WISH: /CASE and /TAIL refinements for REPLACE

[9214]

Original issue: red/red#2845

Title.

alphabet: {AaBbCc}
replace alphabet #"a" 1
probe alphabet

Results in

"1aBbCc"

And as of now it's impossible to replace the lowercase a, because /case refinement is absent. For the sake of compatability, I propose to implement /tail too, thus making replace identical to R2 version:

REFINEMENTS:
     /all -- Replace all occurrences
     /case -- Case-sensitive replacement
     /tail -- Return target after the last replacement position

[Palaing]

Rebol's version of replace (source replace) seems to work fine in Red. It has /case, /tail and even allows a function as third argument.

[endo64]

@Palaing In Red's version there is also a deep refinement which is very useful. Red's version also supports parse rules. So Rebol's version is a bit limited.

I've added case refinement and will do PR,

[endo64]

Here is my PR, @9214 @greggirwin can you please check? Note that didn't add tail yet. red/red#3573

[endo64]

This can be closed now as /case added to replace. We can create another one for /tail.

[hiiamboris]

Don't wanna make a separate wish for this...
I'd like to also have /only refinement for replace on blocks. It would then treat the replacement block as a single value.

My example:

#macro LEN3?: func [waht [path! word! block!]] [
	replace/deep [((length? waht) / 3)] 'waht reduce [waht]
]

Here I have to use reduce [waht] to do what could have been done by replace/deep/only.

[greggirwin]

Adding /only support to replace makes a lot of sense. We don't have a lot of tests for replace right now, so step 1, IMO, is to make sure we have good coverage for the current behavior. Then it seems like it could use a review, to see if we can simplify things a bit, and add comments. It's got a lot of subtlety in there already, and this will add more. system/words/all is used enough that a local shortcut also makes sense to me.

[hiiamboris]

Another point is deep replacement vs using a parse rule: ☝️ June 7, 2019 9:33 PM

[greggirwin]

Looking at the original ticket, /tail is a tough call. I don't recall ever using it, but there is no good way to get that info otherwise. Only needed for advanced processing, but then is it useful with /deep or /all? What are the top use cases or examples?

[hiiamboris]

R2:

>> replace/all/tail "acbcd" "c" "-"
== "d"

with /deep it should probably error out, and if one day somebody finds a meaning for that - we add it.

[greggirwin]

I understand how it works, but how/where has anyone used it in the real world? In a search of R2 code here, I found zero instances of it. And that includes all Rebol SDKs, IOS worlds, and a lot of other code.

[hiiamboris]

And what about replace without /all refinement - what is the use case for it? Does just replace make more sense than replace/tail? The latter at least I can imagine being used in a loop, even intermittent with find/tail maybe. And will it hurt to return /tail by default when /all is not used?

[greggirwin]

The most common use case for replace is replace/all, but reversing the logic and having a /first refinement (or other name) would a) be a huge Rebol incompatibility, and b) not fit quite as nicely with other uses of single vs /all behavior. Tough decision, even then, because I've tripped over that myself more than once.

I also think the vast majority of replace calls don't chain the return value. Only a few cases turned up where "replace/all replace/all" was found. The common pattern is strictly imperative. Call it, then call it again.