Re: [RFC] Revert/extend/postpone original RFC about read_only, lazy_write sessions

On Thu, Mar 20, 2014 at 5:19 AM, Andrey Andreev <narf@devilix.net> wrote: > Hi all, > > I did this prototype/concept about SessionHandler, > SessionHandlerInterface in PHP 5.6: > https://gist.github.com/narfbg/9643416 > In the context of this discussion, the important part is the > updateTimestamp() method (in relation to 'lazy_write'), you should > ignore the rest (for now). I want to discuss it here first before > updating the RFC with it. > > So, I had a chat with Yasuo about it to identify the problems and/or > limitations and this is the most important one: > > The patch (https://github.com/php/php-src/pull/628) doesn't allow > calling parent::updateTimestamp() and users should call > parent::write() instead. This is because there are PECL extensions > that provide a session handler and they won't have support for that, > which is hard to detect at runtime. > However, I don't think that this is convenient, nor that documenting > this is better than documenting that PECL extensions might not support > the parent::updateTimestamp() call. To me, that's an unnecessary > limitation and here are the alternatives: > > 1. Don't declare SessionHandler::updateTimestamp() if > session.save_handler doesn't support it and put a big red warning in > the docs that PECL extensions might lack the method. > > I think this is a good trade-off, given that most users trust (or > have) only "stock" PHP extensions and can benefit from this. > Also, this wouldn't be a precedent - you can take mysqli and it's > dependancies on mysqlnd for example (where the docs mention this very > briefly). > > 2. Scrap the updateWhatever() method and just modify write() to only > update the timestamp when necessary. > > I have no idea why nobody has thought about this previously. The > 'lazy_write' idea is inspired by a feature request from 2002 > (https://bugs.php.net/bug.php?id=17860) that simply asks for an API > identifying whether $_SESSION has been changed or not, or in other > words - and API saying whether session data or only the timestamp > should be updated. > > Now, Yasuo prefers providing explicit API for the lazy_write feature > (namely, the updateTimestamp() method), but simply providing such a > flag via i.e. yet another magic constant, similar to SID, solves ALL > of the problems: > > - No API changes > - No BC breaks > - No effect on PECL extensions that don't support it > > IMO, that's way more flexible and would allow doing lazy_write by > default, no option - after all, the feature represents a performance > improvement. > It also solves another, side-effect problem: the > SessionUpdateTimestampInterface, that currently exists only because of > internal implementation details. > > Thoughts? More suggestions? Concerns? > > Cheers, > Andrey. > > -- > PHP Internals - PHP Runtime Development Mailing List > To unsubscribe, visit: http://www.php.net/unsub.php > > Just an unimportant note on the name of the function. Having done this in a custom session handler, I have always thought of this as an analogue to the "touch" command for the filesystem. Naming this SessionHandler::touch() would probably appeal to those familiar with linux or other posix compliant systems. http://pubs.opengroup.org/onlinepubs/9699919799/utilities/touch.html Thanks, John