find_and_load() may load different file than stat()'ed

[Martin Husemann]

Hi folks,

I found a reproducable way in the NetBSD build when creating info for
a cross-gdb (this is optional when cross compiling NetBSD, so the problem
went unnoticed some time). I've reported the bug as a NetBSD PR:

  http://www.netbsd.org/cgi-bin/query-pr-single.pl?number=26791

I think what happens is that makeinfo/node.c:split_file() uses stat() to
determine the file_size variable and then calls find_and_load() with use_path
= 1, and this finds a different sized (older) file with the same name (due to
the way makeinfo is invoked). Later, when accessing data beyound the loaded
size.

I worked around it by changing the call to find_and_load() to not use the
path (see attached patch). I suspect the right fix, however - or something
that should be done additionally - would be to change find_and_load() to 

  char *find_and_load (char *filename, int use_path, off_t *loaded_size);

and while there clean up all mis-uses of "long" for file size values ;-}


Martin