fix ora-03131 on bigendian 64bit machines #7422
Conversation
out-variables cannot be used on bigendian 64bit machines and cause ORA-03131 error This is caused by a uc4-value (length of OCI8-data) stored into the upper half of a sizte_t-value (length of zend_string).
ext/oci8/oci8_statement.c
Outdated
| @@ -1500,6 +1500,10 @@ sb4 php_oci_bind_out_callback( | |||
|
|
|||
| /* XXX we assume that zend-zval len has 4 bytes */ | |||
| *alenpp = (ub4*) &Z_STRLEN_P(val); | |||
| if(sizeof(size_t)==2*sizeof(ub4)){ | |||
There was a problem hiding this comment.
A solution using the config.m4 macro PHP_C_BIGENDIAN would be better.
There was a problem hiding this comment.
Right, this can check #ifdef WORDS_BIGENDIAN.
There was a problem hiding this comment.
I'm using WORDS_BIGENDIAN now.
Are there bigendian-machines with sizeof(size_t)==sizeof(ub4) or sizeof(size_t)==4*sizeof(ub4) ?
Hence I'm increasing the ub4-pointer only if sizeof(size_t)==2*sizeof(ub4)
There was a problem hiding this comment.
@pk1234, I don't know the answer to that question, sorry. Overall I still feel this if test needs to be done at compile time and a macro defined. See how AC_COMPILE_IFELSE is used in other configure files. Sorry I don't have access to a big endian machine to check this.
There was a problem hiding this comment.
@cjbj, current code does not work at all on 64bit bigendian machines. So everything that changes the behaviour on bigendian systems only, will be better than what have now.
#ifdef WORDS_BIGENDIAN makes sure that the proposed changes do not influence littleendian systems.
Creating a new macro that will be defined if and only if sizeof(size_t)==2*sizeof(ub4) seems to be too much effort to me.
sizeof(size_t)==2*sizeof(ub4) can be evaluated at compile time and the optimizer will either do or remove the increment-statement unconditionally. That's exactly the effect of a #ifdef SIZE_T_HAS_TWICE_THE_SIZE_OF_UB4-macro
I therefore feel, that my combination of #ifdef WORDS_BIGENDIAN and if-statement is the best way to go without lots of bigendian/littleendian test-machines.
instead of reading first 4 bytes of (size_t)1 to detect bigendieness use WORDS_BIGENDIAN macro
another typo - sorry for 3 commits
|
@pk1234 I'm out for a few days. I'll get back to this next week. |
|
@cjbj no problem |
|
Is there anything I can do? Peter |
OCI8 stores the length of retrieved data as a uc4-value into the first 4 bytes of the length of a zend_string (size_t). This fails on bigendian 64bit machines where the length must be stored into the LAST 4 bytes of zend_string-length.
This patches fixes the problem by increasing the *uc4-pointer if the first 4 bytes of (sizte_t)1 are equal to 0.