2015-04-24 12:59 GMT+03:00 Johannes Schlüter <johan...@schlueters.de>:
> On Fri, 2015-04-24 at 09:16 +0300, Arvids Godjuks wrote: > > May I question the sanity of the words written in this email? :D (it's a > > joke). > > > > The whole point of mysqlnd drivers and other improvements was to cut down > > on data copying, improving performance and doing a lot of other stuff. > > Moving PDO to a PHP implementation will kill it all: preformance will > > suffer, memory usage will skyrocket, dealing with charsets - I don't even > > wana pretend I understand how to deal with that part in a proper fasion. > > Doesn't it require access to internal PHP api's to do a lot of what PDO > and > > other native drivers do? > > Well, the Zephyr could pitch in here, MAYBE, depending on how good it > > actually is and what it can do, but still, it feels more like a cruch to > me. > > In many many different topics I stressed that we should do things in > userland and use extensions only when needed for performance. > > Doing things in userland gives > * better debugability > * better understanding for users > * lower entry barrier > * faster development time > * better ability to evolve (different library versions can coexist > on a system, for old and new code) > > With the changes in PHP 7 this is viable for even more areas. What PDO > does is very thin. And mind: If a user truly cares about the last bit of > performance they won't use an abstraction, they use DBMS-specific SQL > and cut out abstractions. But for many cases that level of performance > matters, as usage of ORMs etc. show. Also mind that processing in the > database server and network traffic probably cost way more time than a > thin wrapper even in userspace (in real life, not in "SELECT 1") > > johannes > > > It may have a merit to try it out, and as I said - Zephyr is an interesting idea to try - may work quite well considering.
