I forgot to CC everyone. ---------- Forwarded message --------- From: Will Bush Date: Sat, Apr 25, 2020 at 8:30 AM Subject: Re: bug#40733: 28.0.50; Emacs locks up on paste (yank) of unicode characters To: Robert Pluim Robert> ...we implicitly do '--with-cairo' now. Will> Is that since 27.0.50? Think I just answered my own question the hard way. I was able to narrow the performance issue to starting after rev `6100f9a19e9d8d8e688ad8bbec2233bd6782cbde` and before or at `a75047634697acbc37a9ecd58cc5e7ea9d89d91f` in the master branch in Emacs repository. Which corresponds to these commits: 06caa3b7e5 | * | Refactor Tramp async process code 88efc736f5 | * | Default cairo to enabled 4fc0bc9678 | * | Update from gnulib 0abda558bc | * | Port configure.ac to future Gnulib 6100f9a19e | * | * src/pdumper.c (dump_vectorlike): Unbreak build after 724af7671590c Lol "Default cairo to enabled" really stands out there. It's probably safe to assume that's what it is. Following goes into extraneous detail showing how I verified it's between those two revisions: λ ~/system/nixos/ emacs/revamp* niv update emacs-overlay -a rev=0feda8b31b52f3ea008555dfe79dba3989d3e585 Update emacs-overlay Reading sources file Done: Update emacs-overlay λ ~/system/nixos/ emacs/revamp* home-manager switch ... home manager spam goes here λ ~/system/nixos/ emacs/revamp* time emacs -Q --eval '(message "︵")' -kill emacs -Q --eval '(message "︵")' -kill 0.42s user 0.02s system 64% cpu 0.690 total λ ~/system/nixos/ emacs/revamp* time emacs -Q --eval '(message "︵")' -kill emacs -Q --eval '(message "︵")' -kill 0.43s user 0.02s system 94% cpu 0.472 total λ ~/system/nixos/ emacs/revamp* niv update emacs-overlay -a rev=a75047634697acbc37a9ecd58cc5e7ea9d89d91f Update emacs-overlay Reading sources file Done: Update emacs-overlay λ ~/system/nixos/ emacs/revamp* home-manager switch ... more home manager spam λ ~/system/nixos/ emacs/revamp* time emacs -Q --eval '(message "︵")' -kill emacs -Q --eval '(message "︵")' -kill 80.73s user 0.02s system 99% cpu 1:20.97 total λ ~/system/nixos/ emacs/revamp* The `niv update emacs-overlay -a rev=a75047634697acbc37a9ecd58cc5e7ea9d89d91f` command is using a tool called niv to update emacs-overlay to a pinned version. The rev in the diff below is for the Emacs repository. *# perf issue:* a75047634697acbc37a9ecd58cc5e7ea9d89d91f Author: emacs-overlay AuthorDate: Tue Jan 14 12:10:25 2020 +0000 Commit: emacs-overlay CommitDate: Tue Jan 14 12:10:25 2020 +0000 Parent: 9351772 Updated repos/melpa Contained: master Updated repos/emacs modified repos/emacs/emacs.json @@ -1 +1 @@ -{"rev": "4fc0bc96787252b1b3e14a7f747ef556273b5979", "sha256": "0syw4xcbps6i62fa7l88zyvyc3kiggx2kpa2n41p8y2pl01vdqqs", "version": "20200114.0"} +{"rev": "06caa3b7e5e9fe91b6918f8567adbd5501d6dbdd", "sha256": "0kzk30660xky0zj7v5sr9a49pnnz609jda4s8x86pjk91x1wrv2i", "version": "20200114.0"} *# no perf issue:* 0feda8b31b52f3ea008555dfe79dba3989d3e585 Author: emacs-overlay AuthorDate: Mon Jan 13 00:10:30 2020 +0000 Commit: emacs-overlay CommitDate: Mon Jan 13 00:10:30 2020 +0000 Parent: e38dc3b Updated repos/melpa Contained: master Updated repos/emacs modified repos/emacs/emacs.json @@ -1 +1 @@ -{"rev": "41d9d51cf5ac5b76c09802388e1691cf489d9d9d", "sha256": "1vy1fcw2m7lbw8wcwmp04zwkqra835vdbxbgnms3wgrwviqm14zd", "version": "20200111.0"} +{"rev": "6100f9a19e9d8d8e688ad8bbec2233bd6782cbde", "sha256": "01fvxplljwnz11sizlfpl219dvrg7yf790zmr69wvkn6wlgxif76", "version": "20200112.0"} On Sat, Apr 25, 2020 at 5:34 AM Will Bush wrote: > Robert> Which font specifically does emacs end up using for that character? > Robert> Emacs ends up using 'Noto Sans CJK KR' for me here. > > When google fonts is removed? > > This is what `C-u C-x =` says: > > ftcrhb:-PfEd-Unifont-normal-normal-normal-*-15-*-*-*-d-0-iso10646-1 > (#xDD38ftcrhb:-PfEd-Unifont-normal-normal-normal-*-15-*-*-*-d-0-iso10646-1 > (#xDD38) > > Note on the above: For the hell of it, I tried installing `noto-fonts` > font pack > from nixpkgs and it didn't make a difference. Then again, `fc-list > --verbose | > rg "Noto Sans CJK" -i` produced no results so that specific font probably > isn't > in that font pack. > > When google fonts are installed: > > ftcrhb:-GNU-Unifont-normal-normal-normal-Sans-Serif-16-*-*-*-c-80-iso10646-1 > (#xDD36) > > Robert> BTW, if you want to ignore that font, you can set > Robert> 'face-ignored-fonts' to match it, and you won't have to uninstall > it. > > Thanks, I didn't know that! Maybe I can use that to narrow down to the > specific > font that's causing problems because adding `google-fonts` adds 2905 fonts > for > me, and many I would like to have. > > Robert> I donʼt think thereʼs much point in that: emacs-26 uses Xft for > font > Robert> handling, emacs-27 uses Cairo+Harfbuzz[1]; theyʼre fundamentally > doing > Robert> very different things, so I donʼt think this is caused by a single > Robert> identifiable change. > > I'm not trying to prove you wrong or anything. It's just easy for me to try > different versions because I'm using > (https://github.com/nix-community/emacs-overlay). However, I tried Emacs > 27.0.50 > and it's behaving exactly the same as Emacs 26. I glanced at the > `report-emacs-bug` output and the build inputs look the same. I can > include it > if desired. > > λ ~/ time emacs -Q --eval '(message "hi")' -kill > emacs -Q --eval '(message "hi")' -kill 0.18s user 0.02s system 67% cpu > 0.303 total > λ ~/ time emacs -Q --eval '(message "︵")' -kill > emacs -Q --eval '(message "︵")' -kill 0.44s user 0.03s system 95% cpu > 0.494 total > λ ~/ emacs --version > GNU Emacs 27.0.50 > > Robert> ...we implicitly do '--with-cairo' now. > > Is that since 27.0.50? > > Were either Cairo+Harfbuzz libraries updated since 27.0.50 (perhaps a > regression > in those libraries)? I'll follow up with an update later after testing > more versions. > > Robert> Although you can still build it with Xft if you want, but I > Robert> wouldnʼt recommend that, since it will crash once you start > Robert> processing Emojis and other 'interesting' Unicode characters. > > Just to verity I understand. Building with Xft is what `--with-xft` is > doing in > the following from my initial email? > > Configured using: > 'configure > --prefix=/nix/store/5v0fp6vikajaqc2v0ppkm51hfc054mnm-emacs-git-20190910.0 > --disable-build-details --with-modules --with-x-toolkit=gtk3 --with-xft > CFLAGS=-DMAC_OS_X_VERSION_MAX_ALLOWED=101200' > > Eli> I'm not sure I understand: you are saying that slow, but correct > Eli> display is _worse_ than displaying a white space instead of the > Eli> correct glyph, i.e. producing incorrect display? To me, it sounds > Eli> like Emacs 27+ actually _improves_ things in this case. > > Let me quantify the performance because I've been ambiguous about it so > far: > > λ ~/ time emacs -Q --eval '(message "hi")' -kill > emacs -Q --eval '(message "hi")' -kill 0.19s user 0.02s system 55% cpu > 0.371 total > λ ~/ time emacs -Q --eval '(message "︵")' -kill > emacs -Q --eval '(message "︵")' -kill 81.64s user 0.03s system 99% cpu > 1:21.91 total > > It takes ~81 seconds to do something while locking up the UI. That's > personally > beyond my threshold for killing the process. > > > On Wed, Apr 22, 2020 at 2:35 AM Robert Pluim wrote: > >> >>>>> On Tue, 21 Apr 2020 15:35:23 -0400, James Cloos >> said: >> >> >>>>> "RP" == Robert Pluim writes: >> RP> Footnotes: >> RP> [1] Although you can still build it with Xft if you want, but I >> RP> wouldnʼt recommend that, since it will crash once you start >> RP> processing Emojis and other 'interesting' Unicode characters. >> >> James> note that master will also crash when using cr+hb on some code >> points. >> >> James> such as some private use characters. >> >> Examples? Eli fixed one such case with Bug#39892, but if there are >> more we should fix them (please open a separate bug report for that). >> >> Robert >> >