give more information in StringIndexError

[KristofferC]

Before

julia> str = "abcdefghκijklmno"
"abcdefghκijklmno"

julia> str[10]
ERROR: StringIndexError("abcdefghκijklmno", 10)
Stacktrace:

Now

julia> str = "abcdefghκijklmno"
"abcdefghκijklmno"

julia> str[10]
ERROR: StringIndexError: invalid index [10], valid previous index [9] ('κ'), valid next index [11] ('i')

[KristofferC]

Bikeshed over valid previous index [1] ('α')?

Perhaps valid previous index [1]=='α'?

[mbauman]

StringIndexError is exported and may be (mis)used by user code — maybe defensively check for exc.index < firstindex(s) && return invoke(showerror, Tuple{IO, Any}, io, exc) before doing something that may throw?

[StefanKarpinski]

Bikeshed over valid previous index [1] ('α')?

I like that version fine, it's certainly clearer to me than [1]=='α'.

[StefanKarpinski]

Oh but "previous valid index" reads more naturally than "valid previous index". There's some rule about how adjectives get sorted in English and "valid previous index" violates them (in some way that I only know intuitively, I have no idea what the rules actually are, just that they exist).

[StefanKarpinski]

Same for "valid next index" which would be "next valid index" instead.

[mbauman]

We use pairs to connect indices to values in some cases; it could be nearby valid indices 9=>'κ', 11=>'i'? Of course the irregular plural is annoying at the end of the string, but the previous and next aren't doing much heavy lifting since I can see that 9<10<11.

[KristofferC]

Oh but "previous valid index" reads more naturally than "valid previous index".

Yeah, I thought about this too but decided on this to sort of emphasise the "valid nextind" and "valid prevind" connection but I can change the order if it is too weird.

[KristofferC]

Did some tweaks based on comments