[Uim] Re: Massive slowdown after revision 129

[Etsushi Kato]

On 2005/01/07, at 13:47, YamaKen wrote:

> At Fri, 07 Jan 2005 10:25:44 +0900,
> yamaken at bp.iij4u.or.jp wrote:
>> At Fri, 7 Jan 2005 08:29:07 +0900,
>> ekato at ees.hokudai.ac.jp wrote:
>>> I checked the behavior in r129 and I found that one to three GC occur
>>> with every key press in the revision.  This seems to causes the
>>> massive slowdown.
>>>
>>> With r128, GC occurs once per 20 to 55 key presses.
>>> With r146 (I now updated from r134), it occurs one per 4 to 11.
>>>
>>> Isn't it related to memory usage of the interpreter and the plugin
>>> system?  r146 has usable responsiveness for me, but it is still 
>>> slower
>>> than r128.
>>
>> I've now understood what you felt in r143. Yes, my tuning may
>> affect performance for such situation.
>>
>> After testing r146, I've found two distinct problems.
>>
>> 1. too short GC cycle (since r143)
>> 2. massive slowdown (since r129?)
>>
>> I'll resolve 1. soon.
>>
>> And I also fix 2. which causes exponential slowdown by
>> autorepeating a hundreds of kana-char in uim-skk. It seems to be
>> caused by improper memory consumption.
>
> I've resolved the 1. by r147. But for memory consumption vs GC
> frequency tradeoff, GC frequency is tuned more than r128. Try
> it.

OK.  With r147, GC takes place about once per 6 to 23 key presses with 
same environment mentioned above (using uim-skk's hiragana-direct mode 
and alphabet-direct mode).  It's better than r146.  Still r128 shows 
good response, but I can say that the difference of feeling is really 
little.


> 2. is my misunderstanding. It's not from uim but from wordwrap
> feature of text widget of GTK. Breaking long japanese-word line
> makes the slowdown away.

Hmm, I experienced slowdown through r129 to r142.  Do you have any idea 
about the reason?

Cheers,
-- 
Etsushi Kato
ekato at ees.hokudai.ac.jp