Re: com php-src: Fix bug #67436: Zend/zend_compile.c

On Mon, Jun 16, 2014 at 1:40 AM, Bob Weinand <bobwei9@hotmail.com> wrote: > Am 16.6.2014 um 01:29 schrieb Ferenc Kovacs <tyra3l@gmail.com>: > > On Mon, Jun 16, 2014 at 1:25 AM, Bob Weinand <bobwei9@hotmail.com> wrote: > >> Am 16.6.2014 um 01:21 schrieb Ferenc Kovacs <tyra3l@gmail.com>: >> >> On Mon, Jun 16, 2014 at 12:27 AM, Bob Weinand <bobwei9@hotmail.com> >> wrote: >> >>> Am 16.6.2014 um 00:11 schrieb Stas Malyshev <smalyshev@sugarcrm.com>: >>> > Hi! >>> > >>> >> Commit: ca29063da5599f463454578abe629bc21ab33b3b >>> >> Author: Bob Weinand <bobwei9@hotmail.com> Sun, 15 Jun >>> 2014 22:18:25 +0200 >>> >> Parents: b053c4f4f6db57c7687e700de7aa4abadef2c128 >>> >> Branches: PHP-5.4 >>> >> >>> >> Link: >>> >>> http://git.php.net/?p=php-src.git;a=commitdiff;h=ca29063da5599f463454578abe629bc21ab33b3b >>> >> >>> >> Log: >>> >> Fix bug #67436 >>> >> >>> >> Bugs: >>> >> https://bugs.php.net/67436 >>> > >>> > Could you please explain this fix? The bug is about autoloader, but the >>> > fix seems nothing to do with it. Also, adding a test case would be >>> > really good, bugfixes without test cases tend to regress later. >>> > >>> > -- >>> > Stanislav Malyshev, Software Architect >>> > SugarCRM: http://www.sugarcrm.com/ >>> > (408)454-6900 ext. 227 >>> >>> Levi wanted to provide some tests. I'm sure they're ready in a few >>> minutes.... >>> >>> And yes, bit complicated. >>> >>> The issue is that we cannot invoke the autoloader during compile time >>> (where early binding is done, when class "a" was loaded before). (That's >>> all what it had to do with autoloaders) >>> And so the class "c" (in that example) is not defined. >>> >>> In this case the class "c" was (tried to be) fetched by this >>> zval_update_constant_ex which was invoked there by >>> zend_get_function_declaration called by do_inheritance_check_on_method as >>> the function signatures didn't match (E_STRICT error). >>> >>> So, as the autoloader couldn't be invoked (compile-time...), but the >>> class constant was tried to be resolved, we ended up with a fatal error >>> instead of the notice. >>> >>> Logical fix: don't try to resolve the class constant, but just write the >>> constants name there. (which is maybe even more informative to the user >>> than the value) >>> >>> >>> Bob >> >> >> hm, >> I thought this was fixed with >> 1µ¯ùuÊg,tè& >> æÃhttp://git.php.net/?p=php-src.git;a=commit;h=f0c926564c5f7de9462d9ca7bd75014b14a63f56 >> >> -- >> Ferenc Kovács >> @Tyr43l - http://tyrael.hu >> >> >> Nope, that was some totally unrelated bug. >> >> That bug occurred during an error handler at run-time, this bug here >> during compile-time. >> >> Bob >> > > did you read https://bugs.php.net/bug.php?id=65322 ? > it was caused by a compile time error triggered by mismatching class > definition, and that the error handler called for that won't be able to > trigger the autloader (we also tracked it back to the "if (!use_autoload || > zend_is_compiling(TSRMLS_C)) {" check). > maybe I'm missing something but this seems like the exact case here. > > -- > Ferenc Kovács > @Tyr43l - http://tyrael.hu > > > No. Still two different things. Bug 65322 is caused inside the error > handler, inside zend_error (the error was that compilation mode wasn't > turned off) while bug 67436 happens during compilation time, before > zend_error is reached. > > Bob > right, I was too quick to read the attached testscripts for 67436, I missed that in this case the autoloader is triggered by the class definition/resolving the ancestors while in 65322 it was triggered by the error handler caused by the error in the error in the incompatible class definition. -- Ferenc Kovács @Tyr43l - http://tyrael.hu