Re: backtrace

At 21:22 31-08-00, Sascha Schumann wrote:

On Thu, 31 Aug 2000, Zeev Suraski wrote: At 18:33 31-08-00, Sascha Schumann wrote:

On Thu, 31 Aug 2000, Stanislav Malyshev wrote:

j>> called with a null list of var_args, which is causing it to barf. j>> Problem is, the variable orig_args is being cleared on line 369 j>> (va_end()), and j>> then being reused on line 444. Sorry, don't see any va_end in current CVS.

    The old_args stuff was removed some weeks ago (if we are
    talking about main/main.c). The author of the original code
    did not seem to be aware of the correct va_args use.

Actually, the original version was quite alright, except for under very weird platforms, which define va_list as a type that cannot be copied (probably some sort of an array).

    It was not necessary at all to copy it. That is where I saw
    the basic misunderstanding. A va_list can be traversed
    multiple times without copying it.

A va_list cannot be traversed multiple times, which is why the code was originally there (I think). What you meant is that if you send a va_list to a function (by value), that function can obviously not modify it. Thus, the code in php_error was redundant.

    And if there would have been any need to copy it, the macro
    va_copy should have been used.

va_copy() is not portable.

    But admittedly, while fixing the standard compliance, I broke
    some other odd configurations :(

Many platforms that define va_list as a simple pointer, reset it to NULL on va_end(), which makes it useless afterwards. Others don't do anything. Anyway, shit happens, this time it happened to both of us, but I hope we got it right this time (even though theoretically, it could still be broken under some implementations, but I don't think such implementations exist). Zeev -- Zeev Suraski <zeev@zend.com> http://www.zend.com/