Fix 'ucs-normalize' tests following Unciode 17.0 import

* lisp/international/ucs-normalize.el
(ucs-normalize-composition-exclusions): Doc fix.

* test/lisp/international/ucs-normalize-tests.el
(ucs-normalize-tests--failing-lines-part1): Update to _really_
match Unicode 17.0.

* admin/notes/unicode: Update instructions.

1 files changed, 13 insertions, 8 deletions

diff --git a/admin/notes/unicode b/admin/notes/unicode
index ae94b5c0a3f..21ce2c4cda3 100644
--- a/admin/notes/unicode
+++ b/admin/notes/unicode
@@ -97,18 +97,23 @@ might need to be updated because it knows about used and unused ranges
 of Unicode codepoints, which a new release of the Unicode Standard
 could change.
 
-The data used by ucs-normalize.el might need to be updated.
-Specifically, the values of 'ucs-normalize-composition-exclusions' and
-'check-range", defined at the beginning of ucs-normalize.el, should be
-verified against the latest Unicode data files.
-
-Run the ucs-names test:
+Next, run the ucs-names test:
 
   make -C test lisp/international/mule-tests
 
 If it fails, the exclusion ranges of codepoints in
-'mule-cmds-tests--ucs-names-missing-names' may need to be updated to the
-added Unicode codepoints.
+'mule-cmds-tests--ucs-names-missing-names' may need to be updated to
+account for the added Unicode codepoints.
+
+The data used by ucs-normalize.el might need to be updated.
+Specifically, the values of 'ucs-normalize-composition-exclusions' and
+'check-range", defined at the beginning of ucs-normalize.el, should be
+verified against the latest Unicode data files.  But even if
+ucs-normalize.el is not modified, make sure it has been byte-compiled
+_after_ the lisp/international/uni-*.el files were regenerated, because
+it depends on uni-decomposition.el and uni-combining.el.  (We don't have
+this dependency in lisp/Makefile.in because recompiling ucs-normalize.el
+is expensive, and updates for a new Unicode version are rare.)
 
 Next, test normalization functions against NormalizationTests.txt,
 in the top-level directory run: