https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91307
--- Comment #8 from Richard Biener <rguenth at gcc dot gnu.org> ---
I wonder why we can't simply use .init_array and thus get away with a
local symbol on targets that support this. Uh, so the symbol is already
local but we're keeping it in our stripping process it seems. And
the name is fancy "for debugging purposes". Not sure if gdb has special
breakpoints for ctors/dtors or if the user is expected to know the
mangling.
Btw, for !targetm.have_ctors_dtors we're appending "some randomness"
anyways so I wonder if we shouldn't simply pass down appropriate
deterministic randomness by producing it from the symbols of the individual
ctors/dtors we call here:
/* Find the next batch of constructors/destructors with the same
initialization priority. */
for (;i < j; i++)
{
tree call;
fn = cdtors[i];
call = build_call_expr (fn, 0);
if (ctor_p)
DECL_STATIC_CONSTRUCTOR (fn) = 0;
else
DECL_STATIC_DESTRUCTOR (fn) = 0;
/* We do not want to optimize away pure/const calls here.
When optimizing, these should be already removed, when not
optimizing, we want user to be able to breakpoint in them. */
TREE_SIDE_EFFECTS (call) = 1;
append_to_statement_list (call, &body);
}
like
Index: gcc/ipa.c
===================================================================
--- gcc/ipa.c (revision 274527)
+++ gcc/ipa.c (working copy)
@@ -827,7 +827,7 @@ ipa_discover_variable_flags (void)
static void
cgraph_build_static_cdtor_1 (char which, tree body, int priority, bool final,
tree optimization,
- tree target)
+ tree target, hashval_t hstate)
{
static int counter = 0;
char which_buf[16];
@@ -842,7 +842,7 @@ cgraph_build_static_cdtor_1 (char which,
/* Proudce sane name but one not recognizable by collect2, just for the
case we fail to inline the function. */
sprintf (which_buf, "sub_%c_%.5d_%d", which, priority, counter++);
- name = get_file_function_name (which_buf);
+ name = get_file_function_name (which_buf, hstate);
decl = build_decl (input_location, FUNCTION_DECL, name,
build_function_type_list (void_type_node, NULL_TREE));
@@ -973,6 +973,7 @@ build_cdtor (bool ctor_p, const vec<tree
}
/* Find the next batch of constructors/destructors with the same
initialization priority. */
+ inchash::hash hstate;
for (;i < j; i++)
{
tree call;
@@ -987,13 +988,15 @@ build_cdtor (bool ctor_p, const vec<tree
optimizing, we want user to be able to breakpoint in them. */
TREE_SIDE_EFFECTS (call) = 1;
append_to_statement_list (call, &body);
+ hstate.merge_hash (IDENTIFIER_HASH_VALUE (DECL_NAME (fn)));
}
gcc_assert (body != NULL_TREE);
/* Generate a function to call all the function of like
priority. */
cgraph_build_static_cdtor_1 (ctor_p ? 'I' : 'D', body, priority, true,
DECL_FUNCTION_SPECIFIC_OPTIMIZATION
(cdtors[0]),
- DECL_FUNCTION_SPECIFIC_TARGET (cdtors[0]));
+ DECL_FUNCTION_SPECIFIC_TARGET (cdtors[0]),
+ hstate.end);
}
}
plus appropriate changes to get_file_function_name.
It also seems to me we may not need to produce a unique symbol here
so Martins original idea of using "wpa" for LTO might work fine
(but I'd change the signature of get_file_function_name to pass in
the actual "file name" then, keeping the original use of main_input_filename
via an overload).
Note with !targetm.have_ctors_dtors we are going to produce
either symbol clashes or random symbols (with and without -flto,
"conveniently" IPA CDTOR merging is only run for -flto or
!targetm.have_ctors_dtors for whathever reasons).
