this-month-in-org/content/2022-05-31-folding.org

225 lines
11 KiB
Org Mode

# -*- org-plot/gnuplot-term-extra: "background rgb '#fafafa' size 800,500 font 'Alegreya Sans, 16'"; -*-
#+title: May 2022
#+subtitle: Folding more improvement into Org
#+author: TEC
#+date: 2022-05-31
Finding time as of late has been more difficult than I anticipated, and on top
of that, just as I was thinking of writing last month's post, I got distracted
by an exciting patchset that has been in the works for over a year finally
getting sorted out and landing. So, I hope that some of the fun developments in
this post will make up the absense of the last one 🙂.
Since it's been longer than I thought since the last standard post, we've got a
fair few commits to catch up on --- about 200. Most of these are miscellaneous
minor improvements and bugfixes, but a few notable changes have arrived too.
* Folding
The fabulous new folding engine (=org-fold-core=) should noticeably improve Org's
performance with large files. It contains a number of key optimisations to
improve speed, namely:
+ Deferring fontification of folded regions
+ Using text properties (\(\mathcal{O}(n \log n)\)) instead of overlays (\(\mathcal{O}(n^2)\)) for folded regions
+ A collection of aggressive optimisations available under ~org-fold-core--optimise-for-huge-buffers~
+ Convert text properties to overlays for =isearch= (which currently only supports overlays)
How noticeable is the overall performance impact? Well, I poked Ihor and he was
kind enough to whip up some benchmarks.
#+caption: The scaling of ~org-shifttab~ showing file contents, as file size increases, with and without org-fold.
#+attr_html: :class invertible
[[file:figures/org-fold-perf-shifttab-contents.svg]]
#+caption: The scaling of ~org-shifttab~ showing the entire file, as file size increases, with and without org-fold.
#+attr_html: :class invertible
[[file:figures/org-fold-perf-shifttab-showall.svg]]
Well this looks very promising[fn::Note the difference in scale, org-fold makes
the most difference in the graph where the times are an order of magnitude
more.]! Let's see how much of an improvement this is overall.
#+caption: Time to run =org-shifttab= twice, cycling through all three display modes (in seconds).
| File size (Mb) | Headings (thousands) | Bugfix (no org-fold) | Main (with org-fold) | Improvement |
|----------------+----------------------+----------------------+----------------------+-------------|
| 18 | 36 | 115.31 | 0.89 | 99% |
| 8.8 | 24 | 19.03 | 0.48 | 97% |
| 4.4 | 5 | 3.79 | 0.13 | 97% |
| 2.2 | 2 | 1.29 | 0.08 | 94% |
| 1.1 | 1 | 0.50 | 0.045 | 91% |
#+TBLFM: $5=100*(1 - $4/$3) ; %.0f%%
To be clear, even the smallest file in this data --- a 1.1 Mb Org file with around
a thousand headings, is fairly large. So, it's unlikely you'll notice much of a
difference with small--medium files, but if you a few large+ files this should
be a /fantastic/ improvement. Once again, thanks Ihor!
#+begin_warning
The change to text properties instead of overlays breaks a number of third party
packages like =evil-search= and =consult='s ~consult-line~.
If you are involved in any packages affected by this, you'll either want to
consider supporting invisible text, or look at ~isearch-filter-predicate~ and
~isearch-mode-end-hook~, which =org-fold= now uses.
If you're an end-user, perhaps politely make an issue on the repo for a project
/if no issue currently exists/, and either:
+ Stay off Org's bleeding edge till the package ecosystem has adapted to this change
+ Help the packages you use adapt to this change
+ Set ~org-fold-core-style~ to ~overlays~ to restore the old behaviour
#+end_warning
** Benchmark data :noexport:
#+plot: ind:2 deps:(3 4) with:linespoints file:"figures/org-fold-perf-shifttab-contents.svg"
#+plot: set:"title '{/*1.6 Running org-shifttab to CONTENTS}'" set:"xlabel '{/*1.2 File size (Mb)}'" set:"ylabel '{/*1.2 Load time (seconds)}'"
#+plot: set:"logscale x" set:"logscale y" set:"linetype 1 pt 5" set:"linetype 2 pt 7"
| Headings (k) | File Size (Mb) | Bugfix (no org-fold) | Main (org-fold) |
|--------------+----------------+----------------------+-----------------|
| 217 | 150 | 173.28 | 10.73 |
| 36 | 18 | 1.97 | 0.92 |
| 24 | 8.8 | 0.81 | 0.46 |
| 5 | 4.4 | 0.09 | 0.21 |
| 2 | 2.2 | 0.03 | 0.07 |
| 1 | 1.1 | 0.02 | 0.04 |
#+plot: ind:2 deps:(3 4) with:linespoints file:"figures/org-fold-perf-shifttab-showall.svg"
#+plot: set:"title '{/*1.6 Running org-shifttab to SHOW-ALL}'" set:"xlabel '{/*1.2 File size (Mb)}'" set:"ylabel '{/*1.2 Load time (seconds)}'"
#+plot: set:"logscale x" set:"logscale y" set:"linetype 1 pt 5" set:"linetype 2 pt 7"
| Headings (k) | File Size (Mb) | Bugfix (no org-fold) | Main (org-fold) |
|--------------+----------------+----------------------+-----------------|
| 217 | 150 | 8921.02 | 0.09 |
| 36 | 18 | 100.96 | 0.02 |
| 24 | 8.8 | 17.35 | 0.01 |
| 5 | 4.4 | 3.79 | 0.005 |
| 2 | 2.2 | 1.20 | 0.003 |
| 1 | 1.1 | 0.49 | 0.003 |
* /Engraved/ source code blocks in LaTeX
All too often exporting code to LaTeX has been a disappointment, with lovely
syntax highlighting from Emacs major modes replaced with a markedly inferior
attempt by pygments (setting ~org-latex-listings~ to ~minted~) in a colour scheme I
don't really like.
A bit over a year ago, a project called [[https://github.com/tecosaur/engrave-faces][engrave-faces]] started with the aim of
making Emacs' font-lock more exportable, like a generalised =htmlize.el=. This has
recently been used to provide a new option for inline and block source code
exports in LaTeX.
#+caption: A screenshot of an Org code block, exported to a PDF,
#+caption: using =engrave-faces= and the =doom-one-light= theme.
#+attr_html: :class invertible
[[file:figures/engraved-faces-sample.png]]
To use this, simply install the package and set ~org-latex-src-block-backend~ (a
rename of ~org-latex-listings~ to better reflect its usage) to ~engraved~.
While this is sufficient to get started, this new backend also allows for some
new options. The theme used for /engraving/ a source block can be set globally
with the new variable ~org-latex-engraved-theme~, or per-file with the
=#+latex_engraved_theme= keyword. It takes either the name of a theme, or the
symbol =t= as a stand-in for the current theme.
The theme can also be set on a per-block level using the LaTeX attribute
=:engraved-theme=.
#+caption: Seven code blocks exported to LaTeX, each with a different engrave-faces theme.
[[file:figures/engraved-faces-multitheme.png]]
Here's what using these new capabilities looks like in practice.
#+begin_src org
,#+title: Engraving source blocks
,#+latex_engraved_theme: modus-operandi
,#+begin_src emacs-lisp
(message "look ma, some %s" 'code)
,#+end_src
,#+attr_latex: :engraved-theme modus-viviandi
,#+begin_src shell
echo "This is shell code"
,#+end_src
#+end_src
This may well be the best syntax-highlighting solution available for PDFs/LaTeX
currently available, but I am a tad biased 😛.
* TexInfo export improvements
Jonas Bernoulli has been using a custom TexInfo backend for Magit's
documentation for a while now, and over the past few months he's worked the
features he was missing into Org's built-in TexInfo exporter.
Upstreaming like this always takes a fair bit of effort, so thank you Jonas for
going through with this!
* Toggle noweb prefix handling
Previously, whenever a noweb reference appeared on a non-empty line, a
multi-line replacement would duplicate the content before the noweb reference.
Clearly, this is not always desirable, and this behaviour can now be turned of
by setting the new header argument =:noweb-prefix no=.
#+begin_src org
,#+begin_src emacs-lisp :noweb yes :noweb-prefix no
(setq example-data "<<example>>")
,#+end_src
Will now expand to
,#+begin_src emacs-lisp
(setq example-data "some
multi-line
content")
,#+end_src
Instead of
,#+begin_src emacs-lisp
(setq example-data "some
(setq example-data "multiline
(setq example-data "content")
,#+end_src
#+end_src
* Package highlight: org-modern
I think we've all [[https://github.com/integral-dw/org-superstar-mode][seen]] [[https://github.com/sabof/org-bullets][plenty]] of =org-mode= [[https://github.com/Fuco1/org-pretty-table][prettification]] [[https://github.com/harrybournis/org-fancy-priorities][packages]] [[https://gitlab.com/marcowahl/org-pretty-tags][before]], so
what makes Minad's [[https://github.com/minad/org-modern][org-modern]] special? It's actually doing something similar to
Ihor's org-fold improvements, switching out slower overlay-based approaches for
text properties. I can confirm that switching out =org-superstar-mode= for
=org-modern= has made a substantial improvement in my experience, halving the
first-load time of my =config.org= to around 20 seconds. If you're a fan of Org
prettification and haven't taken a look at this package, I highly recommend
giving it a shot.
#+caption: A demonstration of org-modern taken from the project README.
#+attr_html: :class invertible
[[file:figures/org-modern-readme-demo.gif]]
* Other improvements
+ Clean up some magic numbers in =org-attach= _Marco Wahl_
+ Allow /any/ command form in ~org-attach-commands~ (including keyboard macros) _Marco Wahl_
+ Allow =dest= in ~org-list-send-item~ to be a buffer position _Sacha Chua_
+ Improve CSL-JSON date handling in =oc-basic= _David Lukes_
+ Add =TOML= and =desktop= language aliases _TEC_
+ Speed up cached bibliography retrieval in =oc-basic= _Ihor Radchenko_
+ Allow setting PlantUML jar arguments _Ihor Radchenko_
+ Allow for customisation of property separators with ~org-property-separators~ _Tyler Grinn_
+ New =ox-latex= maintainer, Daniel Fleischer
+ More unit tests _Kyle Keyer, Nick Dokos_
+ Documentation improvements _Kyle Meyer, Juan Manuel Macias, Bastien, Karl
Fogel, Cody Harris_
* Bugfixes
+ An Emacs <28 bug in =org-persist= _Ihor Radchenko_
+ Author extraction in =oc-basic= _Nicolas Goaziou_
+ Fix behaviour of ~org-copy-visible~ with adjacent tex and ~buffer-invisibility-spec~ _Kyle Meyer_
+ Parsing of inline footnotes with parentheses _Nicolas Goaziou_
+ Honor ~default-directory~ in =ob-gnuplot= _Ihor Radchenko_
+ Heading fontification bug _Anders Johansson_
+ Template expansion where one key is a substring of another _Andrew Arensburger_