Fontconfig - Feb 2006 - performance patch: reduce stat () call if all cache files can be trusted

Takashi IWAI <tiwai@suse.de> made a patch to reduce the amount of stat
() calls in the case when one is absolutely sure that all cache files
are OK and up-to-date and can be trusted. 

For details see

    http://bugzilla.novell.com/show_bug.cgi?id=148361

from comment #20 on.

There are problems if the cache files are not correct of course,
but nevertheless Takashi''s patch might be a useful performance hack
if a user is able to make sure that all cache files can be trusted.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: fc-validatecache-option.diff
Type: text/x-patch
Size: 4114 bytes
Desc: not available
Url :
http://lists.freedesktop.org/archives/fontconfig/attachments/20060210/ea88726a/fc-validatecache-option.bin
-------------- next part --------------

-- 
Mike FABIAN   <mfabian@suse.de>   http://www.suse.de/~mfabian
?????????????

Mike FABIAN wrote:
> Takashi IWAI <tiwai@suse.de> made a patch to reduce the amount of
stat
> () calls in the case when one is absolutely sure that all cache files
> are OK and up-to-date and can be trusted. 
> 
> For details see
> 
>     http://bugzilla.novell.com/show_bug.cgi?id=148361
> 
> from comment #20 on.
> 
> There are problems if the cache files are not correct of course,
> but nevertheless Takashi''s patch might be a useful performance
hack
> if a user is able to make sure that all cache files can be trusted.
This bug is not publicly readable.  Doesn''t it tend to be a bad idea to
trust cache files, in that it causes a pretty annoying failure mode?
How much performance can we potentially gain?

pat

Patrick Lam <plam@MIT.EDU> ????????:
> Mike FABIAN wrote:
>> Takashi IWAI <tiwai@suse.de> made a patch to reduce the amount of
stat
>> () calls in the case when one is absolutely sure that all cache files
>> are OK and up-to-date and can be trusted. 
>> 
>> For details see
>> 
>>     http://bugzilla.novell.com/show_bug.cgi?id=148361
>> 
>> from comment #20 on.
>> 
>> There are problems if the cache files are not correct of course,
>> but nevertheless Takashi''s patch might be a useful performance
hack
>> if a user is able to make sure that all cache files can be trusted.
>
> This bug is not publicly readable.
I changed something to make it publicly readable just now.

-- 
Mike FABIAN   <mfabian@suse.de>   http://www.suse.de/~mfabian
?????????????

Takashi Iwai wrote:
> However, such an option might be helpful for almost read-only sytems
> like thin clients. 
It sounds like you''d want to merge all config files into one on a thin
client to get this perf gain without adding unsafe options.

pat

Takashi Iwai wrote:
> At Fri, 10 Feb 2006 17:16:17 -0500,
> Patrick Lam wrote:
> 
>>Takashi Iwai wrote:
>>
>>>However, such an option might be helpful for almost read-only sytems
>>>like thin clients. 
>>
>>It sounds like you''d want to merge all config files into one on
a thin
>>client to get this perf gain without adding unsafe options.
> 
> Imagine all files are mounted as read-only over NFS, including fonts
> and font cache files.  Why bother to check and create local caches at
> each time you open apps in such a case?
So I''ve read the discussion in bug 148361 now.  I think it''s
definitely
leaving a lot of rope around for people to hang themselves with to trust
cache files; this isn''t a failure mode that fontconfig ought to have,
in
my opinion.

We''re only talking about checking global (not local) caches; if
they''re
wrong, then we''re in trouble.

If you really wanted to have an option like that, one kind of
off-the-wall way of limiting the damage would be to randomly check the
caches, say with a probability of 1%.  Then it would complain loudly if
you said the caches were up to date but they weren''t.

pat