Re: RE: [RFC] Fast Parameter Parsing API

Hey: Sorry for top post, first. I am going to talk about a thought I have.. we can reuse the arginfo as the only key struct for function arguments declaration/parsing.. something like, take strlen as example: declaring: ZEND_ARG_INFO(strlen_arginfo, 0, 0, 1) Z_ARG_STR() ZEND_ARG_INFO_END(); then when doing register internal function. Zend vm will bind the Z_ARG_STR to int num_args = 0; (zend_internl_functions *)(strlen_function)->argument_handlers[num_args++] = zend_parse_parameter_str; the zend_parse_parameter_str here is the handler only handling with str(which will be effective) then while in zif_strlen(), only needs call : zend_string *str; zend_parser_parameters(stren_arginfo, &str); the only thing I am not sure now is, if there are more than 1 arguments of one function, let's say it's 5.. this implementation will doing 5 functions call..... which maybe slower than previous zend_parse_parameter thoughts? thanks On Tue, May 27, 2014 at 10:06 AM, Stas Malyshev <smalyshev@sugarcrm.com> wrote: > Hi! > >> zpp current usage is much more readable to me and less error prone >> while writing (clang plugin helping to check the syntax, see Johannes >> tools). > > Not really - I have to go to the docs each time to remember what \ or / > are and what is the difference between H and h or z and Z. And putting a > wrong arg type is extremely easy. I've just learned about clang thing > this month, so I'm pretty sure I can count number of people who use it > on one hand, and it doesn't seem to be mentioned anywhere in the docs. > >> May I point you to my questions in this thread about extending the >> reflection struct, add the necessary info (types) and maybe do the >> naming args while being at it? > > This would be great (again, laying ground for named args support) but > I'm not sure how to do it. We have a bit of a contradiction here since > the speedup requires unrolled code (that's what the current ideas do) > and reflection requires static data struct. I'm not sure how to make > both work. Maybe with some kind of function pointers in the data struct > for parsing? But I'm afraid then it would lose the speedup, since it > won't be inlined unrolled code anymore. > -- > Stanislav Malyshev, Software Architect > SugarCRM: http://www.sugarcrm.com/ > (408)454-6900 ext. 227 -- Xinchen Hui Laruence Consultant at Zend http://www.laruence.com/