Re: msi: remove limit on number of handles

[Michael Stefaniuc]

Dan Kegel wrote:
> Changelog:
>  msi: remove limit on number of handles
>
> handle.c  |   30 +++++++++++++++++++++++-------
> msipriv.h |    1 -
> 2 files changed, 23 insertions(+), 8 deletions(-)
>
> diff --git a/dlls/msi/handle.c b/dlls/msi/handle.c
> index 1cd82b3..79664f5 100644
> @@ -66,18 +67,29 @@ MSIHANDLE alloc_msihandle( MSIOBJECTHDR
>  EnterCriticalSection( &MSI_handle_cs );
>
> - if (msihandletable == NULL)
> - {
> - /* start table off small to avoid hiding resize bugs */
> -
>    msihandletable_size = 1;

Isn't this very small start value something to be used only for
debugging while one writes the resizing code? For "production" I would
use a value that is big enough to handle the common case (handles at
least 50% of the real life cases). As the old MSIMAXHANDLES was good
enough for a whole lot of programs i would define that as common case to
save me work of finding what the common case is.
> -
>    msihandletable = msi_alloc_zero(1*sizeof(msi_handle_info));
> - }
> -

bye
    michael

msi: remove limit on number of handles

[Dan Kegel]

On 8/23/06, Michael Stefaniuc  wrote:
> > -
> > /* start table off small to avoid hiding resize bugs */
> > -
> >    msihandletable_size = 1;
>
> Isn't this very small start value something to be used only for
> debugging while one writes the resizing code? For "production" I would
> use a value that is big enough to handle the common case (handles at
> least 50% of the real life cases).

OK, once wine hits 1.0 and is in production, let's go back and do that.

Seriously, what's the disadvantage to keeping the initial size small?
The overhead of six realloc's is utterly insignificant, and there's
a fairly good chance somebody will get the locking wrong in an update
to that code, so making the reallocs happen in the common case
is a *good* thing that will help guard against bugs.
- Dan

msi: remove limit on number of handles

[Alexandre Julliard]

"Dan Kegel"  writes:
> On 8/23/06, Michael Stefaniuc  wrote:
> > > -
> > > /* start table off small to avoid hiding resize bugs */
> > > -
> > >    msihandletable_size = 1;
> >
> > Isn't this very small start value something to be used only for
> > debugging while one writes the resizing code? For "production" I would
> > use a value that is big enough to handle the common case (handles at
> > least 50% of the real life cases).
>
> OK, once wine hits 1.0 and is in production, let's go back and do that.

Nobody is going to remember to do that. It's OK to use 1 for testing,
but once you have done that please go back to a more reasonable value.