ixml parser: keep a line feed inside a value

[oblomov-dev]

What is wrong

if_ixml_parser~parse removes every literal line feed from the document
before it starts tokenizing, src/ixml/cl_ixml.clas.locals_imp.abap:

REPLACE ALL OCCURRENCES OF |\n| IN lv_xml WITH ||.

A LF is not markup, it is character data, so the strip also hits the text
nodes. Nothing writes a LF as a character reference on the way out either
(lcl_data_to_xml=>escape_text writes &, < and >, added in #1193, and
leaves the LF as it is), so there is no form in which a line break survives
the document.

The visible effect is the asXML roundtrip: a two-line value comes back as one
line.

DATA lv_xml TYPE string.
DATA: BEGIN OF ls_data,
        field TYPE string,
      END OF ls_data.

ls_data-field = |a{ cl_abap_char_utilities=>newline }b|.

CALL TRANSFORMATION id SOURCE data = ls_data RESULT XML lv_xml.
CLEAR ls_data.
CALL TRANSFORMATION id SOURCE XML lv_xml RESULT data = ls_data.

" system: `a\nb`, here: `ab`

The change

The strip is not needed. The parse loop already skips the whitespace between
tags itself, at the end of every iteration:

lv_rest = lv_xml.
SHIFT lv_rest LEFT DELETING LEADING lv_whitespace.
IF lv_rest IS INITIAL OR lv_rest(1) = '<'.
  lv_xml = lv_rest.
ENDIF.

and lv_whitespace is cl_abap_char_utilities=>get_simple_spaces_for_cur_cp( ),
which contains newline and the CR. So a pretty-printed document is read
exactly as before — the indentation is dropped because it is whitespace
between tags, not because a LF was removed from the string. What changes is
only the text node: the line break stays where a system keeps it.

One line removed. The write side is untouched — a LF is legal raw in character
data, so no escaping is added and every value serializes byte for byte as it
does today.

Tests

• kernel_call_transformation: newline_roundtrip — the two-line value above,
  through CALL TRANSFORMATION id and back
• cl_ixml: parse_value_with_newline — the parser keeps the LF in the
  #text node

The two existing tests that cover this area, root_element_after_crlf and
first_child_after_crlf, are unchanged and still pass.

npm test is green: abaplint: 0 issue(s) found, 735 file(s) analyzed, and
the full unit run passes.

Why it matters downstream

abap2UI5 is stateless per request: the app instance — every attribute of the
app class, including the strings bound to a TextArea — is serialized with
CALL TRANSFORMATION id into the draft and parsed back on the next request
(z2ui5_cl_ui5_util_context=>xml_stringify / xml_parse). On the transpiled
backend that means the user types two lines and the next request has one.
abap2UI5 carries a local pre-transpile patch for this today; with this merged
the last one goes away.