Re: Re: svn: /php/php-src/ branches/PHP_5_4/Zend/zend_execute_API.c trunk/Zend/zend_execute_API.c

On Tue, Feb 7, 2012 at 1:47 AM, Stas Malyshev <smalyshev@sugarcrm.com> wrote: > Hi! > > >> there comes a new bug relate to this change,  #60978,   in php_cli.c > > > I'm not sure this bug is critical enough to keep messing with the code base > this close to the release. Could you please describe what is the current > status, what was before the fix and what exactly the problems are here? Hi Stats, let me try to describe the whole thing. before any change. the php -r implemented like: exit_code = 0; zend_try { if ( zend_eval_stringl() == FAILUE) { [1] exit_code = 254; [2] } } zend_end_try(); if (!exit_code) { exit_code = EG(exit_code); [3] } in zend_eval_stringl: op_array = compile(*); [4] zend_execute(op_array); [5] free(op_array); return SUCCESS; then there comes a bug, if longjmp(via die, exit or zend_bailout directly) called in [5] (zend_execute), the [4] (op_array) will be leaked. so dsp change the zend_eval_stringl to: op_array = compile(**); zend_try { zend_execute(op_array); } zend_end_try(); free(op_array); return SUCCESS; then Derick found this make his xdebug code won't works any more. in xdebug, it expect zend_eval_stringl longjmp when error occurred. but after dsp's fix, obviously, zend_eval_stringl will not jmp anymore. so Derick change the codes to : op_array = compile(**); retval = SUCCESS; zend_try { zend_execute(op_array); } zend_catch { retval = FAILURE; [6] } zend_end_try(); free(op_array); return retval; let look at [1], before any change, a call to exit(or die) in a eval string, will let the process goto [3], then make the exit_code eqaul to EG(exit_code). after Derick's fix [6], a call to exit(or die) in a eval string will lead the process goto [2], which cause the exit_code always be 254 [2]. so I proposal change the fix to: op_array = compile(**); zend_try { zend_execute(op_array); } zend_catch { free(op_array); zend_bailout(); } zend_end_try(); free(op_array); return SUCCESS; this change should can meet all the needs... and won't break any BCs. do I describe this clearly ? thanks > -- > Stanislav Malyshev, Software Architect > SugarCRM: http://www.sugarcrm.com/ > (408)454-6900 ext. 227 -- Laruence  Xinchen Hui http://www.laruence.com/