Re: [RFC] Zend Signal Handling

How shall we continue with this patch? Arnaud, I need an updated HEAD patch to match the changes made in 5.3. There's valid concern about performance in ZTS, which seems hard to address without the __thread changes. It seems you are working on to get this into 6.0. Dmitry, Lukas, Arnaud and others interested, would it make sense to proceed with the current non-ZTS signals implementation in 5.3 and work on ZTS in HEAD with a goal to improve performance? -lucas On 8/14/08 4:22 PM, "Arnaud Le Blanc" <arnaud.lb@gmail.com> wrote: > On Thursday 14 August 2008 02:59:48 Arnaud Le Blanc wrote: >> Hi, >> >> I made some changes to the patch: >> >> >> http://arnaud.lb.s3.amazonaws.com/php-5.3.0-alarms-0808141122.patch >> >> - Apache effectively seems to resets the signals after MINIT, so original >> handlers are now saved in RINIT in the first request of the process. >> - Removed some unneeded blocks/unblocks. >> - Changed all added TSRMLS_FETCH() so that they affect only ZTS+ZEND_SIGNAL >> builds. >> - Removed some added TSRMLS_FETCH() (all in zend_alloc and a few in >> zend_hash). >> >> Regards, >> >> Arnaud > > > I made an other patch that caches tsrm_ls in a thread-local variable when > "calling" TSRMLS_FETCH(), on platforms that support the __thread specifier: > > http://arnaud.lb.s3.amazonaws.com/tsrm_ls_cache.patch > > This gives a 10 to 15% speedup to ZTS builds on bench.php. It also makes > ZEND_SIGNAL+ZTS builds run as fast as ZTS without ZEND_SIGNAL. > > Regards, > > Arnaud