Final Release Notes for 1.0.4.

Author: carlosame

RELEASE_NOTES.txt

@@ -2,7 +2,7 @@
                             CSS4J RELEASE NOTES
                             ===================
 
-Release 1.0.4 - April 13, 2020 [End of Life - Unsupported]
+Release 1.0.4 - April 14, 2020 [End of Life - Unsupported]
 ----------------------------------------------------------
 
 Release Highlights
@@ -12,28 +12,26 @@ although users should upgrade to 2.0 or later as soon as possible (1.x is not
 formally maintained anymore). When upgrading, please keep in mind that 2.x
 releases require Java 8 or higher.
 
- * Adapted from the 2.0.2 Release Notes: *
+ * The following is adapted from the 2.0.2 Release Notes: *
 
  This library handled font-size values a bit different than the other values.
 When a font-size was expressed as a percentage, if the value of the parent font
-size could not be accurately determined the value was left as a precentage. This
-did not apply to computations that depended on the font size, like 'em' values
-in 'margin' or 'padding' properties: in that case default values were forced for
-that font size, and an absolute length was obtained.
-
- This procedure left a bit of potential inconsistency between computed values
-for the font-size and other properties, but was considered a trade-off that gave
-flexibility; in that way, users could apply their own font-size defaults after
-the computed style tree was evaluated, for example when applying a XSL sheet
-later.
-
- However, the way that var() values are handled left a code path for an infinite
-loop when a default font size was not set by the style sheets, and percentages
-are used inside a var(). The issue was detected and, while a specific fix for
-var() (and also var-like stuff like env() and attr()) could be rolled, I decided
-to force font-size percentage resolution in all cases, as the safest and most
-consistent approach. So the behaviour of the library now changes a bit, but the
-reasons for it are solid.
+size could not be accurately determined the value was left as a precentage in
+the serialized style. This did not apply to computations that depended on the
+font size, like 'em' values in 'margin' or 'padding' properties: in that case
+default values were forced for that font size, and an absolute length was
+obtained.
+
+ This procedure allows easy inspection of the specified vaues, but is inadequate
+when the serialized style is used as the input for other software packages.
+Changes to the library were made in the past to address this, but percentages of
+font-size values were kept.
+
+ Not only the serialization is inadequate: it also introduces a degradation in
+performance that can be highly noticeable with heavy documents (thousands of
+nodes, several tens of thousand rules). So the serialization of computed font
+sizes now avoids percentages: if you had unit tests depending on that, you may
+have to change them.
 
  Other than that, this release comes with interesting fixes, including one for a
 regression introduced in 1.0.3 (triggered by var() values inside gradients). All