[PATCH] XKB: cache xkbcomp output for fast start-up v3 for 1.6.0

Li, Yan yanli at infradead.org
Wed Apr 8 00:22:39 PDT 2009

On Fri, Apr 03, 2009 at 07:55:34AM -0700, Alan Coopersmith wrote:
> Is there a race condition if two servers start at once (two vt's or
> multiseat, etc.) and the first server has started writing the .xkm
> but not finished when the second one finds it and tries to read?
> (Perhaps it should write to a temporary name and then rename() when
>  writing is complete.)    We had problems with that with the old
> code before we switched to exclusively using server-<display>.xkm
> naming.

Thanks for pointing this out.  Using a temp file and then rename() is
OK for me.  I'll post a new patch to fix this soon.

> I'd also rather see the Sha1 helper function moved to a file under
> the os dir and shared by both render/glyph.c & this code instead
> of duplicating the code here and giving people two places to make
> the same updates when adding new sha1 library implementations.

Sure.  But how about putting this into another patch (I can make it),
since I don't know how to test the change to render/glyph.c. By
putting this into another patch we can test the impact to
render/glyph.c separately.

Li, Yan

More information about the xorg mailing list