Ping: Andrew, did I answer your questions? Do you (or anyone else)
have any other comments on the latest patch below?
https://gcc.gnu.org/pipermail/gcc-patches/2021-August/577865.html
On 8/20/21 4:16 PM, Martin Sebor wrote:
On 8/20/21 7:09 AM, Andrew MacLeod wrote:
On 8/19/21 7:09 PM, Martin Sebor via Gcc-patches wrote:
The attached patch changes the new access warning pass to use
the per-function ranger instance. To do that it makes a number
of the global static functions members of the pass (that involved
moving one to a later point in the file, increasing the diff;
the body of the function hasn't changed otherwise). Still more
functions remain. At the same time, the patch also enables
the simple pointer_query cache to avoid repeatedly recomputing
the properties of related pointers into the same objects, and
makes the cache more effective (trunk fails to cache a bunch of
intermediate results). Finally, the patch enhances the debugging
support for the cache.
Other than the ranger/caching the changes have no user-visible
effect.
Why are you calling enable/disable ranger if you are passing a ranger
instance around instead of using the get_range_query (cfun)->range*
calls?
The pass stores an instance of the pointer_query class which in
turn stores a pointer to range_query (which is a copy of the ranger).
So storing it also in pass_waccess isn't necessary and can be
removed. I've made that change in the attached update. I'm not
sure the corresponding pointer should at some point also be removed
from the pointer_query class and replaced by calls to get_range_query
(cfun). If so, that would take some surgery to the strlen pass which
also uses pointer_query and isn't quite ready to make this switch.
Are you planning to transition to using the get_range_query()
interface instead of keeping a range_query pointer in the
pointer_query class?
This pass and to a smaller extent the pointer_query class that's
used by it and the strlen pass are still a work in progress.
I also still need to convert the strlen pass to use Ranger and
I expect it will take some changes to pointer_query. So at that
point, if going through get_range_query (cfun) everywhere is what
you recommend, I'm happy to do it.
Anyway, attached is an updated revision with the m_ranger member
removed and a few helpers changed to take a range_query argument
to use the pointer_query member instead. It was retested on
x86_64-linux.
Martin
PS There has been an effort to get rid of global variables from GCC,
or, as the first step, to avoid accessing them directly(*). If and
when that happens, it seems like each pass will have to store either
the ranger instance as a member (directly or indirectly, via a member
of a class that stores it) or the function passed to pass::execute()
if it wants to access either.
[*] https://gcc.gnu.org/pipermail/gcc-patches/2021-June/573975.html
The patch at the link above wasn't approved but IIUC removing globals
from GCC is still a goal.
