Re: hmtx optimization over the Google Fonts collection

FYI, I asked the compression team about it and was advised this type of
response to small changes in the input is pretty normal.

On Wed, Jan 20, 2016 at 10:11 AM, Behdad Esfahbod <behdad@google.com> wrote:

> Sure.  But I'm sure they are interested in taking a look :)
>
> On Wed, Jan 20, 2016 at 6:58 PM, Roderick Sheeter <rsheeter@google.com>
> wrote:
>
>> Yup, will do. Fingers crossed they can work further magic for us :D
>>
>> However, I would think this expected behavior as it's hard to predict how
>> well brotli can compress a given input.
>>
>>
>> On Wed, Jan 20, 2016 at 9:27 AM, Behdad Esfahbod <behdad@google.com>
>> wrote:
>>
>>> Right... But the ones, say, growing 2k in size are interesting.  Can you
>>> ping Brotli people so they are at least aware of this?
>>>
>>> On Wed, Jan 20, 2016 at 6:13 PM, Roderick Sheeter <rsheeter@google.com>
>>> wrote:
>>>
>>>> I think it's because the result is can be an input buffer that is less
>>>> friendly to brotli.
>>>>
>>>> To give an example, lets take ArbutusSlab-Regular.ttf. It's hmtx barely
>>>> saves anything (Was 1734 now 1733 [bytes]). The main compression step gets
>>>> a smaller input but isn't able to compress it quite as well:
>>>>
>>>> hmtx_opt: Compressed 63150 to 29992.
>>>> not opt: Compressed 63151 to 29939.
>>>>
>>>> Plus we need an additional UIntBase128 to store the transform length.
>>>>
>>>> On Wed, Jan 20, 2016 at 9:03 AM, Behdad Esfahbod <behdad@google.com>
>>>> wrote:
>>>>
>>>>> Hey,
>>>>>
>>>>> I'm sure everyone wants to know: why would any font get larger?
>>>>>
>>>>> On Wed, Jan 20, 2016 at 5:45 PM, Roderick Sheeter <rsheeter@google.com
>>>>> > wrote:
>>>>>
>>>>>> I did a test run of hmtx optimization over the Google Fonts
>>>>>> collection and thought the results might be of interest. A few key results:
>>>>>>
>>>>>>    - Of 1754 font files, 80.4% (1411) got smaller, 16.4% (288) had
>>>>>>    no change, and 3.1% (55) got larger.
>>>>>>    - For fonts with savings, average was 466 bytes or 1.08% of size
>>>>>>       - Across all fonts, average was 368 bytes or 0.86% of size
>>>>>>
>>>>>> Cheers, Rod S.
>>>>>>
>>>>>> Per-font results can be seen in
>>>>>> https://docs.google.com/spreadsheets/d/1dgL-il6fIHaHJghlzXz7aM_HEtes9G7Pt7TsnlsxsGc/edit?usp=sharing
>>>>>> .
>>>>>>
>>>>>>
>>>>>>
>>>>>
>>>>
>>>
>>
>