From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: by sourceware.org (Postfix, from userid 48) id DC3B93858D39; Thu, 8 Sep 2022 09:28:29 +0000 (GMT) DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org DC3B93858D39 DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=sourceware.org; s=default; t=1662629309; bh=PkaJsqh+27vM71QKpC10c5E0m/FOt4nNZiICrjOyTCI=; h=From:To:Subject:Date:In-Reply-To:References:From; b=kfPUsYdrsBsPotLRKOiLLyi+7WyR5GUx2nresntCqyP89vnRkXB/+w0f5zSuOAgdK SON/cd7KuHWSnUa77wwUe8Gb5ZnqIG9mMjyBF6fuvgzc88SbjYsZXZdESLVJr4kw2V 4cgEFILx5CsJbuxtXikV2gTB9uPMPAXmRa6rJKqU= From: "gprocida at google dot com" To: libabigail@sourceware.org Subject: [Bug default/29464] abidw performance regression on vmlinux Date: Thu, 08 Sep 2022 09:28:29 +0000 X-Bugzilla-Reason: CC X-Bugzilla-Type: changed X-Bugzilla-Watch-Reason: None X-Bugzilla-Product: libabigail X-Bugzilla-Component: default X-Bugzilla-Version: unspecified X-Bugzilla-Keywords: X-Bugzilla-Severity: normal X-Bugzilla-Who: gprocida at google dot com X-Bugzilla-Status: ASSIGNED X-Bugzilla-Resolution: X-Bugzilla-Priority: P2 X-Bugzilla-Assigned-To: dodji at redhat dot com X-Bugzilla-Target-Milestone: --- X-Bugzilla-Flags: X-Bugzilla-Changed-Fields: Message-ID: In-Reply-To: References: Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: quoted-printable X-Bugzilla-URL: http://sourceware.org/bugzilla/ Auto-Submitted: auto-generated MIME-Version: 1.0 List-Id: https://sourceware.org/bugzilla/show_bug.cgi?id=3D29464 --- Comment #6 from gprocida at google dot com --- It turns out that there is now a new infinite loop in the XML writer when compiled with Clang. This is our old "friend" where containers have inconsistent equality and hash functions. In particular: Commit 7ecef636 "Canonicalize DIEs w/o assuming ODR & handle typedefs transparently" introduced new sets and a map in the XML writer to track referencing, emission and type ids for "non canonicalized" types as a separ= ate category. It added non_canonicalized_type_equal and non_canonicalized_type_hash funct= ion objects. The former is a deep equality and the latter hashes the non-intern= al name. The net effect is that I see a typedef being repeatedly emitted in a tight loop. It's recorded as emitted but not found when checked to see if it was emitted. The last time these issues occurred they were eventually resolved by using = much safer and more efficient bare pointers as the keys to the containers. I don't understand why new sets and maps were added or why they need to have bespoke equality and hash functions. --=20 You are receiving this mail because: You are on the CC list for the bug.=