#bobzhang/difflib

• ports of CPython's Lib/test/test_difflib.py and the module's doctests;
• a byte-for-byte comparison with CPython's test_difflib_expect.html;
• a differential corpus of 1456 randomized cases generated by running the pinned CPython source (tools/gen_corpus.py). It covers opcodes, grouped opcodes, ratios, junk and autojunk handling, stateful set_seq1/set_seq2 call sequences, ndiff, restore, unified/context/colored diffs, diff_bytes, HTML tables and files, close matches and error messages. Inputs include tabs, Unicode whitespace, astral characters, lone surrogates and literal NUL/SOH characters (which collide with HtmlDiff's internal markers, exactly as in Python).

#Python → MoonBit

#Examples

#SequenceMatcher

///|
test "sequence matcher" {
  let s = @difflib.SequenceMatcher::new(
    a="qabxcd".to_array(),
    b="abycdf".to_array(),
  )
  inspect(
    s
    .get_opcodes()
    .map(op => "\{op.tag} a[\{op.i1}:\{op.i2}] b[\{op.j1}:\{op.j2}]")
    .join("\n"),
    content=(
      #|delete a[0:1] b[0:0]
      #|equal a[1:3] b[0:2]
      #|replace a[3:4] b[2:3]
      #|equal a[4:6] b[3:5]
      #|insert a[6:6] b[5:6]
    ),
  )
  inspect(s.ratio(), content="0.6666666666666666")
}

#Unified diff

///|
test "unified diff" {
  let before = ["one\n", "two\n", "three\n", "four\n"]
  let after = ["zero\n", "one\n", "tree\n", "four\n"]
  inspect(
    @difflib.unified_diff(
      before,
      after,
      fromfile="before.txt",
      tofile="after.txt",
    ).join(""),
    content=(
      #|--- before.txt
      #|+++ after.txt
      #|@@ -1,4 +1,4 @@
      #|+zero
      #| one
      #|-two
      #|-three
      #|+tree
      #| four
      #|
    ),
  )
}

#ndiff and restore

///|
test "ndiff" {
  let a = ["one\n", "two\n", "three\n"]
  let b = ["ore\n", "tree\n", "emu\n"]
  let delta = @difflib.ndiff(a, b).to_array()
  inspect(
    delta.join(""),
    content=(
      #|- one
      #|?  ^
      #|+ ore
      #|?  ^
      #|- two
      #|- three
      #|?  -
      #|+ tree
      #|+ emu
      #|
    ),
  )
  assert_eq(@difflib.restore(delta.iter(), 1).to_array(), a)
  assert_eq(@difflib.restore(delta.iter(), 2).to_array(), b)
}

#Close matches

///|
test "close matches" {
  debug_inspect(
    @difflib.get_close_matches("appel", ["ape", "apple", "peach", "puppy"]),
    content=(
      #|["apple", "ape"]
    ),
  )
}

#HTML side-by-side diff

///|
test "html diff" {
  let html = @difflib.HtmlDiff::new(wrapcolumn=40).make_file(
    ["a\n", "b\n"],
    ["a\n", "c\n"],
    fromdesc="old",
    todesc="new",
  )
  assert_true(html.has_prefix("\n<!DOCTYPE html>"))
}

#Differences from Python

• Python's restore checks which when its generator is first advanced; here the ValueError is raised immediately.
• Python's get_grouped_opcodes overwrites the first and last entries of the opcode list cached by get_opcodes(). This port works on a copy and leaves the cache unchanged.
• charjunk defaults to is_character_junk in ndiff and HtmlDiff. To get Python's charjunk=None, pass charjunk=_ => false, which behaves the same.
• unified_diff(color=true) always emits ANSI escapes. It does not check NO_COLOR, FORCE_COLOR or whether the output is a terminal.
• diff_bytes takes a DiffFormat (Unified or Context) rather than a diff function. Each byte becomes one character in the range U+0000–U+00FF and is encoded back afterwards. This is lossless and gives the same output as Python's ascii/surrogateescape round trip.
• HtmlDiff::make_file(charset=...) emulates encode(charset,"xmlcharrefreplace") for the ASCII, Latin-1 and UTF-8/16/32 codecs and all of their Python aliases (US-ASCII, ISO-8859-1, latin1, utf8, ...). It assumes that any other charset name can encode all of Unicode, where Python would use that codec or raise LookupError.
• Like Python's class attribute HtmlDiff._default_prefix, the anchor prefix counter is global and increases on every make_table/make_file call. HtmlDiff::reset_prefix_counter() resets it, for example to get reproducible output in tests.
• Negative context sizes (n in get_grouped_opcodes, unified_diff, context_diff; numlines in HtmlDiff) and a negative wrapcolumn panic. Depending on the input, Python can return meaningless hunks or raise IndexError, ZeroDivisionError or RecursionError for these.
• MoonBit strings are UTF-16, so a high surrogate directly followed by a low surrogate is the astral character they encode. Python can keep them as two separate code points; lone surrogates otherwise behave as in Python.
• Python's runtime type checks (_check_types) are unnecessary under MoonBit's static types and were dropped.

#Development

moon test                 # one backend; add --target js, native, wasm or all

git init .repos/cpython && cd .repos/cpython
git remote add origin https://github.com/python/cpython
git sparse-checkout set --no-cone /Lib/difflib.py /Lib/test/test_difflib.py /Lib/test/test_difflib_expect.html
git fetch --depth 1 --filter=blob:none origin 763b6edb0ec959bdfb78f098b76829365de36ca5
git checkout FETCH_HEAD && cd ../..
python3 tools/gen_corpus.py > corpus/corpus_fixtures_test.mbt
python3 tools/gen_html_fixtures.py .repos/cpython/Lib tools > html_fixtures_test.mbt
python3 tools/gen_templates.py .repos/cpython/Lib > html_templates.mbt
python3 tools/gen_charset.py > charset.mbt
moon fmt                  # generated files are committed in formatted form

#License