Re: Performance regression in export to LaTeX?
Am Sonntag, dem 19.12.2021 um 19:21 + schrieb José Abílio Matos: > If this proves to be a problem we can speed lyx2lyx code (using > profiling to identify the bottlenecks). No, this doesn't seem to be involved. Jürgen signature.asc Description: This is a digitally signed message part -- lyx-devel mailing list lyx-devel@lists.lyx.org http://lists.lyx.org/mailman/listinfo/lyx-devel
Re: Performance regression in export to LaTeX?
On Sunday, 19 December 2021 17.15.19 WET Jürgen Spitzmüller wrote: > The first thing to check, since this is a 2.3.x file, is how much time > is taken by lyx2lyx. If this proves to be a problem we can speed lyx2lyx code (using profiling to identify the bottlenecks). -- José Abílio-- lyx-devel mailing list lyx-devel@lists.lyx.org http://lists.lyx.org/mailman/listinfo/lyx-devel
Re: Performance regression in export to LaTeX?
Le 19/12/2021 à 19:22, Scott Kostyshak a écrit : On Sun, Dec 19, 2021 at 07:08:56PM +0100, Jürgen Spitzmüller wrote: Am Sonntag, dem 19.12.2021 um 18:15 +0100 schrieb Jürgen Spitzmüller: I can bisect as well. I gave up. I'll give it a try. Might not be able to do it until tomorrow though. If we find a really big document, we can also try to run it under a profiler. I am not sure whether 5 seconds of UserGuide.lyx will be enough. Is this with runtime debugging disables? JMarc -- lyx-devel mailing list lyx-devel@lists.lyx.org http://lists.lyx.org/mailman/listinfo/lyx-devel
Re: Performance regression in export to LaTeX?
On Sun, Dec 19, 2021 at 07:08:56PM +0100, Jürgen Spitzmüller wrote: > Am Sonntag, dem 19.12.2021 um 18:15 +0100 schrieb Jürgen Spitzmüller: > > I can bisect as well. > > I gave up. I'll give it a try. Might not be able to do it until tomorrow though. Scott signature.asc Description: PGP signature -- lyx-devel mailing list lyx-devel@lists.lyx.org http://lists.lyx.org/mailman/listinfo/lyx-devel
Re: Performance regression in export to LaTeX?
Am Sonntag, dem 19.12.2021 um 18:15 +0100 schrieb Jürgen Spitzmüller: > I can bisect as well. I gave up. Jürgen signature.asc Description: This is a digitally signed message part -- lyx-devel mailing list lyx-devel@lists.lyx.org http://lists.lyx.org/mailman/listinfo/lyx-devel
Re: Performance regression in export to LaTeX?
Am Sonntag, dem 19.12.2021 um 11:55 -0500 schrieb Scott Kostyshak: > I can try to do another bisect if there's interest. I'm just not sure > this is something even related to a use case. I guess the first thing > to check would be if indeed the issue only shows up with a very long > line. The first thing to check, since this is a 2.3.x file, is how much time is taken by lyx2lyx. If I convert your example file to master file format, I get real0m6,868s user0m6,828s sys 0m0,020s So, yes, it seems worthwile to investigate. I can bisect as well. Jürgen signature.asc Description: This is a digitally signed message part -- lyx-devel mailing list lyx-devel@lists.lyx.org http://lists.lyx.org/mailman/listinfo/lyx-devel
Re: Performance regression in export to LaTeX?
On Sun, Dec 19, 2021 at 05:39:16PM +0100, Jürgen Spitzmüller wrote: > Am Sonntag, dem 19.12.2021 um 11:37 -0500 schrieb Scott Kostyshak: > > Thanks for checking. Indeed I think my computer has some performance > > issues (especially with I/O, which I think is relevant in this case). > > > > And what do you get with 2.3.0? > > real 0m1,108s > user 0m1,077s > sys 0m0,031s > > So there is a significant difference here, as well. I can try to do another bisect if there's interest. I'm just not sure this is something even related to a use case. I guess the first thing to check would be if indeed the issue only shows up with a very long line. Scott signature.asc Description: PGP signature -- lyx-devel mailing list lyx-devel@lists.lyx.org http://lists.lyx.org/mailman/listinfo/lyx-devel
Re: Performance regression in export to LaTeX?
Am Sonntag, dem 19.12.2021 um 11:37 -0500 schrieb Scott Kostyshak: > Thanks for checking. Indeed I think my computer has some performance > issues (especially with I/O, which I think is relevant in this case). > > And what do you get with 2.3.0? real0m1,108s user0m1,077s sys 0m0,031s So there is a significant difference here, as well. Jürgen signature.asc Description: This is a digitally signed message part -- lyx-devel mailing list lyx-devel@lists.lyx.org http://lists.lyx.org/mailman/listinfo/lyx-devel
Re: Performance regression in export to LaTeX?
On Sun, Dec 19, 2021 at 05:29:07PM +0100, Jürgen Spitzmüller wrote: > Am Sonntag, dem 19.12.2021 um 17:24 +0100 schrieb Jürgen Spitzmüller: > > Though I am getting > > > > real0m5,935s > > user0m5,739s > > sys 0m0,066s > > > > With this on master (with -e pdflatex). > > (i.e., with your testfile) Thanks for checking. Indeed I think my computer has some performance issues (especially with I/O, which I think is relevant in this case). And what do you get with 2.3.0? Scott signature.asc Description: PGP signature -- lyx-devel mailing list lyx-devel@lists.lyx.org http://lists.lyx.org/mailman/listinfo/lyx-devel
Re: Performance regression in export to LaTeX?
Am Sonntag, dem 19.12.2021 um 17:24 +0100 schrieb Jürgen Spitzmüller: > Though I am getting > > real 0m5,935s > user 0m5,739s > sys 0m0,066s > > With this on master (with -e pdflatex). (i.e., with your testfile) Jürgen signature.asc Description: This is a digitally signed message part -- lyx-devel mailing list lyx-devel@lists.lyx.org http://lists.lyx.org/mailman/listinfo/lyx-devel
Re: Performance regression in export to LaTeX?
Am Sonntag, dem 19.12.2021 um 11:10 -0500 schrieb Scott Kostyshak: > On master (with your recent fix), I get times like the following: > > real0m44.676s > user0m44.575s > sys 0m0.092s > > I'm still using the following command when testing: > > time mylyx 2.3.x -e pdflatex UserGuide.lyx > > The format "pdflatex" is the shortname for "LaTeX (pdflatex)", where > pdf2 is the shortname for "PDF (pdflatex)". i.e., I'm checking just > LyX's LaTeX export without compiling. You're right. Though I am getting real0m5,935s user0m5,739s sys 0m0,066s With this on master (with -e pdflatex). Jürgen signature.asc Description: This is a digitally signed message part -- lyx-devel mailing list lyx-devel@lists.lyx.org http://lists.lyx.org/mailman/listinfo/lyx-devel
Re: LyX 2.4.0
On 12/18/21 08:21, Daniel wrote: On 2021-12-14 10:45, Daniel wrote: On 14/12/2021 10:34, Richard Kimberly Heck wrote: On 12/12/21 08:55, Daniel wrote: On 2021-12-10 06:52, Richard Kimberly Heck wrote: On 12/9/21 03:00, Daniel wrote: On 2021-12-07 23:04, Richard Kimberly Heck wrote: On 12/6/21 22:00, Daniel wrote: On 2021-12-06 22:58, Richard Kimberly Heck wrote: Do you mean by "safe enough to include" that they should be more or less done? For example, there were a couple of tickets where you said that you take a look at while there were things that needed to be done still. I take it that this is not going to happen, right? Safe as in: Not likely to cause bugs we need to solve before the release. Once we hit beta, we are in bug-fixing mode, and the fewer the better. Riki Okay. (Though it is not what I have expected beta to be. I would have thought that software in beta is actually "likely" to have bug rather than unlikely. I would have thought the latter is release candidate.) Well, we know beta will have bugs---indeed, we know of some already---but the idea is that we do our best not to create new bugs at that point. New features bring new bugs. That's why beta release goes with feature freeze. That said, at the moment we're talking about what will make it into beta, so some risk is all right. Riki I have to do the tickets one by one. Should I post them here (with reference) on the list, create a (temporary) meta bug with the list of tickets, or just ping you on the tickets themselves? Maybe assemble a list of them, and then post it here. I'll see it, I hope! You could also add a 'triage' keyword to the bugs themselves. I've used that for this kind of purpose. Riki Here is a list of enhancements with milestone 2.4.0 and patches. I guess it is worthwhile to go though them anyway and either apply or re-target them: https://www.lyx.org/trac/query?status=accepted&status=assigned&status=new&status=reopened&description=~&reporter=~&summary=~&milestone=2.4.0&keywords=~patch&type=enhancement&col=id&col=summary&col=keywords&col=reporter&col=status&col=type&col=severity&desc=1&order=id I will not manage to do more than give you this list. Maybe you just ask me if something is unclear that seems more efficient since you will look at them carefully anyway? I could also give more info on them but that is as much as I can do for this weekend. (Maybe I missed the road map but some warning would have been helpful a bit in advance. I guess it was communicated on some internal list.) It appears that JMarc's recent changes have introduced some serious bugs. Unsurprisingly. So there is probably some leeway for these patches. I'll also try to have a look. If you have the time, it would be really helpful to re-target bugs that clearly are NOT format changes. Those can definitely wait until 2.4.1. Riki It is a bit more complicated, I think. There are non-format changes that are (arguably) improvements on new features introduced in 2.4.0dev. Wouldn't it be odd to introduce such new features without the improvements in the first place? One concrete example are the new drop-down buttons called "DynamicMenu" [sic]. Currently, they aren't customizable but one of my patches makes them customizable. Daniel Down to 16 bugs. Does that seem manageable for you? https://www.lyx.org/trac/query?status=accepted&status=assigned&status=new&status=reopened&description=~&reporter=~&summary=~&milestone=2.4.0&keywords=~patch&type=enhancement&col=id&col=summary&col=keywords&col=reporter&col=status&col=type&col=severity&desc=1&order=id Otherwise, I can try to take a look again. I'll have a look this week. We still have a bit of time, as JMarc's big patch gets tested. Riki -- lyx-devel mailing list lyx-devel@lists.lyx.org http://lists.lyx.org/mailman/listinfo/lyx-devel
Re: Looking for KDE testers (Re: [PATCH] allow to force backing store for display)
On 12/17/21 09:25, Jean-Marc Lasgouttes wrote: Le 16/12/2021 à 16:39, Jean-Marc Lasgouttes a écrit : This patch allows to enforce the use of of a backing store for display. This has no effect for Wayland and macOS, when backing store is always on. The main use for it is x11 with a window manager theme with translucent windows (KDE) https://www.lyx.org/trac/ticket/12119 At this point, there is no UI for this, because I think this is a niche case. I would be open to the idea of adding a checkbox somewhere. The reworked patch is now in master, along with a new checkbox for X11 and Windows users. I would be grateful of a KDE/X11 user could try to install a translucid windows theme like Layan and check 1/ that it creates display issues 2/ that the new checkbox in Preferences|Display fixes it (at the cost of losing subpixel aliasing). This is candidate for 2.3.7 (by default it does nothing). I can't test, unfortunately, but it's fine for 2.3.x if someone else can. Riki -- lyx-devel mailing list lyx-devel@lists.lyx.org http://lists.lyx.org/mailman/listinfo/lyx-devel
Re: Performance regression in export to LaTeX?
On Sun, Dec 19, 2021 at 03:45:40PM +0100, Jürgen Spitzmüller wrote: > Am Sonntag, dem 19.12.2021 um 15:03 +0100 schrieb Jürgen Spitzmüller: > > So it's the cprotection checks that seem to cost time. Need to check > > whether this can be optimized. > > After 61b8afd893ec we're back at > > real 0m19,685s > user 0m15,633s > sys 0m3,025s > > Which is still a bit slower than stable, but a lot faster that where we > have been. Thank you, Jürgen! That is indeed much better. Here is an example file that might show a different (but related?) issue: https://www.dropbox.com/s/cwaj2ms14fln8oi/example.23.lyx?dl=0 It's a contrived example file. It just pastes a string like "h" over and over. I think it is all on the same line. And pdflatex refuses to compile it because of the extremely long line. I get: "! Unable to read an entire line---bufsize=20." Thus, a fair response would be "find me a more reasonable use case before making noise" :) Nonetheless, the difference in timings are quite large: On 2.3.0 I'm getting times like the following: real 0m6.476s user 0m6.396s sys 0m0.081s On master (with your recent fix), I get times like the following: real 0m44.676s user 0m44.575s sys 0m0.092s I'm still using the following command when testing: time mylyx 2.3.x -e pdflatex UserGuide.lyx The format "pdflatex" is the shortname for "LaTeX (pdflatex)", where pdf2 is the shortname for "PDF (pdflatex)". i.e., I'm checking just LyX's LaTeX export without compiling. Scott signature.asc Description: PGP signature -- lyx-devel mailing list lyx-devel@lists.lyx.org http://lists.lyx.org/mailman/listinfo/lyx-devel
Re: Performance regression in export to LaTeX?
Am Sonntag, dem 19.12.2021 um 15:03 +0100 schrieb Jürgen Spitzmüller: > So it's the cprotection checks that seem to cost time. Need to check > whether this can be optimized. After 61b8afd893ec we're back at real0m19,685s user0m15,633s sys 0m3,025s Which is still a bit slower than stable, but a lot faster that where we have been. Jürgen signature.asc Description: This is a digitally signed message part -- lyx-devel mailing list lyx-devel@lists.lyx.org http://lists.lyx.org/mailman/listinfo/lyx-devel
Re: Performance regression in export to LaTeX?
Am Sonntag, dem 19.12.2021 um 11:45 +0100 schrieb Jürgen Spitzmüller: > I try to bisect myself. This leads to: commit b814c4fda732c8b5ee019692eb881c35b9335da6 Author: Juergen Spitzmueller Date: Sun Sep 20 08:45:42 2020 +0200 Fix unnecessary cprotect Note, however, that this is not accountable for all the increment. There seem to be several minor leaps involved additionally. But the one above is the most significant (by far), ca. 10 secs. So it's the cprotection checks that seem to cost time. Need to check whether this can be optimized. Jürgen signature.asc Description: This is a digitally signed message part -- lyx-devel mailing list lyx-devel@lists.lyx.org http://lists.lyx.org/mailman/listinfo/lyx-devel
Re: Performance regression in export to LaTeX?
Am Samstag, dem 18.12.2021 um 13:26 -0500 schrieb Scott Kostyshak: > Thanks, Jean-Pierre. I wonder why you must export to pdf2. Using > "pdflatex" instead gives an error? Because -e takes a file format as argument, and the file format for pdflatex output is pdf2, not pdflatex. > > In any case, I see roughly the same ratio. I tried to do a bisect and > got to the following: > > commit 26ea1e14966c092a7b87c75c19b83a369e79aeb8 > Author: Juergen Spitzmueller > Date: Sun Apr 22 19:06:46 2018 +0200 > > Align fontenc with document fonts > > But I'm not so confident in my bisect since performance issues are > hard to bisect. This commit also changes file format and prefs > change, so that might play a role. I do not see such a huge difference. At 26ea1e14966c real0m18,531s user0m13,952s sys 0m3,168s At 337ec5830a041 (the commit before): real0m16,335s user0m11,609s sys 0m3,259s So a bit more, but not that much. Stable gives: real0m16,731s user0m11,244s sys 0m4,251s And current master: real0m33,165s user0m28,740s sys 0m3,420s So the cuplrit must be later (or it's a cumulative increasement). I try to bisect myself. Jürgen signature.asc Description: This is a digitally signed message part -- lyx-devel mailing list lyx-devel@lists.lyx.org http://lists.lyx.org/mailman/listinfo/lyx-devel
