Crush spaces?

[timbray]

There are lots of places in the grammar where optional space is allowed, e.g.

slice-selector =  [start S] ":" S [end S] [":" [S step ]]

Would it be a good idea to crush them all out? We wouldn't lose any backward compatibility, we'd remove a lot of optional operations for implementers, and we'd get more compatible expressions. We might lose readability, but not much.

[glyn]

Crushing out all spaces would simplify testing and I agree it wouldn't impact the readability of simple paths much.

On the other hand, implementers using parser generators won't have to write code to handle whitespace and those using hand-crafter parsers or parser combinators can easily code up a whitespace handling routine which can be used where necessary.

When it comes to complex paths, I think the readability impact of crushing out all whitespace may be significant. E.g. compare

$.a[?@ < 2 || @.b == "k", ?@.c > $.d]

to

$.a[?@<2||@.b=="k",?@.c>$.d]

That said, I think it would be unusual to use whitespace in slices, e.g. [1: 10: 2] or [1 : 10 : 2] don't seem more readable than [1:10:2], so there may be a case for crushing whitespace out of specific productions such as slice-selector.

[timbray]

Not passionate about this one, but I do like the idea of removing them from slices.

[glyn]

In addition to removing whitespace from slices, we might also consider removing all whitespace, except that following the comma selector separator, from child (and hence descendant) segments, i.e. replacing:

child-longhand            = "[" S selector 1*(S "," S selector) S "]"

with

child-longhand            = "[" selector 1*("," S selector) "]"

[glyn]

I did think of a readability concern for slices. Suppose there is some kind of list of paths and it was more readable to line up the slice components, e.g.:

$[  0 :   5 : 2]
$[ 10 :  15 : 2]
$[ 20 :  25 : 2]
$[ 30 :  35 : 2]
...
$[100 : 105 : 2]

Compare this to the crushed alternative:

$[0:5:2]
$[10:15:2]
$[20:25:2]
$[30:35:2]
...
$[100:105:2]

I'm not sure that's a deal-breaker for "crushed" slices, but it's worth thinking about. (It is also relevant to child segment whitespacing because of the whitespace after [.)

[cabo]

Blank space also includes newlines. Some queries may be long enough that you want to break lines; not being able to do this on natural breaks such as : can be counterintuitive.

[cabo]

After deciding this issue, continue in #344

[cabo]

2023-01-10 Interim: Do not do this. Leave S in place.