Have `search` throw an exception instead of returning zero/range

[Hydrotoast]

Hi, I have some thoughts on the edge case return values for the current search API.

Currently, search describes a failure to match as follows:

search(string, "substring") = start:end such that string[start:end] == "substring", or 0:-1 if unmatched.

search(string, 'c') = index such that string[index] == 'c', or 0 if unmatched.

This means that users of this API must handle the case of unmatching with some non-intuitive check. For example:

str = "haystack"
result = search(str, "needle")
if result == 0:-1 # or result[1] == 0 or first(result) == 0
    # handle case where no result was found
else
    # handle case where result was found
end

It turns out that first(result) == 0 is the condition that handles the three cases for when the object to search for (the needle) is either a string, a character, or a vector of characters since the API has two different return types depending on the the needle type.

The proposal here is to instead throw some type of exception e.g. UnmatchedException.

[nalimilan]

I think raising an exception is not possible because it would kill performance. Note that you can just check for isempty(result). If that's not clear in the manual and in the docstrings, they should be improved.

Regarding the search variant returning a single index, there have been discussions to decide whether they should return a Nullable instead of 0 as currently. See #15755 and the first point of this section of the Find & Search Julep).

[StefanKarpinski]

I think raising an exception is not possible because it would kill performance.

This is also a matter of language philosophy – we don't use exceptions to indicate unexceptional situations.

[StefanKarpinski]

There are also situations where search* returning an empty range of indices is extremely convenient. For example, inserting into sorted vectors: http://stackoverflow.com/questions/25678112/insert-item-into-a-sorted-list-with-julia-with-and-without-duplicates/25688266.