how can 'the compiler' always warn you??

There is no 'law' staing that the use of the buffer and the ISR has to be in the same module. If the compiler did warn when the two were in the same module, the web would be flooded with complaints "the compiler did not warn" all of which would be referred to the above. The next post would be "the tools should be fixed". How do you fix soemthing that can not be detected by "machine means".
It is extremely dangerous, with all the idiots out there to implement anything that can be misconstrued as malfunctioning by anyone under any circumstances.
I can not count how many posts with the heading "compiler error" re Keil I have seen that ALL have been due to the one reporting being cocksure that it could not in any way be his code that was the problem. I can recall NONE of those that actually were a 'compiler error'. Now, for certain other compilers, that is another story.

If someone can not figure out how to handle atomicity, that someone should choose a chip with a data word as long as the longest entity (s)he uses (e.g. if no longs but ints use minimum 16 bit chip, if longs - use 32 bit chip....).

Jan, this is not directed at you, but so many (PC) programmers want the tools to alleviate them from the horrendous task known as thinking.

Erik