Re: [ZEND-ENGINE-CVS] cvs: TSRM / tsrm_virtual_cwd.h

At 02:06 AM 12/11/2002 -0500, Ilia A. wrote:

On December 11, 2002 12:16 am, Andi Gutmans wrote: This is not the correct patch but adds other bugs. You should check for UNC separetely in the code like in virtual_realpath(). IS_ABSOLUTE_PATH() is only meant to detect the drive letter and is always followed by use of COPY_WHEN_ABSOLUTE() (which happens to also be 2 but it's still not right. Stuff might break). The patch only affects the behavior of IS_ABSOLUTE_PATH on Win32 where UNC path is absolute. PHP's behavior is the same when either IS_ABSOLUTE_PATH or IS_UNC_PATH occur. Separation of the two, only adds another len >= 2 check, beyond that nothing else changes. I do not see how this can break any existing code on Win32 platform, perhaps you could offer me some hints about situations where this may cause problems? I've looked at the usage of IS_ABSOLUTE_PATH macro and it is only used 4 times outside of TSRM/tsrm_virtual_cwd.c, where it only used in 2 functions. From what I've been able to see the change does not break any existing code. It can however prevent problems for paths on Win32 that begin with \\ or //, since unless the UNC check is explicitly done the code will try to 'resolve' those paths breaking them in the process.

I didn't mean it'll break things right away. I'm just saying the fact that by chance sizeof("c:") == sizeof("\\") should not be taken advantage of by the code (IMO). It means the code doesn't stay very clean and if we ever make a change there things might break. I think we should change COPY_WHEN_ABSOLUTE(path) to patch_copy[1] == ':' ? 2 : 2 and then we'll be ready for making future changes and it'll be obvious how it works to people reading the code instead of relying on a coincidence. The compiler will even optimize this (most probably). Andi