Clang says:
stripped_scanner.ll:198:46: error: for loop has empty body [-Werror,-Wempty-body]
stripped_scanner.ll:198:46: note: put the semicolon on a separate line to silence this warning [-Wempty-body]

gcc uses a register for the function call--and it tried r9 here...

The "mr" stores r11 to indeterminate register (r9 in my case) before
storing it into the output variable, thus overwriting one of our input
arguments... This later leads to nice segfault in
testtools/source/bridgetest ...

I suppose there is a better way to get the variable, but I do not know
it .-)

The copy/paste formulas code is getting a bit complex. I will try to
write some test cases for it.

regression from 5dd4f9ec
The change:
-    if(m_GlyphsList.size()>1)
+    if(!m_GlyphsList.empty())
is incorrect, so change it back.

If we sort the keywords once in the constructor, then we can do binary
search when looking up RTF keywords, and that speeds up the import by
about 20% using the first testcase from the bug.

I broke that when I added CB_CONTEXTUALSPACING

To fix the build. I can't test it on win32, thought it would work...