[Epic] The Great Indexing Compromise

[rebolbot]

Submitted by: abolka

Rebol 2, Rebol 3, and Red currently all use slightly differing indexing models -- the stuff that underlies how path-based indexing, pick, poke, index? work.

Under much sweat and tears, this issue was intensely and extensively debated on the REBOL4 AltME world during Nov 2012. Without trying to summarise or even repeat the argument here, none of the two prominent positions could reach convincing consensus. But behold: many participants sensed eventually, that a compromise was reached.

This is a meta-issue to summarise and track the steps toward that compromise.

The corner stones of this compromise:

1. Non-contiguous ordinal indexing ("1-based") with zero hole for paths, pick, poke, index? ("like R2").
2. Contiguous offset indexing ("0-based") with pickz, pokez, indexz? (to be newly introduced). Details in PICKZ and POKEZ functions for 0-based indexing. #613.
3. "Backwards" ordinal accessors: first-back, second-back, third-back. Details in Ordinal functions in the other direction #1983.
4. Errors upon falling into the zero hole: so pick [] 0 would cause an error (and not silently return NONE as in R2).

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

[rebolbot]

Submitted by: BrianH

Note: The bounds checking changes introduced in Rebol 3 (first "" == none rather than triggering an error) are pointedly not part of this compromise. If we need to tweak those, that's a separate debate that hasn't happened yet (though Ladislav and I discussed some ideas a while ago, to no conclusion yet). This is just about the overall indexing model(s).

[ladislav]

• I think that the current state already is a "compromise" in the sense that everybody tolerates it to some extent, although there are simple cases that are obvious bugs
• I think that we need a deeper analysis to make qualified decisions

[hostilefork]

Implemented in Ren-C, without significant concern over compatibility. The body of affected code that is in use which can't be patched is too small to worry over, and the goal of compatibility is generally to match Rebol2 and Red semantics vs. the (non-existent) R3-Alpha codebases that don't already use Ren-C in 2018.