On 03/01/2016 03:40 AM, Ashod Nakashian wrote:
On 02/29/2016 05:42 PM, Ashod Nakashian wrote:
On 02/29/2016 11:44 AM, Stephan Bergmann wrote:
Ashod, can you get that fixed? I assume that it would be a common
issue that such wrapper include files need special treatment in
update_pch (and an indicator that automatic generation of such
precompiled_*.hxx just doesn't work), but don't find anything related
in that script.
Sure.
If the diff of precompiled_xsec_xmlsec.hxx is the desired result, then
the fix is very simple. update_pch has provision to exclude any header
that we deem problematic. We only need to add 'xmlsec/*' as exclusion
and it will never get added to the pch file on subsequent runs.
Unfortunately, this doesn't work under Linux. The wrapper header is
resulting in errors when it's included (which is why I had excluded it).
I know pch is less important under Linux, but it does improve build
times there too.
So the choice is between fixing the issue you address with your patch
(which I'm not sure I understand) or maintaining pch support on Linux.
It seems both would not be possible without modifying the wrapper header.
Perhaps if you shed some light on what you are addressing I might be
able to come with a solution?
What is required is that no xmlsec/* ever get included from outside of
xmlsecurity/xmlsec-wrapper.h (because the latter controls a macro that
the former #if on). That means that no xmlsec/* may show up in a
precompiled_*.hxx. Whether or not xmlsecurity/xmlsec-wrapper.h shows up
in a precompiled_*.hxx is irrelevant for correctness, so whichever is
easier to implement in update_pch should be fine. (I just arbitrarily
did include it in my manual modifications of
xmlsecurity/inc/pch/precompiled_xsec_xmlsec.hxx to approximate what was
there before.)
_______________________________________________
LibreOffice mailing list
LibreOffice@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/libreoffice
