Re: FFI, Preload, and performance

On 20.12.2019 at 23:08, Larry Garfield wrote: > The documentation doesn't really cover this, so I'm hoping it's OK to ask here. > > The FFI docs say that for web requests, you really really ought to use preloading and then > load/scope to manage an FFI library because it's going to be too slow to link up otherwise. > See the example here: > > https://www.php.net/manual/en/ffi.examples-complete.php > > What is unclear, though, is the performance impact of calling scope(). It happens at runtime > in the example so I presume it's "fast enough" at runtime. However, is it expected > that it's only ever called once for a given scope? Or would it be safe to call multiple times > in different places just to get an unshared reference to it? Basically, would this be safe or a bad > idea, and why? > > > final class PrintProxy { > private static $ffi = null; > function __construct() { > if (is_null(self::$ffi)) { > self::$ffi = FFI::scope("DUMMY"); > } > } > function printf($format, ...$args) { > return (int)self::$ffi->printf($format, ...$args); > } > } > > final class SprintfProxy { > private static $ffi = null; > function __construct() { > if (is_null(self::$ffi)) { > self::$ffi = FFI::scope("DUMMY"); > } > } > function printf($format, ...$args) { > return (string)self::$ffi->sprintf($format, ...$args); > } > } > > > The alternative being to have a single object that wraps scope("DUMMY") just once, > and then inject that into both of those classes. Is that worth the extra hassle or no, and why? From looking at the implementation[1], FFI::scope() looks like a pretty cheap operation; basically it's just looking up the scope from a HashTable, and constructing a thin wrapper object around that. But as always: when in doubt, measure! :) [1] <https://github.com/php/php-src/blob/php-7.4.1/ext/ffi/ffi.c#L3387-L3415> -- Christoph M. Becker