Does C30 optimisation break __prog__ pointers?

Hello All. We have a snippet of code that goes something like this:-

 #define CONFIG_START 0x2800  static const __prog__ Byte *rulePointer;  Byte getFirstRule() {      if (getAreaLength())      {          rulePointer = (const __prog__ Byte *)CONFIG_START;          return *rulePointer;      }       return 0; } 

In other words, there is some data stored in flash memory at 0x2800 onwards, and this bit of code reads the first byte. Debug builds consistently work correctly. Most release builds do not, and the 'return *rulePointer' statement resolves to machine instructions that fetch a byte from 0x2800 in *ram* not *flash*. I say most, because we can occasionally get release builds to work, if we include/exclude other blocks of data stored in flash memory. As yet, we haven't spotted a pattern in terms of what makes it work and what makes it fail. The MPU is the PIC24FJ128GB110. Current compiler version in v3.23, although the latest v3.31 behaves in the same way for us. Any suggestions would be welcome... Regards, A.

Try reserve space for "flash memory at 0x2800 onwards" only by a section at gld, like this or manual edit gld

That works a treat - thanks. A.

You can't use an integer for a pointer; sorry.  Take the address of a symbol.

It is never guaranteed in C that a literal value represents a value of a memory location and it almost never does for pointers that are __eds__, __prog__, __psv__, __pmp__ or __external__ since there needs to be alot more information carried in the pointer than an offset from 0.

Regards
Calum