Re: double patch [was Re: libgsf: very preliminary patch]

From: Ryan Pavlik <abiryan_at_ryand.net>
Date: Wed Feb 22 2006 - 07:11:50 CET

J.M. Maurer wrote:

>On Tue, 2006-02-21 at 19:56 +0000, Tomas Frydrych wrote:
>
>
>>J.M. Maurer wrote:
>>
>>
>>>On Tue, 2006-02-21 at 19:15 +0000, Tomas Frydrych wrote:
>>>
>>>
>>>
>>>>J.M. Maurer wrote:
>>>>
>>>>
>>>>
>>>>>I'm willing to update the double-branch if I'm allowed to merge it into
>>>>>HEAD after that....
>>>>>
>>>>>
>>>>No objections from me, but we will almost certainly need to continue
>>>>maintaining an int branch after that -- there is no FP on the 770, and
>>>>the soft FP is not great performance wise.
>>>>
>>>>
>>>Trust me, that won't be possible.. the patch touches almost every line
>>>in af/gr/* and text/* ... it would be a 100% fork, where every single
>>>patch has to be rewritten for the int branch.
>>>
>>>
>>I don't think we can get around this, the FP is not viable for the many
>>of the embedded platforms, and I would like us to keep a foothold there.
>>At the same time we want to take advantage of the FP apis that are
>>becoming available on the desktop like cairo.
>>
>>I think we could reduce the impact of the change if we typedef UT_Unit
>>as double / int and have some comparison macros, like UT_Unit_eq(),
>>UT_Unit_gt(), _lt () _ge(), _le().
>>
>>
>
>Sure, but I've fairly sure that no-one wants to redo 100000+ lines of
>code again (which took me about 2 complete months when I had all time of
>the world being a student)
>
>I think the branch is dead then, and we will have to live with the fact
>that we can't do kerning, or even draw a line from 0,0 to 9,0 and expect
>10 pixels to be drawn.
>
>Marc
>
>
While I appreciate the amount of work it will take to resolve this, I
think that perhaps a defeatist attitude will not get us any further, and
this is the least- desirable result... I know I have been awaiting the
doubles patch for a while on Win32. Certainly we aren't the only
fp-using app: has anyone actually tried the performance of doubles Abi
on a fpu-free device like the 770? Shouldn't the int vs. doubles be
able to be switched at compile (or even run-) time, though an
inheritance or something?

Not wanting to give up,
Ryan

-- 
Ryan Pavlik
AbiWord Win32 Platform Maintainer
www.abisource.com
"Optimism is the father that leads to achievement." - Helen Keller
Received on Wed Feb 22 07:11:55 2006

This archive was generated by hypermail 2.1.8 : Wed Feb 22 2006 - 07:11:56 CET