Documentation for out-of-bound accesses with single index vs slices

[Poshi]

According to the documentation:

out-of-bounds index accesses are errors, but out-of-bounds slice accesses result in trimming the indices, resulting in a short string or even the empty string

This is somewhat ambiguous: if out-of-bound index accesses are errors, it is surprising the an slice access does not trigger the same error. Would not make sense to return the same as in the slice access: an absent value?

Also, the examples below show:

mlr -n put '
  end {
    x = "abcde";
    print x[1:2];
    print x[1:6];
    print x[10:20];
  }
'
ab

If we take for good that out-of-bound slice accesses can result in short strings, should not the second access (x[1:6]) result in the short string "abcde"? (It is a short string because the slice should return a 6 characters string but the result only contains 5).

[johnkerl]

@Poshi you're right, the documentation is not clear.

I very intentionally imitated Python's behavior here, but didn't document that fact ...

[johnkerl]

OK I didn't entirely even mimic Python behavior as intended. :(

Keeping in mind Python indices are 0-up, with lower-inclusive-upper-exclusive, while Miller indices are 1-up, with lower-inclusive-upper-inclusive:

$ python
>>> x = "abcde"

>>> x[0:2]
'ab'

>>> x[0:5]
'abcde'

>>> x[0:6]
'abcde'

>>> x[9:20]
''

$ mlr repl
[mlr] x = "abcde"

[mlr] x = "abcde"

[mlr]  x[1:2]
"ab"

[mlr]  x[1:5]
"abcde"

[mlr]  x[1:6]
""
[mlr]  x[10:20]
""

So for the first thing

Would not make sense to return the same as in the slice access: an absent value?

-- I stand by my intention to mimic Python here.

For the second thing

If we take for good that out-of-bound slice accesses can result in short strings, should not the second access (x[1:6]) result in the short string "abcde"?

-- you are quite right, and I'm embarrassed to realize I missed this. :(

[johnkerl]

For arrays, I did what I intended:

[mlr] x = [1,2,3,4,5]

[mlr] x[1:2]
[1, 2]

[mlr] x[1:5]
[1, 2, 3, 4, 5]

[mlr] x[1:6]
[1, 2, 3, 4, 5]

[mlr] x[10:20]
[]

[johnkerl]

@Poshi please let me know what you think of #960.

[Poshi]

@Poshi please let me know what you think of #960.

Just checked the three commits. Looks good! The new code seems OK. I'm not sure if some checks can be reduced, but they look fine. (I've also been lost a couple minutes when the index went down to zero until I realized the difference between "index" and "Zindex").
And regarding que documentation, adding quotes to the examples makes the empty strings stand out! Good!