shape: More tweaks to space handling

It is a difficult dance between HarfBuzz and pango
which glyphs we let hb synthesize, and which ones
we need to fake ourselves.

HarfBuzz knows how to create other spaces from 0x20,
so if the font is lacking spaces, we want to tell
HarfBuzz that we don't have that glyph *except* for
0x20 where we need to fake one, and provide HarfBuzz
with a suitable width.

Update affected tests.

1 files changed, 16 insertions, 4 deletions

diff --git a/pango/shape.c b/pango/shape.c
index 90334cc1..30bceb80 100644
--- a/pango/shape.c
+++ b/pango/shape.c
@@ -137,12 +137,24 @@ pango_hb_font_get_nominal_glyph (hb_font_t      *font,
   if (hb_font_get_nominal_glyph (context->parent, unicode, glyph))
     return TRUE;
 
-  /* HarfBuzz knows how to synthesize spaces, so never replace them
-   * with unknown glyphs, but we do need to tell HarfBuzz that the
-   * font does not have a glyph.
+  /* HarfBuzz knows how to synthesize other spaces from 0x20, so never
+   * replace them with unknown glyphs, just tell HarfBuzz that we don't
+   * have a glyph.
+   *
+   * For 0x20, on the other hand, we need to pretend that we have a glyph
+   * and rely on our glyph extents code to provide a reasonable width for
+   * PANGO_GET_UNKNOWN_WIDTH (0x20).
    */
   if (g_unichar_type (unicode) == G_UNICODE_SPACE_SEPARATOR)
-    return FALSE;
+    {
+      if (unicode == 0x20)
+        {
+          *glyph = PANGO_GET_UNKNOWN_GLYPH (0x20);
+          return TRUE;
+        }
+
+      return FALSE;
+    }
 
   *glyph = PANGO_GET_UNKNOWN_GLYPH (unicode);