From: Theodore Ts'o Subject: Re: [PATCH 1/4] e2fsprogs/libext2fs: replace ext2fs_free_inode_cache() argument Date: Tue, 11 Nov 2014 11:12:11 -0500 Message-ID: <20141111161211.GB17027@thunk.org> References: <1415689169-19393-1-git-send-email-wangxg.fnst@cn.fujitsu.com> <20141111081451.GD10043@birch.djwong.org> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Xiaoguang Wang , linux-ext4@vger.kernel.org To: "Darrick J. Wong" Return-path: Received: from imap.thunk.org ([74.207.234.97]:51664 "EHLO imap.thunk.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751264AbaKKQTM (ORCPT ); Tue, 11 Nov 2014 11:19:12 -0500 Content-Disposition: inline In-Reply-To: <20141111081451.GD10043@birch.djwong.org> Sender: linux-ext4-owner@vger.kernel.org List-ID: On Tue, Nov 11, 2014 at 12:14:51AM -0800, Darrick J. Wong wrote: > Ideally you'd change the old function to return an error code, but it returns > void... sigh. I guess client programs are on their own. > > Really, free_inode_cache is a poor interface since the ctor doesn't return a > struct ext2_inode_cache * directly, preferring to attach it to fs->icache > instead. Why are the inode cache ctor/dtor exported in ext2fs.h anyway? > Nobody seems to use them. Ted? Looking at the history, the change happened in commit 603e5ebc, which changed the internal representation of the ext2_inode_cache. At that point, I moved the inode cache freeing function from lib/ext2fs/freefs.c to lib/ext2fs/inode.c, on the theory that it made more sense to keep all of the code that handled the representation of the inode cache was kept in the same place. I agree I should have declared the function in ext2fsP.h instead of ext2fs.h; the interface was not great, but that was probably because it was originally intended as an internal interface, and I didn't bother to fix it up before moving it function to another .c file. At this point, we have a couple of choices. (1) define a new interface with a new name (ext2fs_free_inode_cache2) (2) move the function back to freefs.c and make it be static, so that the function signature disappears from the shared library. (3) ignore the problem because it's highly unlikely anyone outside of libext2fs will need to use it, and improving the API/ABI doesn't really matter all that much. Instead, just change the callers to clear fs->icache after calling ext2fs_free_inode_cache(). My preference is (2) or (3). Cheers, - Ted