[HarfBuzz] Static initialization of C++ classes in harfbuzz

[Evan Martin]

I know of one project (Chrome) that is trying to make it a
compile-time error if you have any static initializers.

http://neugierig.org/software/chromium/notes/2011/08/static-initializers.html
discusses more of the reasons why.

On Sat, Jan 7, 2012 at 9:40 AM, Petr Kobalíček <kobalicek.petr at gmail.com> wrote:
> Hi list,
>
> While porting harfbuzz to be used with Fog I encountered that the
> static initialization of some C++ classes (global mutex and langs
> list) happened before Fog and this caused runtime error which I'm
> unable to fix without making changes to harfbuzz. Static
> initialization is unpredictable and the behavior is different across
> C++ compilers.
>
> There is no way how to fix this error without making changes to
> harfbuzz and I fixed it using initialization functions called:
>
>  - hb_init()
>  - hb_shutdown()
>
> These functions can be called from static class if the static
> initialization behavior is preferred:
>
>  class HbInitializer
>  {
>    HbInitializer() { hb_init(); }
>    ~HbInitializer() { hb_shutdown(); }
>  } hb_initializer;
>
> or separately by using my code.
>
> The question is: Is this good idea to include in harfbuzz or I should
> keep my changes in Fog, making this change internal and not visible to
> the master?
>
> Because if designed in such way, one can use something like:
>
> #if !defined(HB_NO_STATIC_INIT)
>  class HbInitializer
>  {
>    HbInitializer() { hb_init(); }
>    ~HbInitializer() { hb_shutdown(); }
>  } hb_initializer;
> #endif /* HB_NO_STATIC_INIT */
>
> to have complete control of the initialization process.
>
> Best regards
> Petr Kobalicek
> _______________________________________________
> HarfBuzz mailing list
> HarfBuzz at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/harfbuzz