Re: Re: How do I save TSRMLS_CACHE for a new thread in PHP7?

On Thu, 26 Nov 2015 12:22:23 +0500 Ruslan Osmanov <rrosmanov@gmail.com> wrote: > On Wed, 25 Nov 2015 22:45:29 +0100 > Ángel González <keisial@gmail.com> wrote: > > > On 25/11/15 20:25, Ruslan Osmanov wrote: > > > I'll try to explain the problem. > > > > > > I have a function(let's call it execF), which is > > > passed to a > > > library(libeio) as a callback. This function is invoked afterwards > > > by the library in separate thread. > > > > > > The function is calling Zend API, particularly ZVAL_STRINGL(). The > > > Zend API uses TSRMLS_CACHE internally. For instance, > > > ZVAL_STRINGL() allocates zend_string which implies using AG() > > > macro. The AG() macro relies on TSRMLS_CACHE. > > > > > > Since TSRMLS_CACHE becomes invalid in new thread, the AG() macro > > > (as well as other macros relying on TSRMLS_CACHE) fails with > > > segmentation fault. > > Why are you allocating a string? Are you expecting to later use it > > on the php request? > > What's the goal of your extension? I suspect you may have an > > earlier misstep. > > I'm allocating a string for zend_fcall_info.function_name: > https://goo.gl/MYoJcw (just like zend_call_method). Called by > php_eio_custom_execute(). > > Well, I can pre-allocate zend_string for the callback. Then a few > lines further I'm trying to fetch EG(function_table), which > eventually refers to TSRMLS_CACHE. Since the latter is invalid, I > catch a segmentation fault, again. > > So I'm asking how do I save the context for a thread created by libeio > library. > > The flow looks like the following: > > 1. User passes a callable to eio_custom() > 2. The callable is saved as libeio user custom data(void*): > eio_custom(php_eio_custom_execute, pri, php_eio_res_cb_custom, > eio_cb); 3. User calls eio_event_loop() > 4. Libeio invokes php_eio_custom_execute() *in separate thread* > 5. php_eio_custom_execute() unsuccessfully tries to use Zend API By the way, TSRMLS_SET_CTX and TSRMLS_FETCH_FROM_CTX macros look redundant. They're making just a pointer assignment which is scarcely useful in context of thread-local variable in global scope. -- Ruslan Osmanov

Attachment: [application/pgp-signature] signature.asc