From 06e771fab2d8feb78f8c93a92d45bf8c94db0197 Mon Sep 17 00:00:00 2001
From: Sanjay Santhanam <51058514+Sanjays2402@users.noreply.github.com>
Date: Thu, 2 Jul 2026 23:29:51 -0700
Subject: [PATCH] Insert separator for more token pairs that would otherwise
 merge

serialize() inserts a `/**/` comment between two adjacent component values
whenever emitting them back-to-back would let the tokenizer re-read them as a
single, different token. The pairs that need this are listed in BAD_PAIRS.

BAD_PAIRS was missing seven pairs from the CSS Syntax serialization table, so
serialize() silently changed the token stream on round trip once the two tokens
were adjacent (e.g. after whitespace was dropped by a transform):

    tokens                       serialize()   re-parses as
    number  + '%'  delim         '1%'          percentage
    number  + '-->' (CDC)        '1-->'        dimension '1--' + '>'
    '#'     + '-'  delim         '#-'          hash '#-'
    '#'     + '-->' (CDC)        '#-->'        hash '#--' + '>'
    '-'     + '-'  delim         '--'          ident '--'
    '-'     + '-->' (CDC)        '--->'        ident '---' + '>'
    '@'     + '-->' (CDC)        '@-->'        at-keyword '@--' + '>'

Root cause: a lone '-' delimiter merges after '#'/'-' (forming a hash or ident),
a CDC ('-->') merges after '#'/'-'/number/'@' (its leading '--' is consumed as
hash / ident / dimension-unit / at-keyword characters), and a '%' delimiter
merges into a preceding number to form a percentage. dimension+CDC and the
number+ident/function/url pairs were already covered; these seven were the gap.

Add the missing pairs (mirroring the existing spec table) and a parametrized
regression test that drops the whitespace between each pair and asserts the
serialized output still re-parses to the same two tokens.

https://drafts.csswg.org/css-syntax/#serialization
---
 tests/test_tinycss2.py | 25 +++++++++++++++++++++++++
 tinycss2/serializer.py |  9 +++++++++
 2 files changed, 34 insertions(+)

diff --git a/tests/test_tinycss2.py b/tests/test_tinycss2.py
index fb1130a..c02ed72 100644
--- a/tests/test_tinycss2.py
+++ b/tests/test_tinycss2.py
@@ -703,3 +703,28 @@ def test_escape_in_function_name():
     function, = parse_component_value_list('\\dddf()')
     assert function.type == 'function'
     assert function.name == '\udddf'
+
+
+@pytest.mark.parametrize('source', [
+    '1 %',       # number + '%'   would merge into a percentage
+    '1.5 %',     # (float) number + '%'
+    '+1 %',      # (signed) number + '%'
+    '1 -->',     # number + CDC    '--' would extend the number's unit
+    '# -',       # '#' + '-'       would merge into a hash token
+    '# -->',     # '#' + CDC
+    '- -',       # '-' + '-'       would merge into an ident '--'
+    '- -->',     # '-' + CDC
+    '@ -->',     # '@' + CDC       would merge into an at-keyword
+])
+def test_serialize_adjacent_tokens_do_not_merge(source):
+    # A pair of tokens separated only by whitespace must survive a round trip
+    # even once that whitespace is dropped (e.g. by a minifier): serialize()
+    # has to insert a separator, otherwise the two tokens re-parse as one.
+    # These pairs are listed in the CSS Syntax serialization table but were
+    # missing from BAD_PAIRS. See https://drafts.csswg.org/css-syntax/#serialization
+    tokens = parse_component_value_list(source)
+    del tokens[1]  # drop the whitespace token between the two component values
+    reparsed = parse_component_value_list(serialize(tokens))
+    reparsed = [t for t in reparsed if t.type not in ('whitespace', 'comment')]
+    assert [t.type for t in reparsed] == [t.type for t in tokens]
+    assert [t.serialize() for t in reparsed] == [t.serialize() for t in tokens]
diff --git a/tinycss2/serializer.py b/tinycss2/serializer.py
index 032b924..a1c5338 100644
--- a/tinycss2/serializer.py
+++ b/tinycss2/serializer.py
@@ -133,6 +133,15 @@ def _serialize_to(nodes, write):
     [(a, b)
         for a in ('ident', 'at-keyword', 'hash', 'dimension')
         for b in ('-', '-->')] +
+    # Per the CSS Syntax serialization table, a lone U+002D HYPHEN-MINUS
+    # delimiter also merges after '#'/'-' (forming a hash or ident), and a CDC
+    # ('-->') merges after '#'/'-'/number/'@' (its leading '--' is consumed as
+    # hash/ident/dimension-unit/at-keyword characters). A '%' delimiter merges
+    # into a preceding number to form a percentage.
+    # https://drafts.csswg.org/css-syntax/#serialization
+    [(a, '-') for a in ('#', '-')] +
+    [(a, '-->') for a in ('#', '-', 'number', '@')] +
+    [('number', '%')] +
     [(a, b)
         for a in ('#', '-', 'number', '@')
         for b in ('ident', 'function', 'url')] +