[TYPO3-dev] set_no_cache is bad. What's next?

[Peter Russ]

--- Original Nachricht ---
Absender:   Franz Holzinger
Datum:       27.10.2009 21:28:
> Martin Kutschker a écrit :
>> Steffen Kamper schrieb:
>>> "Martin Kutschker" <martin.kutschker-n0spam at no5pam-blackbox.net> 
>>> schrieb im Newsbeitrag 
>>> news:mailman.1.1175458083.11183.typo3-dev at lists.netfielders.de...
>>>> Steffen Kamper schrieb:
>>>>> "Dmitry Dulepov" <9f4eetb02 at sneakemail.com> schrieb im Newsbeitrag 
>>>>> news:mailman.1.1175173443.25187.typo3-dev at lists.netfielders.de...
>>>>>> Hi!
>>>>>>
>>>>>> From time to time I see that people try to use 
>>>>>> $TSFE->set_no_cache() in extensions. You know what happens next: 
>>>>>> no caching for pages, performance is terrible, web server is not 
>>>>>> responding, etc. At the same time, sometimes, in really rare cases 
>>>>>> caching must be disabled. But definitely not this way from 
>>>>>> extensions.
>>>>>>
>>>>>> Should we change this function in some way? Should it explain 
>>>>>> people that using it is no-good?
>>>>>>
>>>>>> Personally I would vote for radical methods like (1) keeping it 
>>>>>> but emptying or (2) writing commented phrase inside generated FE 
>>>>>> page that set_no_cache was used, so do not expect good performance.
>>>>>>
>>>>>> Any other ideas/opinions?
>>>>>>
>>>>>> -- 
>>>>>> Dmitry Dulepov
>>>>>>
>>>>>> Web: http://typo3bloke.net/
>>>>>> Skype: callto:liels_bugs
>>>>>>
>>>>>> "It is our choices, that show what we truly are,
>>>>>> far more than our abilities." (A.P.W.B.D.)
>>>>> Ok, here i have a question for a special case, and alos an example, 
>>>>> that shows the difficults of right caching:
>>>>>
>>>>> On a list-view with a form, that filters the records, what is the 
>>>>> best technique that the search results will be cached ?
>>>> Store the search result somewhere. Possible locations are the 
>>>> session or a table. If you use a table you can share the results of 
>>>> popular searches across sessions.
>>>>
>>>> I thought about creating this for the indexed search. and I even 
>>>> thought about making a framework for other extensions. Maybe this is 
>>>> something for ECT?
>>>>
>>>> Masi
>>>>
>>>
>>> hi Masi,
>>>
>>> sounds interesting, i'm not sure how this can be handled. Storing 
>>> search results in a session is quite easy, but how do you present 
>>> this on pages ?
>>> Storing search result in tables may be a good idea, but it sounds 
>>> nearly same like creating cHashes for piVar-combinations.
>>>
>>> Can you give us a deeper thought of this ?
>>
>> For displaying the search result you need a USER_INT (possibly a 
>> USER_INT within a USER) so that no cHashes are needed. Cached are only 
>> the raw data, not the formatted output of the search. I don't think 
>> this is a big problem as a search result is usually shorted lived.
> 
> I still wonder how this should work.
> I have seen this link which has contents copied from a book by Dmitry 
> Dulepov:
> 
> http://www.wiki-oblady.com/index.php/Appeler_un_plugin_en_USER_INT_dans_un_plugin_en_USER 
> 
> 
> But does this really work? When a USER object is cached, then normally 
> the plugin is not called any more. But how should the USER_INT part of a 
> USER object be changed then?
> Maybe it has always to be one USER object called by a USER_INT object.
> 
>> The point is that you create from the search query (piVars) a key for 
>> the search result. Whenever the PI has to show a result it will look 
>> it up in the search cache. In this cache only the ids of the found 
>> rows are stored (sorted!).
> How do you implement a search cache?
> 
> 
> - Franz
> 

IMHO USER_INT as part of a USER object/extension will never be called 
again if cached!

Implement a search cache:

search form -> chash calculation -> result page

search form will "forward" to chash calculation which will "forward" 
with the calculated chash to the search result page which will be 
cachable due to chash enabled. BUT ensure that only this part will 
require chash otherwise NOTHING (=0=zero) gets cached. At least before 
4.3 ;-)
chash calculation could be done in TS on the fly.

Regs. Peter.

-- 
Fiat lux! Docendo discimus.
_____________________________
uon GbR

http://www.uon.li
http://www.xing.com/profile/Peter_Russ