Re: [ft-devel] [PATCH] Always check the checksum to identify tricky fonts

[suzuki toshiya]

Dear Kan-Ru Chen,

Thank you for advice, I've restored the original order;
check by the familyname, then, check by the checksums.

However, if we care the latency by the checksum comparison,
also we should care the latency of the non-tricky fonts,
like TimesNewRoman. By this modification (check the tricky-
ness by the checksums, even when family name doesn't indicate
its trickyness), the loading time of most fonts will get
the extra latency.

For future improvement to reduce such extra latency,
I didn't wrap 2 checks. Sorry Алексей.

Regards,
mpsuzuki

Kan-Ru Chen wrote:
> address@hidden writes:
> 
>> Dear Kan-Ru Chen,
>>
>> On Sat, 23 Apr 2011 18:37:53 +0800
>> Kan-Ru Chen <address@hidden> wrote:
>>> example PDF file attached.
>> Thank you very much for providing interesting sample,
>> I have to fix this issue. Your suggestion to prioritize
>> the tricky font detection checksum than that by family
>> name helps another issue (tricky font with non-ASCII
>> family),
>>      http://lists.gnu.org/archive/html/freetype-devel/2011-01/msg00003.html
>> so I will apply your patch.
> 
> Thank you.
> 
> I find you changed the order of the checks. I think it is better to keep
> the original order, because the family_name check is faster than the
> checksum check (string comparison vs. some 32-bit arithmetic), and more
> accurate, consider that checksums may have collisions.
> 
>> But, there was a discussion that the tricky font detection
>> by the checksum might be inaccurate for too short table,
>>      http://lists.gnu.org/archive/html/freetype-devel/2010-11/msg00033.html
>> so there is a possiblity that I have to revert the change
>> and improve the condition to detect mangled family name.
> 
> I see. Would it help if we use stronger hash method?
> 
>> # Among the bugzillas of poppler library and its client
>> # applications, "please enable freetype hinting always",
>> # "please disable freetype hinting always" are popular
>> # requests.
>>
>> Your sample PDF seems to be generated by "Crystal Reports",
>> if you know any other softwares generating a PDF including
>> mangled family names, please let me know.
> 
> OK.
> 
> Cheers,
> Kanru