From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mp1 ([2001:41d0:2:4a6f::]) (using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits)) by ms11 with LMTPS id yBI6E27UzF6WNQAA0tVLHw (envelope-from ) for ; Tue, 26 May 2020 08:33:50 +0000 Received: from aspmx1.migadu.com ([2001:41d0:2:4a6f::]) (using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits)) by mp1 with LMTPS id mO/wDm7UzF7vSQAAbx9fmQ (envelope-from ) for ; Tue, 26 May 2020 08:33:50 +0000 Received: from lists.gnu.org (lists.gnu.org [209.51.188.17]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by aspmx1.migadu.com (Postfix) with ESMTPS id 95E75940607 for ; Tue, 26 May 2020 08:33:49 +0000 (UTC) Received: from localhost ([::1]:35508 helo=lists1p.gnu.org) by lists.gnu.org with esmtp (Exim 4.90_1) (envelope-from ) id 1jdV1v-00030S-Gt for larch@yhetil.org; Tue, 26 May 2020 04:33:47 -0400 Received: from eggs.gnu.org ([2001:470:142:3::10]:49364) by lists.gnu.org with esmtps (TLS1.2:ECDHE_RSA_AES_256_GCM_SHA384:256) (Exim 4.90_1) (envelope-from ) id 1jdV1a-00030F-EG for emacs-orgmode@gnu.org; Tue, 26 May 2020 04:33:26 -0400 Received: from relay6-d.mail.gandi.net ([217.70.183.198]:45437) by eggs.gnu.org with esmtps (TLS1.2:ECDHE_RSA_AES_256_GCM_SHA384:256) (Exim 4.90_1) (envelope-from ) id 1jdV1Y-0001Bv-4Z for emacs-orgmode@gnu.org; Tue, 26 May 2020 04:33:25 -0400 X-Originating-IP: 185.131.40.67 Received: from localhost (40-67.ipv4.commingeshautdebit.fr [185.131.40.67]) (Authenticated sender: admin@nicolasgoaziou.fr) by relay6-d.mail.gandi.net (Postfix) with ESMTPSA id 6D79FC001E; Tue, 26 May 2020 08:33:20 +0000 (UTC) From: Nicolas Goaziou To: Ihor Radchenko Subject: Re: [patch suggestion] Mitigating the poor Emacs performance on huge org files: Do not use overlays for PROPERTY and LOGBOOK drawers References: <87h7x9e5jo.fsf@localhost> <875zdpia5i.fsf@nicolasgoaziou.fr> <87y2qi8c8w.fsf@localhost> <87r1vu5qmc.fsf@nicolasgoaziou.fr> <87imh5w1zt.fsf@localhost> <87blmxjckl.fsf@localhost> <87y2q13tgs.fsf@nicolasgoaziou.fr> <878si1j83x.fsf@localhost> <87d07bzvhd.fsf@nicolasgoaziou.fr> <87imh34usq.fsf@localhost> <87pnbby49m.fsf@nicolasgoaziou.fr> <87tv0efvyd.fsf@localhost> <874kse1seu.fsf@localhost> <87r1vhqpja.fsf@nicolasgoaziou.fr> <87tv0d2nk7.fsf@localhost> <87o8qkhy3g.fsf@nicolasgoaziou.fr> <87sgfqu5av.fsf@localhost> Mail-Followup-To: Ihor Radchenko , emacs-orgmode@gnu.org Date: Tue, 26 May 2020 10:33:19 +0200 In-Reply-To: <87sgfqu5av.fsf@localhost> (Ihor Radchenko's message of "Sat, 23 May 2020 21:52:40 +0800") Message-ID: <87sgfn6qpc.fsf@nicolasgoaziou.fr> User-Agent: Gnus/5.13 (Gnus v5.13) Emacs/26.3 (gnu/linux) MIME-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: quoted-printable Received-SPF: pass client-ip=217.70.183.198; envelope-from=mail@nicolasgoaziou.fr; helo=relay6-d.mail.gandi.net X-detected-operating-system: by eggs.gnu.org: First seen = 2020/05/26 04:33:21 X-ACL-Warn: Detected OS = Linux 3.11 and newer X-Spam_score_int: -25 X-Spam_score: -2.6 X-Spam_bar: -- X-Spam_report: (-2.6 / 5.0 requ) BAYES_00=-1.9, RCVD_IN_DNSWL_LOW=-0.7, RCVD_IN_MSPIKE_H2=-0.001, SPF_PASS=-0.001 autolearn=_AUTOLEARN X-Spam_action: no action X-BeenThere: emacs-orgmode@gnu.org X-Mailman-Version: 2.1.23 Precedence: list List-Id: "General discussions about Org-mode." List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Cc: emacs-orgmode@gnu.org Errors-To: emacs-orgmode-bounces+larch=yhetil.org@gnu.org Sender: "Emacs-orgmode" X-Scanner: scn0 Authentication-Results: aspmx1.migadu.com; dkim=none; dmarc=none; spf=pass (aspmx1.migadu.com: domain of emacs-orgmode-bounces@gnu.org designates 209.51.188.17 as permitted sender) smtp.mailfrom=emacs-orgmode-bounces@gnu.org X-Spam-Score: -1.01 X-TUID: Xp1CA4E4FZP1 Hello, Ihor Radchenko writes: > I have five updates from the previous version of the patch: Thank you. > 1. I implemented a simplified version of element parsing to detect > changes in folded drawers or blocks. No computationally expensive calls > of org-element-at-point or org-element-parse-buffer are needed now. > > 2. The patch is now compatible with master (commit 2e96dc639). I > reverted the earlier change in folding drawers and blocks. Now, they are > back to using 'org-hide-block and 'org-hide-drawer. Using 'outline would > achieve nothing when we use text properties. > > 3. 'invisible text property can now be nested. This is important, for > example, when text inside drawers contains fontified links (which also > use 'invisible text property to hide parts of the link). Now, the old > 'invisible spec is recovered after unfolding. Interesting. I'm running out of time, so I cannot properly inspect the code right now. I'll try to do that before the end of the week. > 4. Some outline-* function calls in org referred to outline-flag-region > implementation, which is not in sync with org-flag-region in this patch. > I have implemented their org-* versions and replaced the calls > throughout .el files. Actually, some org-* versions were already > implemented in org, but not used for some reason (or not mentioned in > the manual). I have updated the relevant sections of manual. These > changes might be relevant to org independently of this feature branch. Yes, we certainly want to move to org-specific versions in all cases. > 5. I have managed to get a working version of outline folding via text > properties. However, that approach has a big downside - folding state > cannot be different in indirect buffer when we use text properties. I > have seen packages relying on this feature of org and I do not see any > obvious way to achieve different folding state in indirect buffer while > using text properties for outline folding. Hmm. Good point. This is a serious issue to consider. Even if we don't use text properties for outline, this also affects drawers and blocks. > For now, I still used before/after-change-functions combination. You shouldn't. > I see the following problems with using only after-change-functions:=20 > > 1. They are not guaranteed to be called after every single change: Of course they are! See below. > From (elisp) Change Hooks: > "... some complex primitives call =E2=80=98before-change-functions=E2=80= =99 once before > making changes, and then call =E2=80=98after-change-functions=E2=80=99 ze= ro or more > times" "zero" means there are no changes at all, so, `after-change-functions' are not called, which is expected. > The consequence of it is a possibility that region passed to the > after-change-functions is quite big (including all the singular changes, > even if they are distant). This region may contain changed drawers as > well and unchanged drawers and needs to be parsed to determine which > drawers need to be re-folded. It seems you're getting it backwards. `before-change-functions' are the functions being called with a possibly wide, imprecise, region to handle: When that happens, the arguments to =E2=80=98before-change-functions=E2= =80=99 will enclose a region in which the individual changes are made, but won=E2= =80=99t necessarily be the minimal such region however, after-change-functions calls are always minimal: and the arguments to each successive call of =E2=80=98after-change-functions=E2=80=99 will then delimit the part of = text being changed exactly. If you stick to `after-change-functions', there will be no such thing as you describe. >> And, more importantly, they are not meant to be used together, i.e., you >> cannot assume that a single call to `before-change-functions' always >> happens before calling `after-change-functions'. This can be tricky if >> you want to use the former to pass information to the latter. > > The fact that before-change-functions can be called multiple times > before after-change-functions, is trivially solved by using buffer-local > changes register (see org--modified-elements). Famous last words. Been there, done that, and it failed. Let me quote the manual: In general, we advise to use either before- or the after-change hooks, but not both. So, let me insist: don't do that. If you don't agree with me, let's at least agree with Emacs developers. > The register is populated by before-change-functions and cleared by > after-change-functions. You cannot expect `after-change-functions' to clear what `before-change-functions' did. This is likely to introduce pernicious bugs. Sorry if it sounds like FUD, but bugs in those areas are just horrible to squash. >> Well, `before-change-fuctions' and `after-change-functions' are not >> clean at all: you modify an unrelated part of the buffer, but still call >> those to check if a drawer needs to be unfolded somewhere. > > 2. As you pointed, instead of global before-change-functions, we can use > modification-hooks text property on sensitive parts of the > drawers/blocks. This would work, but I am concerned about one annoying > special case: > > ------------------------------------------------------------------------- > :BLAH: > > > > :DRAWER: > Donec at pede. > :END: > ------------------------------------------------------------------------- > In this example, the user would not be able to unfold the folder DRAWER > because it will technically become a part of a new giant BLAH drawer. > This may be especially annoying if is more than one screen > long and there is no easy way to identify why unfolding does not work > (with point at :DRAWER:). You shouldn't be bothered by the case you're describing here, for multiple reasons. First, this issue already arises in the current implementation. No one bothered so far: this change is very unlikely to happen. If it becomes an issue, we could make sure that `org-reveal' handles this. But, more importantly, we actually /want it/ as a feature. Indeed, if DRAWER is expanded every time ":BLAH:" is inserted above, then inserting a drawer manually would unfold /all/ drawers in the section. The user is more likely to write first ":BLAH:" (everything is unfolded) then ":END:" than ":END:", then ":BLAH:". > Because of this scenario, limiting before-change-functions to folded > drawers is not sufficient. Any change in text may need to trigger > unfolding. after-change-functions is more appropriate than before-change-functions, and local parsing, as explained in this thread, is more efficient than re-inventing the parser. > In the patch, I always register possible modifications in the > blocks/drawers intersecting with the modified region + a drawer/block > right next to the region. > > ----------------------------------------------------------------------- > ----------------------------------------------------------------------- > > More details on the nested 'invisible text property implementation. > > The idea is to keep 'invisible property stack push and popping from it > as we add/remove 'invisible text property. All the work is done in > org-flag-region. This sounds like a good idea. > This was originally intended for folding outlines via text properties. > Since using text properties for folding outlines is not a good idea, > nested text properties have much less use. AFAIU, they have. You mention link fontification, but there are other pieces that we could switch to text properties instead of overlays, e.g., Babel hashes, narrowed table columns=E2=80=A6 > 3. Multiple calls to before/after-change-functions is still a problem. I > am looking into following ways to reduce this number: > - reduce the number of elements registered as potentially modified > + do not add duplicates to org--modified-elements > + do not add unfolded elements to org--modified-elements > + register after-change-function as post-command hook and remove it > from global after-change-functions. This way, it will be called > twice per command only. > - determine common region containing org--modified-elements. if change > is happening within that region, there is no need to parse > drawers/blocks there again. This is over-engineering. Again, please focus on local changes, as discussed before. > Recipe to have different (org-element-at-point) and > (org-element-parse-buffer 'element) > ------------------------------------------------------------------------- > > :PROPERTIES: > :CREATED: [2020-05-23 Sat 02:32] > :END: > > > > ------------------------------------------------------------------------- I didn't look at this situation in particular, but there are cases where different :post-blank values are inevitable, for example at the end of a section. Regards, --=20 Nicolas Goaziou