Speed up whitespace skipping in the lexer

lexer::skip_whitespace() called get() for every whitespace byte, and
get() checks the (almost always false, once past the first character)
next_unget flag on every call. skip_whitespace() now reads its first
character with get() (needed to honor a pending unget() left over from
finishing the previous token, e.g. scan_number() always ungets the
character that terminated the number) and every further whitespace
character with a new get_ignoring_pending_unget() variant that skips
that branch, since nothing in the loop calls unget().

This is a narrower fix than the full contiguous-buffer bulk-skip
suggested in the issue (scan a run of whitespace directly in the
adapter's buffer and update position counters once per run). That
approach depends on bulk-scan adapter infrastructure
(supports_bulk_scan/bulk_data()/bulk_skip()) introduced by the open,
unmerged parser-performance PR #5283, which this change intentionally
does not depend on or replicate. Building new bulk-scan adapter
infrastructure from scratch was judged out of scope/riskier than
warranted here, so this change is limited to the safe, always-correct
improvement of removing redundant per-character bookkeeping from the
existing byte-at-a-time loop; full bulk-skipping is left as future
work once #5283 (or equivalent adapter support) lands.

Line/column/byte-offset bookkeeping is untouched and verified
bit-for-bit identical before and after this change, including for
pretty-printed (dump(4)) input with embedded newlines.

Fixes #5412

Stacked on top of the PR for #5411 (branch
issue-5411-lexer-skip-conversion).

Signed-off-by: Niels Lohmann <mail@nlohmann.me>

include/nlohmann/detail/input/lexer.hpp

@@ -1459,6 +1459,14 @@ scan_number_done:
             current = ia.get_character();
         }
 
+        return track_after_read();
+    }
+
+    /// shared tail of get() / get_ignoring_pending_unget(): capture the
+    /// character for error messages (if needed) and update line/column
+    /// bookkeeping for the character now in `current`
+    char_int_type track_after_read()
+    {
         // seekable adapters reconstruct the token lazily on error (see
         // get_token_string), so the eager per-character copy is skipped
         capture_char(std::integral_constant<bool, lazy_token_string> {});
@@ -1472,6 +1480,30 @@ scan_number_done:
         return current;
     }
 
+    /*!
+    @brief like get(), but for call sites that can prove no unget() is pending
+
+    get() has to check the `next_unget` flag on every call, because a
+    previous token may have ended with unget() (e.g. scan_number() always
+    ungets the character that terminated the number, so the next call to
+    scan() can see it again). skip_whitespace() reads that first,
+    possibly-ungotten character via a plain get(), but every further
+    character it reads is guaranteed to be a fresh read: nothing between
+    those calls invokes unget(). This variant skips the (otherwise always
+    false) next_unget branch for those calls; it is not a general
+    replacement for get().
+    */
+    char_int_type get_ignoring_pending_unget()
+    {
+        JSON_ASSERT(!next_unget);
+
+        ++position.chars_read_total;
+        ++position.chars_read_current_line;
+        current = ia.get_character();
+
+        return track_after_read();
+    }
+
     /// seekable adapter: nothing to capture, the token is rebuilt on error
     void capture_char(std::true_type /*lazy*/) const noexcept {}
 
@@ -1667,11 +1699,16 @@ scan_number_done:
 
     void skip_whitespace()
     {
-        do
+        // the first character may be a pending unget() left over from the
+        // previous token (see get_ignoring_pending_unget()); every
+        // subsequent character read by this loop is guaranteed fresh, since
+        // nothing below calls unget()
+        get();
+
+        while (current == ' ' || current == '\t' || current == '\n' || current == '\r')
         {
-            get();
+            get_ignoring_pending_unget();
         }
-        while (current == ' ' || current == '\t' || current == '\n' || current == '\r');
     }
 
     token_type scan()

single_include/nlohmann/json.hpp

@@ -9242,6 +9242,14 @@ scan_number_done:
             current = ia.get_character();
         }
 
+        return track_after_read();
+    }
+
+    /// shared tail of get() / get_ignoring_pending_unget(): capture the
+    /// character for error messages (if needed) and update line/column
+    /// bookkeeping for the character now in `current`
+    char_int_type track_after_read()
+    {
         // seekable adapters reconstruct the token lazily on error (see
         // get_token_string), so the eager per-character copy is skipped
         capture_char(std::integral_constant<bool, lazy_token_string> {});
@@ -9255,6 +9263,30 @@ scan_number_done:
         return current;
     }
 
+    /*!
+    @brief like get(), but for call sites that can prove no unget() is pending
+
+    get() has to check the `next_unget` flag on every call, because a
+    previous token may have ended with unget() (e.g. scan_number() always
+    ungets the character that terminated the number, so the next call to
+    scan() can see it again). skip_whitespace() reads that first,
+    possibly-ungotten character via a plain get(), but every further
+    character it reads is guaranteed to be a fresh read: nothing between
+    those calls invokes unget(). This variant skips the (otherwise always
+    false) next_unget branch for those calls; it is not a general
+    replacement for get().
+    */
+    char_int_type get_ignoring_pending_unget()
+    {
+        JSON_ASSERT(!next_unget);
+
+        ++position.chars_read_total;
+        ++position.chars_read_current_line;
+        current = ia.get_character();
+
+        return track_after_read();
+    }
+
     /// seekable adapter: nothing to capture, the token is rebuilt on error
     void capture_char(std::true_type /*lazy*/) const noexcept {}
 
@@ -9450,11 +9482,16 @@ scan_number_done:
 
     void skip_whitespace()
     {
-        do
+        // the first character may be a pending unget() left over from the
+        // previous token (see get_ignoring_pending_unget()); every
+        // subsequent character read by this loop is guaranteed fresh, since
+        // nothing below calls unget()
+        get();
+
+        while (current == ' ' || current == '\t' || current == '\n' || current == '\r')
         {
-            get();
+            get_ignoring_pending_unget();
         }
-        while (current == ' ' || current == '\t' || current == '\n' || current == '\r');
     }
 
     token_type scan()

tests/src/unit-class_parser.cpp

@@ -1482,6 +1482,62 @@ TEST_CASE("parser class")
         CHECK(accept_helper("\"\\uD80C\\uFFFF\"") == false);
     }
 
+    SECTION("issue #5412 - whitespace skipping bookkeeping (compact vs. pretty-printed)")
+    {
+        // lexer::skip_whitespace() reads its first character with get() (to
+        // honor a possibly pending unget() from the previous token) and every
+        // further whitespace character with get_ignoring_pending_unget() (a
+        // get() variant that skips the then-always-false next_unget check).
+        // This must not change the reported byte offset, line, or column of
+        // a syntax error, even when a long run of whitespace containing
+        // multiple newlines is skipped beforehand (as with pretty-printed
+        // input). The expected values below were captured from the
+        // unmodified do-while(get()) loop, so any regression that miscounts
+        // characters or newlines while skipping whitespace changes them.
+        const auto check_error = [](const std::string & input, std::size_t expected_byte,
+                                    const std::string & expected_what)
+        {
+            CAPTURE(input)
+            try
+            {
+                json _ = json::parse(input);
+                FAIL_CHECK("expected a parse_error, but parsing succeeded");
+            }
+            catch (const json::parse_error& e)
+            {
+                CHECK(e.byte == expected_byte);
+                CHECK(std::string(e.what()) == expected_what);
+            }
+        };
+
+        // a nested document, serialized both compactly and pretty-printed
+        // (dump(4)), each truncated right before the final closing '}' so
+        // that the parser hits EOF after skipping all of the (in the
+        // pretty-printed case, substantial) indentation whitespace
+        const json doc =
+        {
+            {"a", 1},
+            {"b", json::array({true, false, nullptr, "x"})},
+            {"c", json::object({{"d", 3.14}, {"e", json::array({1, 2, 3})}})}
+        };
+
+        const std::string compact = doc.dump();
+        const std::string pretty = doc.dump(4);
+
+        check_error(compact.substr(0, compact.size() - 1), 60,
+                    "[json.exception.parse_error.101] parse error at line 1, column 60: syntax error while parsing object - unexpected end of input; expected '}'");
+        check_error(pretty.substr(0, pretty.size() - 1), 193,
+                    "[json.exception.parse_error.101] parse error at line 17, column 1: syntax error while parsing object - unexpected end of input; expected '}'");
+
+        // an invalid token appearing after several indented, multi-line
+        // whitespace runs vs. the same document without any of that
+        // whitespace
+        check_error("{\n    \"a\": 1,\n    \"b\": [\n        true,\n        false\n    ],\n    \"c\": @\n}", 70,
+                    "[json.exception.parse_error.101] parse error at line 7, column 10: syntax error while parsing value - invalid literal; last read: '\"c\": @'");
+        check_error("{\"a\":1,\"b\":[true,false],\"c\":@}", 29,
+                    "[json.exception.parse_error.101] parse error at line 1, column 29: syntax error while parsing value - invalid literal; last read: '\"c\":@'");
+    }
+
     SECTION("tests found by mutate++")
     {
         // test case to make sure no comma precedes the first key