Re: Changing PECL signup flow.

On September 18, 2019 3:32:38 PM GMT+02:00, "Christoph M. Becker" <cmbecker69@gmx.de> wrote: >On 18.09.2019 at 15:22, Johannes Schlüter wrote: > >> The Windows builds are also a relevant factor in the discussions >> regarding the PECL site. Afaik (Anatoliy and others know better than >> me) packages uploaded to the pecl site are being built in the same >> setup as our official PHP runtime Windows builds. > >The PECL builds are done on another VM than the Windows builds. Good to know. If not done another got thing is if each build runs in its own container or from a clean VM snapshot or whatever capabilities there are for Windows ... Anyways different topic in itself. Just came to mind while thinking about the pickle tool. And possible future paths. One thought in my head (also away from the initial question) is also whether we could provide Linux builds - since removing TSRM it came closer to being feasible to have a stable common interface for 95% of extensions. (There are the ones which go deep ... but for most the only thing they have to interact with is a zval which is relatively stable and a class_entry+zend_object where most things should be wrapable behind accessor functions with stable API) If we just need one single build with static dependencies covering most systems (some (g)libc dependency ... while glibc has versions APIs and quite good compatibility these days) this could cover the 80/20 case. Or is ffi the big push for those 95% making pecl actually mostly obsolete, solving the whole thing that way round? (I admit - I never looked at FFI) Then only extensions "going deep" (xdebug, mysqlnd stuff, runkit-like, ...) need some distribution channel, while ffi code could use packagist/composer. johannes