we can fix bugs while we're at it, but semantic model changes need to go in before the 3.0 release sets an expectation of continuing compatibility.

Related tickets in the first comment, to make linking easier. Status will be updated there.

^{CC - Data [ Version: r3 master Type: Issue Platform: All Category: Evaluation Reproduce: Always Fixed-in:none ]}

[rebolbot]

Submitted by: BrianH

Missing functions:

• TO-VALUE TO-VALUE function to convert unset to none #2003 - The last bit of the "unset is only erroneous when you need it to be" principle.
• IDENTITY Identity function #1987 - The noop value filter function, for when you have to provide a filter but don't need to do anything. Evaluating version of QUOTE.
• WRAP WRAP function #2119 - Missing part of the new locals model (like Drop current FUNCTION, make it a FUNCT synonym instead #1973): The locals-gathering equivalent of USE.
• ANY-OF and ALL-OF ANY-OF, ALL-OF functions #637 - *EACH conditional functions, data-processing ANY and ALL.
• DO-COMPARE DO-COMPARE function #2110 - Uniform record-oriented function behavior for functions written in Rebol, to promote the uniformity.
• EXPR EXPR mathematical expression evaluator obeying ordinary precedence #2120 - We need an answer to shut down complaints about our precedence rules (a common issue with new users). Fast mezzanine is fine (except No [throw] attribute: Mezzanine control functions capture RETURN and EXIT #539), unless we use it a lot.

Breaks in the model:

• DO/next non-block DO/next doesn't set var when /next doesn't apply #1951 - expected modification doesn't happen, doesn't trigger error either.
• REDUCE/into non-block REDUCE/into of non-block doesn't insert into target #2061 - breaks the /into model, making code that uses it not work.
• COMPOSE/into non-block COMPOSE/into of non-block doesn't insert into target #2062 - breaks the /into model, making code that uses it not work.
• SPLIT misnames /into Rename SPLIT's /INTO refinement #2051 - breaks the /into model, and it's not a good name for this either.
• TRANSCODE/error triggers construction syntax errors TRANSCODE/error of an invalid serialized constructor triggers the error it should insert #1857 - This breaks an invariant that LOAD depends on.
• REDUCE/no-set sets set-words REDUCE/NO-SET working also with set-path! #2122 - no-set means no setting.
• No error for REDUCE/only [unset-word] REDUCE/ONLY emits UNSET for undefined word #1771 - unset-tolerant code should require get-words or GET/any, so you can tell.
• Crash on REDUCE REDUCE [:self] REDUCE REDUCE [:self] crash #1756 - bigger issue: get-word evaluation of 'self keyword doesn't convert from frame to object, like word evaluation does.

Filling in the coverage gaps (some not as critical):

• Set operations on map! Can we have set operations on map! please? #1984 - Increase the coverage of the set functions like UNION.
• FIRST+, ++ and -- for ports FIRST+ for ports #559 - Let ports that support series-like actions act like series. Only if it's efficient enough.

More will links be added later.

#637, #1951, #2003 and #2119 have PRs.

[Oldes]

#2061 fixed in: Oldes/Rebol3@6d14eea

[Oldes]

#2062 fixed in Oldes/Rebol3@27278b6

[Oldes]

#1756 fixed in zsx/r3@b275787 and Oldes/Rebol3@2abd4c2

[Oldes]

#2051 modified in Oldes/Rebol3@3385185 (split/pieces instead of split/into)

[Oldes]

#2122 fixed in Oldes/Rebol3@690dffa