https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82952
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Keywords| |compile-time-hog
Status|UNCONFIRMED |NEW
Last reconfirmed| |2017-11-13
CC| |jakub at gcc dot gnu.org
Ever confirmed|0 |1
--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed. I suspect UBSAN creates a large number of branches and
-Wduplicated-branches isn't very effective. In fact it looks quadratic given
we do
128 if (warn_duplicated_branches)
129 walk_tree_without_duplicates (&DECL_SAVED_TREE (fndecl),
130 do_warn_duplicated_branches_r, NULL);
and do_warn_duplicated_branches_r does itself
/* Compute the hash of the then branch. */
inchash::hash hstate0 (0);
inchash::add_expr (thenb, hstate0);
hashval_t h0 = hstate0.end ();
...
&& !walk_tree_without_duplicates (&thenb, expr_from_macro_expansion_r,
NULL)
thus we process each tree (in COND_EXPRs) a quadratic amount of times.
inchash is very likely also not avoiding walking duplicates multiple times,
as SAVE_EXPR is tcc_expression it even walks those multiple times.
As a band-aid I suggest to limit the depth we walk here somehow...
