sgf-parsing: Add individual tests of escaping/whitespace behaviour

[petertseng]

11c36323-93fc-495d-bb23-c88ee5844b8c crams too much into one test case.
Split it up into multiple ones.

In addition, 11c36323-93fc-495d-bb23-c88ee5844b8c has behaviour that
violates the specification. The violation is that \t and \n (written
as "\\t" and "\\n" in the JSON string, respectively), do not hold
any sort of special significance in SGF, according to the specification:
https://www.red-bean.com/sgf/sgf4.html

Reimplement in 08e4b8ba-bb07-4431-a3d9-b1f4cdea6dab.

The reimplemented case is mostly as it was when the exercise was
originally implemented in
exercism/exercism@7a5075b
and the reimplemented case is in accordance with the specification.

Note that the original case also got it wrong in that newlines should
remain newlines; this is corrected in
08e4b8ba-bb07-4431-a3d9-b1f4cdea6dab.

[petertseng]

Diff to https://github.com/petertseng/exercism-problem-specifications/blob/verify/exercises/sgf-parsing/verify.rb:

diff --git a/exercises/sgf-parsing/verify.rb b/exercises/sgf-parsing/verify.rb
index 45d41aca..f85c5044 100644
--- a/exercises/sgf-parsing/verify.rb
+++ b/exercises/sgf-parsing/verify.rb
@@ -97,12 +97,8 @@ class SGF
       if c == ?\\ && !escape
         escape = true
       else
-        val << ?\\ if escape && c == ?n
-        if escape && c == ?t
-          val << ' '
-        else
-          val << c
-        end
+        to_insert = c == "\n" && escape ? '' : c == "\t" ? ' ' : c
+        val << to_insert
         escape = false
       end
       pos += 1
@@ -114,6 +110,8 @@ end
 
 json = JSON.parse(File.read(File.join(__dir__, 'canonical-data.json')))
 
+json['cases'].reject! { |v| v['uuid'] == '11c36323-93fc-495d-bb23-c88ee5844b8c' }
+
 verify(json['cases'], property: 'parse') { |i, c|
   pos, node = SGF.new(i['encoded']).node(0, paren_required: true)
   raise 'Incomplete parse' unless pos == i['encoded'].size

[petertseng]

you may ask. why the foo and bar? Just to detect the case where a parser silently breaks in a way that finishes the current property value but halts all further parsing. Having additional material to parse after that, and verifying that that was parsed, avoids that case.

[petertseng]

To be very clear: In this case, and in all other cases in this PR as well, it's important to keep in mind the difference between the string as it appears in JSON and the string as it is presented to the SGF parser.

The string "(;A[\\]b\nc\\\nd\t\te\\\\ \\\n\\]])" in JSON means that this string is presented to the SGF parser:

(;A[\]b
c\
d		e\\ \
\]])

Everything between the [ and the ] is the property value, and it contains these characters:

• \], escaped closing bracket: insert closing bracket in property value
• b: insert as-is.
• unescaped newline: insert as-is.
• c: insert as-is.
• escaped newline: insert nothing.
• d: insert as-is
• two tabs: insert two spaces
• e: insert as-is
• \\, escaped backslash: insert backslash
• space: insert as-is.
• escaped newline: insert nothing
• \], escaped closing bracket: insert closing bracket

The property value is thus:

]b
cd  e\ ]

which is written in JSON as it is on line 347.

[ErikSchierboom]

Would it perhaps be worth it to add this comment to the test case? It looks to be very useful in getting people to understand what they're working with.

[petertseng]

Not a bad idea. Added.

[ErikSchierboom]

Would it perhaps be worth it to add this comment to the test case? It looks to be very useful in getting people to understand what they're working with.

[petertseng]

this is verbose, but I am thinking it may be a good idea to avoid backslashes in the comments field if avoiding backslashes will help prevent further confusion.

[ErikSchierboom]

Absolutely.

[ErikSchierboom]

CC @exercism/reviewers

[petertseng]

Please understand that it was necessary for me to rebase this because #1917 reformatted this exercise. My intent is that the only thing changed is whitespace. Believing this to be true (observe how git diff -w 03d447f 427d136 exercises/sgf-parsing is empty), I believe this PR remains reviewable.