Call escape_raw_string to make parsing more predictable

[vtjnash]

The advantage here is that you don't escape it twice, so that styled"\\" doesn't need double the number of \\ each time to get one more \ at the end (the downside is that less common forms such as @styled_str "\\" still apply the unescape_string call twice and we cannot catch that)

julia> styled""
""

julia> styled"\\"
"\\"

julia> styled"\\\\"
"\\"

julia> styled"\\\\\\"
"\\\\"

julia> styled"\\\\\\\\"
"\\\\"

[tecosaur]

Thanks! I wonder, why isn't this fixed by just adding \ to the tuple of chars supplied to unescape_string? If I make that change locally I see:

julia> styled"\\"
"\\"

julia> styled"\\\\"
"\\\\"

julia> styled"\\\\\\"
"\\\\\\"

julia> styled"\\\\\\\\"
"\\\\\\\\"

julia> styled"\a\b\e\u200b"
"\a\b\e\u200b"

[vtjnash]

That applies it always, rather than just at the end of a string

[tecosaur]

Ah, so the bug you're mentioning in the original commit only affects a terminal /s?

Also, perhaps I'm missing the obvious - would you mind providing an example of how adding / to the tuple of unescape chars isn't a good enough fix?

These are the changes I'm considering:

diff --git a/src/stylemacro.jl b/src/stylemacro.jl
index b030569..9abccb4 100644
--- a/src/stylemacro.jl
+++ b/src/stylemacro.jl
@@ -107,7 +107,7 @@ macro styled_str(raw_content::String)
     end =#
 
     # Instead we'll just use a `NamedTuple`
-    state = let content = unescape_string(raw_content, ('{', '}', '$', '\n', '\r'))
+    state = let content = unescape_string(raw_content, ('\\', '{', '}', '$', '\n', '\r'))
          (; content, bytes = Vector{UInt8}(content),
          s = Iterators.Stateful(pairs(content)),
          parts = Any[],
@@ -635,10 +635,10 @@ macro styled_str(raw_content::String)
     function run_state_machine!(state)
         # Run the state machine
         for (i, char) in state.s
-            if char == '\\'
-                state.escape[] = true
-            elseif state.escape[]
+            if state.escape[]
                 escaped!(state, i, char)
+            elseif char == '\\'
+                state.escape[] = true
             elseif char == '$'
                 interpolated!(state, i, char)
             elseif char == '{'

[vtjnash]

These should return the same styled string

julia> styled"\\\\\\\\" * "a"
"\\\\a"

julia> styled"\\\\\\\\a"
"\\\\\\\\a"

[vtjnash]

You can attempt to combine escape_string and unescape_raw_string, but it can make the implementation much more confusing and likely to have bugs since they cannot be tested separately. It doesn't seem like that change you are considering implements any part of unescape_raw_string, but rather just reimplements the unescape_string's handling for \\ needlessly

[tecosaur]

Right. Thanks for clarifying. One more question I'd like to ask is regarding this code comment in the PR

assuming the input was entered with single styled"" markers so this transform is unambiguously reversible and not as @styled_str "." or styled"""."""

I'm not too concerned about the @styled_str "..." case, but what's the potential transformation ambiguity with a triple-quote string macro styled"""..."""?

p.s. if there are any relevant tests you think it would be good to have, more tests would be nice, though not required.

[vtjnash]

IIRC, styled"""...""" has an ambiguity with styled"""... \\\" ...""" vs. styled"""... \\" ...""" since the final \ in that set is optional. But, actually, I think unescape_string would give the same result for either?

[vtjnash]

Bump?

[tecosaur]

Ah yes. Mind if I force push to your branch with some minor changes (moving a comment, adding some a test or two)?

[vtjnash]

Go for it

[tecosaur]

Done. The tests I just added failed before the change, but now pass.

I also expanded on the commit message a bit.